People Analytics in the Era of Big Data

1. 1Start from a business question, not a survey. The source frames engagement as a talent question tied to strategy: 'How engaged are our top performers in our key business strategies?' and 'Do we know which talents are at risk of leaving? What are we doing to keep them?' Write one question and one decision it will change before touching data.
2. 2Inventory every engagement signal you hold. The source lists 'real-time information on employee engagement, actions, sentiment, productivity, location, quality, and aspirations.' Include unstructured survey comments, which one company in the source called 'an untapped resource.' Record owner, refresh rate and employee-level identifier for each.
3. 3Treat the survey as one input, not the measure. The book's own sequence is surveys, then making them predictive, then 'moving beyond the survey' to other engagement measures. Add at least one behavioral or outcome signal (productivity, quality, actions) next to the attitude score.
4. 4Move from isolated metrics to prediction. The source says to go 'beyond simple dashboards, scorecards, and isolated talent metrics' and to use metrics that are 'inherently predictive' with 'high business impact.' Link engagement data to a later outcome you can observe, such as leaving or output, so that a score becomes a forecast.
5. 5Segment before you average. Compare groups that differ in meaning, such as long-tenured versus new employees, and top performers versus the rest. The source's example targeted talent by tenure group to yield 'long-term employee engagement.'
6. 6Reconcile your numbers with other teams before presenting. In the source, the analytics team produced 'statistics that were in conflict with those from other teams in the company,' which eroded trust. Agree on definitions (who counts as engaged, which population, which period) and document them.
7. 7Package output for business readers. Deliver a finding, a recommended action and one clear chart. Do not send 'lots of raw data and a graph or two that was difficult for the business's people to understand.'
8. 8Hand the action to managers and keep the analytics team as partner. The source's emerging paradigm says 'managers and TM jointly apply their experience' and employee development is shared among managers, employees and the talent management organization. Name the manager who acts on each finding.

The source describes a company that believed its unstructured engagement-survey data was an untapped resource for people strategy and business strategy. It searched seven months for an analyst and hired one with a statistics degree, computer science experience and a track record with well-known brands. Four existing HR analysts were placed under him. His first assignment compared long-tenured employees with new ones, to target prospective talent for long-term engagement and a solid return on recruiting dollars. Things started well, then three problems appeared. The team disappeared for weeks with little analysis completed. When it delivered, the output was raw data and a graph or two the business could not follow. Its statistics also conflicted with figures from other teams. Each problem maps to a step above: define the question and a delivery cadence, package for business readers, and reconcile definitions before release. The passage excerpt does not give the outcome, so treat it as a cautionary pattern, not a success case.

• If your only engagement data is an annual survey score, then add a behavioral or outcome signal and a prediction target before you report a trend.
• If your figure disagrees with another team's, then stop and reconcile definitions and populations before anyone sees it.
• If the audience is business leaders, then deliver one finding, one action and one owner, not raw tables.
• If the question concerns top performers or at-risk talent, then segment to that group first, because company-wide averages hide it.
• If an analytics hire is brought in to unlock engagement data, then set interim deliverables and a business partner from week one, so work does not disappear for weeks.

• Dashboard-only engagement: a score and a bar chart with no link to outcomes. The tell: the report cannot say what will happen to anyone's retention or output.
• Ignoring unstructured data: comments are collected but never analyzed. The tell: free-text fields exist and nobody can say what they contain.
• Analyst isolation: the team works apart from the business for weeks. The tell: long silences, then a data dump that nobody asked for.
• Conflicting numbers: the engagement or headcount base differs between teams. The tell: leaders argue over whose figure is right instead of discussing the finding.
• HR-owns-it framing: engagement treated as an HR deliverable rather than a shared task with managers. The tell: results go to HR and no manager has a named action.

• ▸Business question: the one talent question this analysis answers (e.g., how engaged are our top performers in key strategies?).
• ▸Decision it informs: the action or budget that changes depending on the answer.
• ▸Data sources: list survey scores, survey comments, and behavioral or outcome signals, each with owner and refresh rate.
• ▸Outcome to predict: the observable result engagement is linked to (e.g., leaving, productivity, quality).
• ▸Segments compared: groups to split by (e.g., tenure bands, top performers).
• ▸Definitions agreed: how 'engaged' is defined, which population and period, and who signed off from other teams.
• ▸Reconciliation check: other teams' figures compared and any gaps explained.
• ▸Business-ready finding: one sentence, one chart, one recommended action.
• ▸Action owner: the named manager and the shared role of HR/talent management and employee.

1. 1Start with a business question, not a survey. Follow the People Analytics Virtuous Process: question first, then data integration, then analytics, then actionable insight. Write the question in outcome terms, for example: 'How does employee well-being impact productivity, customer lifetime value, and customer up-sell and retention?' or 'What is the impact of employee satisfaction on customer satisfaction and business performance?'
2. 2Break the construct into its components before you measure: morale, sense of welcome, support, connection, and competence, during onboarding and beyond. Assign each component at least one data source, so 'satisfaction' is not read off a single score.
3. 3Build the descriptive layer ('What happened?'). Use the talent equivalents of customer reporting: a talent dashboard, a talent scorecard with KPIs, and triggers and alerts. Add an employee satisfaction survey and 'voice of employee' analysis as the 'What's happening and why?' layer.
4. 4Mine the unstructured data. Run text analytics on open-ended engagement survey comments. The source describes this as an untapped resource for people strategy and business strategy, and it is where the 'why' behind a low score sits.
5. 5Segment before you average. Use talent segmentation and talent profile analysis for turnover and retention. Compare cohorts, such as long-tenured employees versus those new to the organization, rather than reporting one company-wide satisfaction number.
6. 6Measure the early window separately. Treat onboarding as a staged process with its own early employee needs, and pair satisfaction readings with outcome measures such as time to productivity. This tests the 'competence' and 'welcome' components directly.
7. 7Integrate talent data with business data (company and market data) and move to prediction. Predict voluntary attrition ('talent attrition') from satisfaction signals. On the wellness side, ask which employees are more likely to be injured and what well-being, health, and safety do to performance and bottom-line results.
8. 8Close the loop with the IMPACT Cycle's 'track outcomes' step. Record what was done, what the impact was, and what new critical questions arose. Then review these with business partners on a set cadence.

The passages describe a company that saw its unstructured employee engagement survey data as an untapped resource for both people strategy and business strategy. It searched seven months for a data leader with a statistics degree and computer science experience, hired him, and put its four existing HR analysts under him. His first assignment was to compare long-tenured employees with those new to the organization, to learn how to target prospective talent for long-term engagement and a return on recruiting dollars. The passage is cut off before any results, so none are claimed here. The design is still instructive. The cohort split is the segmentation step. The unstructured survey text supplies the 'why' behind the scores. The stated aim, engagement that lasts and a return on recruiting spend, ties the work to a business outcome. To finish it, you would add an outcome measure, such as attrition or time to productivity, per cohort. You would also record what action followed and what changed, as the IMPACT Cycle requires.

• If you cannot state the business outcome (productivity, customer satisfaction, retention, lifetime value) the satisfaction data should explain, then do not start analysis. Return to the question step first.
• If the only data you have is a survey score, then add unstructured comment analysis and cohort segmentation before drawing conclusions. A score tells you what, not why.
• If new hires and long-tenured staff are pooled in one average, then split them. Early-tenure needs (onboarding) differ from established-employee needs.
• If satisfaction or well-being findings do not link to a business metric, then you do not yet have a case for executives. Connect the talent data to business data first.
• If you are early in your analytics journey, then start small and master the human capital data you already have. Consult experts in traditional analytics to plan the approach.

• Treating satisfaction as one survey number. The tell: the report has a single average and no cohort, text, or outcome breakdown.
• Collecting survey data and never analyzing the free text. The tell: comments sit unread while decisions rest on scores alone.
• Hiring analytical talent without a defined question. The tell: a long search and a hire, but the first task is a vague data exploration rather than a tied business question.
• Measuring without follow-up. The tell: no record of what action was taken, what impact it had, or what new questions emerged after the last survey.
• Keeping wellness, health, and safety separate from performance analysis. The tell: injury or well-being data is never linked to productivity or bottom-line results.

• ▸Business question: state the outcome the analysis must explain (e.g., effect of well-being on productivity or retention).
• ▸Construct components tracked: list which of morale, welcome, support, connection, and competence you will measure.
• ▸Data sources: survey, open-text comments, dashboard/scorecard KPIs, wellness/safety data.
• ▸Segments compared: define cohorts (e.g., long-tenured vs new hires) and the reason for each split.
• ▸Early-window measure: name the onboarding stage and outcome measure (e.g., time to productivity).
• ▸Linked business metric: the business data joined to the talent data.
• ▸Predictive target: the outcome to forecast (e.g., voluntary attrition, injury likelihood).
• ▸Action taken: what was done in response to the finding.
• ▸Outcome and next question: impact observed, plus the new question for business partners.

1. 1Start with one business question, not a menu of perks. The source frames every pillar around a question such as whom to attract, hire, engage, or retain. For this pillar, write the question as a sentence that links well-being or safety to an outcome you already track, for example: 'Which employee groups are most likely to leave or be injured, and what can we do before it happens?'
2. 2Name two owners before you build anything. The SAS Institute program was described by its Vice President of HR (Jenn Mann) and its Chief Health Officer (Gale Adcock). The safety story comes from the Workplace Safety and Insurance Board's Vice President and Chief Statistician (Eugene Wen). Pair an HR or health owner with an analytical owner so the program and the measurement are designed together.
3. 3Inventory data in the source's three buckets. Talent data includes workforce well-being and overall wellness, plus performance, engagement, turnover, compensation and benefits, and productivity. Company data includes sales, revenue, loyalty, churn, and Net Promoter Score. Labor market data includes Bureau of Labor Statistics turnover, wages, and job openings. List which fields you hold in each bucket and which wellness or safety fields are missing.
4. 4Link each data stream to your question and keep only the fields that bear on it. The source says this linking is how you see which data points are key to an actionable plan. Drop fields that do not connect, such as unrelated company metrics.
5. 5Move from description to prediction. The source says to use predictive models to anticipate what will happen with your employee base and to put proactive plans in place. Frame the model target as a future event, such as a safety claim, sustained disengagement, or attrition, and test whether wellness and health indicators help predict it.
6. 6Turn each prediction into a measurable goal and a specific action. Write the trigger, the intervention (preventive care, lifestyle support, or a safer environment), the owner, and the metric that will show it worked. A prediction without an assigned action is only a report.
7. 7Review the program against the chapter's own structure: what wellness means for your firm, the business case, best practices, and predictive optimization. Re-run the model on a set cadence and revise the interventions. Check your data collection against the privacy backlash the book flags in its future-of-analytics chapter.
8. 8Read Chapter 12 (pp. 307–328) for the SAS and WSIB interview details before committing budget. The supplied passages contain only the outline and names, not the program mechanics.

The supplied passages name two real cases but give no figures or program details. They are the SAS Institute Employee Wellness and Health Program, discussed with its VP of HR and Chief Health Officer, and a workforce-safety success story from Eugene Wen of Ontario's Workplace Safety and Insurance Board. You can still show the method the source prescribes, using a hypothetical firm with no numbers invented. A manufacturer asks which plant employees are most likely to leave or be hurt in the next year. It pulls talent data (wellness and well-being indicators, engagement, turnover, productivity) and company data (output, churn), then adds labor market turnover and job-opening rates for context. It links only the fields that bear on that question. A predictive model flags higher-risk groups. HR and the health lead assign preventive actions to those groups, each with a named owner and a measurable goal. The model is re-run each cycle to see whether the flagged risk moved. For the real mechanics of SAS and WSIB, use Chapter 12.

• If you cannot state the wellness or safety question as a business outcome (retention, engagement, injury, loyalty), then do not buy or launch the program yet.
• If wellness or well-being data is absent from your talent data, then collect it before modeling. Do not substitute proxies silently.
• If a data field does not connect to your stated question, then exclude it. The source's test is whether the stream links to the question.
• If a model flags risk but no owner or action is attached, then treat the output as unfinished and assign an intervention before sharing it.
• If employees may read the data collection as surveillance, then settle what you collect and why before expanding it. The source lists employee data privacy backlash as a coming trend.

• Running a perks menu with no analytical owner. The tell is that no one can say which outcome the program is meant to move.
• Reporting past participation or claims only, never predicting. The tell is that dashboards describe history and no proactive plan exists.
• Treating wellness as an HR-only project. The tell is that no health or safety expert and no statistician co-owns the work, unlike the SAS and WSIB pairings the source cites.
• Piling on data from all three buckets without linking it to a question. The tell is a large dataset and no clear action plan.
• Ignoring employee privacy concerns. The tell is that employees resist or distrust collection and participation falls.

• ▸Business question: write one sentence tying well-being or safety to an outcome (retain, engage, injury).
• ▸HR/health owner: name the person accountable for the program.
• ▸Analytics owner: name the person accountable for the data and model.
• ▸Talent data available: list the wellness, well-being, engagement, turnover, and productivity fields you hold.
• ▸Company and labor market data linked: list only the fields that connect to the question.
• ▸Predicted event and horizon: state what you are forecasting and over what period.
• ▸Intervention per risk group: name the preventive-care, lifestyle, or safe-environment action tied to each flagged group.
• ▸Measurable goal and review date: give the metric, target, and next model re-run date.
• ▸Privacy check: state what employee data is collected, why, and how it is communicated.

1. 1Start from a business question, not an HR metric. Write the one output or revenue problem you need answered (for example, 'how do we raise plant output without losing quality?'). The source stresses that business questions ignore your org chart and demand answers regardless of where the data sits.
2. 2Link each role's individual objectives to a company objective. Write the chain explicitly: company objective, team objective, individual objective. Drop any objective you cannot trace upward, because it does not measure contribution to revenue or output.
3. 3Define performance measures per role in business-relevant units: output, quality, and other outcomes tied to the objective. The preface names productivity, quality, engagement, actions and sentiment as data now available in near real time. Pick the few that match the objective.
4. 4Split productivity into three measurable phases: time to productivity (ramp), fully ramped performance level, and ongoing performance against objectives. Record a start date and a 'ramped' threshold for every role so ramp can be measured.
5. 5Pull the data into one view. Employee data sits in finance, HR systems and IT. Join hire date, role, objective attainment, output and quality, training history and exit date on a shared employee ID.
6. 6Before blaming skills or equipment, test the other drivers of low output: turnover, ongoing training and development, and onboarding quality. In the Piper case, management blamed equipment, layout and frontline quality control. The root cause was poor employment levels and high turnover.
7. 7Segment results by cohort (hire period, site or unit, manager, training exposure). Look for where ramp is slow or ramped performance is low, then build a predictive onboarding model on those patterns.
8. 8Close the loop with incentives and promotion. Connect the performance measures to reward and career-path decisions, and report findings to senior management in terms of the business objective, not HR activity.

Piper, a window and door manufacturer, wanted more market share by using its capacity fully, so it needed more plant output without losing quality. Management believed the problem was poor productivity and quality control by frontline staff, caused by equipment use and layout. The UK Department of Trade analysis confirmed that productivity and quality were serious threats. It traced them instead to poor levels of employment and high staff turnover. The company had won regional awards for training, but that training was initial training for new hires only, with no ongoing training or staff development. Leadership stayed convinced that better use of operations was the answer and did not accept turnover as a threat. Piper filed for bankruptcy in late 2013. The lesson is to measure ramp and ongoing performance by cohort and tenure, and to test turnover and development as causes before attributing low output to frontline effort or equipment.

• If output or quality is falling, then test turnover and ongoing development as causes before blaming equipment, layout or frontline effort.
• If you cannot trace an individual objective to a company objective, then drop it or rewrite it before measuring it.
• If new hires reach the ramped threshold slowly, then examine onboarding and early employee needs first, not hiring volume.
• If the data sits in finance, HR and IT, then join it under one employee ID before modeling. Do not wait for a single data scientist to solve it.
• If leadership resists a finding that contradicts its belief, then present it as a business-outcome result (output, quality) and name the cost, rather than as an HR metric.

• Attributing low productivity to staff effort or equipment without testing workforce causes. The tell: the fix list contains only operations changes while turnover is high and unexamined.
• Treating training as a one-time event. The tell: awards or programs exist for new-hire training, but nothing exists after onboarding, as at Piper.
• Measuring activity instead of contribution. The tell: metrics cannot be traced to a company objective or to output and revenue.
• Relying on gut feel and 'that's worked in the past'. The tell: performance decisions cite experience rather than data on ramp time and ramped output.
• Expecting one hire to fix it. The tell: you have a data scientist but fragmented data and no agreed performance definitions.

• ▸Business question: the output or revenue problem in one sentence.
• ▸Company objective: the objective the role must serve.
• ▸Individual objective: the role's objective, with its traceable link to the company objective.
• ▸Performance measures: two or three business-relevant measures (output, quality, other outcomes) and their data source.
• ▸Ramp definition: start date and the threshold that counts as 'ramped'.
• ▸Time to productivity: observed time to reach the threshold, by hire cohort.
• ▸Ramped performance level: typical performance of fully ramped employees in the role.
• ▸Ongoing development: training or development offered after onboarding, or 'none'.
• ▸Turnover signal: exits in the cohort or unit, and whether leadership acknowledges them as a risk.
• ▸Next action and owner: the specific change to test and who runs it.

1. 1Start from a business question, not a model. The source defines effective People Analytics as analytics that start with a strategic human capital goal or question. Write one sentence such as 'Which high performers are likely to leave, and what does their exit cost us?' before you pull any data.
2. 2Define retention before you measure it. The retention chapter opens with 'What Is Retention, Anyway?'. Decide in writing who counts as a leaver (voluntary only) and who counts as high-value (your performance-assessment and lifetime-value pillar data). That gives you regretted turnover rather than a blended headcount rate.
3. 3Set a tenure baseline by cohort before you diagnose anything. Median job tenure has fallen across all age groups for at least 50 years, so compare your people against their own age and birth cohort, not against a historical company average. A one-year tenure among 20- to 30-year-olds is not by itself a warning sign, because that is the national median for the 1993 cohort.
4. 4Build the business case in cost terms. The source says turnover is often a substantial business cost. Use the employee lifetime value and cost modeling pillar to price the loss of each high-value segment, so retention spend can be justified against a number.
5. 5Assemble the attrition data set from the other pillars. Retention depends on engagement, onboarding and culture fit, performance, and wellness data, not just exit records. Join survey results, early-tenure outcomes, performance history, and employment history to each employee record.
6. 6Move from a reactive to a proactive attrition model. Predict who is at risk of leaving, rather than reading exit interviews afterward. The source says traditional approaches are failing and points toward predictive models that flag risk before resignation.
7. 7Segment the model output before acting. Group at-risk employees by role, performance tier, and driver of risk. Match each segment to a different intervention instead of one organization-wide program.
8. 8Close the loop. Put the findings into manager and HR action, then check afterward whether regretted turnover moved in the targeted segments. Use that result to refine the model (the IMPACT cycle logic).

The source gives a real baseline for reading tenure. The Federal Reserve Bank of Atlanta, using Current Population Survey data, found that median job tenure for 20- to 30-year-olds was four years for those born in 1953 and one year for those born in 1993. Among 41- to 50-year-olds, tenure fell by five years between 'Depression babies' (born from 1933) and Gen Xers (born from 1973). Suppose a leader says 'millennials are our retention problem.' The data says to check first, because the decline appears in every age group. Your first move is to segment your own leavers by age cohort and compare each to its cohort's norm. If older cohorts are leaving earlier than their cohort baseline, the cause lies elsewhere. Wharton's Adam Cobb offers one candidate: since the 1980s, healthy firms laid off workers and shifted risks such as pensions and health costs onto employees, so workers stopped letting firms control their careers. That points you toward organization-level drivers (security, benefits, career paths) rather than a generational fix. You then build the attrition model on those drivers and weight it toward your high performers.

• If a leader attributes turnover to one generation, then compare each cohort's tenure with its national cohort baseline first, since tenure decline is across the board.
• If an exit is voluntary and the person was a high performer, then count it as regretted turnover and price it with lifetime value. Do not average it with low-performer exits.
• If you only have exit-interview data, then treat your retention work as reactive. Add engagement, onboarding, and performance data to move to prediction.
• If the model flags a large at-risk group, then segment by role, performance tier, and risk driver before choosing interventions. One program for everyone dilutes spend.
• If no human capital business question is attached to the analysis, then stop and write one. Analytics without a strategic question does not meet the source's definition of effective.

• Blaming millennials for turnover. The tell is a retention plan aimed at one age group while tenure is shortening in every cohort.
• Measuring total turnover instead of regretted turnover. The tell is a flat or improving rate while top performers keep leaving.
• Relying on exit interviews and annual surveys alone. The tell is that you learn why people left only after the resignation, and managers are surprised each time.
• Running retention analytics as an isolated project. The tell is a model with no engagement, onboarding, performance, or cost data feeding it, and no one assigned to act on its output.
• Producing a risk score with no segmentation or follow-through. The tell is a dashboard that lists names but triggers no differentiated action and no later check on whether turnover changed.

• ▸Business question: one sentence naming the strategic human capital goal this retention work serves.
• ▸Leaver definition: state who counts as voluntary and who counts as high-value, and where each flag comes from.
• ▸Cohort tenure baseline: for each age or birth cohort in the segment, record your median tenure next to the national cohort median.
• ▸Cost of loss: the lifetime value or replacement cost estimate for one high-value departure, with its source.
• ▸Data sources joined: list engagement, onboarding, performance, and wellness data feeding the attrition model.
• ▸Risk segments: name each segment by role, performance tier, and main risk driver.
• ▸Intervention per segment: the specific action, its owner, and a start date.
• ▸Check measure: the regretted-turnover figure you will compare against, and the review date.
• ▸Model refresh note: what you will change in the model based on the review result.

1. 1Split the construct into the three outcomes in the definition: performance, fit, and retention. Write one measurable indicator for each before you score anyone. The source passages name 'Time to Productivity and Other Outcome Measures' as the onboarding-stage output, so use time to productivity as your first performance indicator.
2. 2Track each hire from the first moment they hear about your company, not from the offer. The passages describe applying customer life cycle management to employees: follow candidates and new hires from first awareness through the employee life cycle (the Nextel case). Tag every hire with sourcing channel, application path, assessment results, interview outcome, and onboarding record so outcomes trace back to the decision that produced them.
3. 3Capture selection signals while you hire. The passages place 'Preinterview Assessment Analytics', interviews, and 'Predictive Analytics for Selection' inside talent acquisition. Store assessment scores and interview ratings in a form you can later compare with performance and tenure. Unrecorded gut feel cannot be validated.
4. 4Set the retention window and a cohort benchmark before you judge anyone. Job tenure has fallen across all age groups. The Atlanta Fed data show median tenure of 4 years for 20-to-30-year-olds born in 1953 versus 1 year for those born in 1993. Compare new hires with their own age cohort, not with a historical norm that no longer applies.
5. 5Measure onboarding as its own link in the chain. Capture early-employee needs and time to productivity, then build an onboarding predictive model, as the source's OPEN framework chapter suggests. Check what happens after initial training: whether ongoing development exists, not only whether onboarding was completed.
6. 6Roll outcomes up by sourcing channel, selection method, and hiring manager, then trace weak results back to the decision that created them. Quality of Hire is a verdict on sourcing, selection, and onboarding, so a low score has to point at one of the three.
7. 7Convert results into decisions at the individual level. When someone announces they are leaving, look at that person's expected lifetime value before deciding on a counter-offer, as the Harris interview describes. Feed the pattern of leavers back into sourcing and selection criteria.
8. 8Repeat per cohort. Re-score each hiring cohort at the end of your retention window and compare cohorts, so the metric tests your sourcing and selection changes over time.

The source's clearest case is Piper, a UK manufacturer. Management believed its growth problems came from poor productivity and quality control by frontline staff, and blamed equipment use and layout. The UK Department of Trade was brought in, and analysis found the basic cause was poor levels of employment and high staff turnover. The company had won regional awards for training, but all of it was initial training for new hires, with no ongoing training or staff development. Read through Quality of Hire: the onboarding looked strong, yet hires did not remain or perform, so the metric would have scored the whole sourcing-to-onboarding chain as failing. Leadership never accepted turnover as a threat, kept pursuing operational fixes, and the company filed for bankruptcy in late 2013. The lesson is to measure retention and performance outcomes by cohort, not onboarding activity or awards, and to treat a high-turnover signal as a first-order finding.

• If new hires leave early despite a well-regarded onboarding program, then audit what follows onboarding (ongoing training and development) before redesigning sourcing. Piper won training awards on initial training alone and still failed on turnover.
• If a hire's tenure looks short, then compare it with the median for their age cohort before calling it a bad hire. Median tenure for 20-to-30-year-olds fell from 4 years (born 1953) to 1 year (born 1993).
• If a valued employee announces they are leaving, then check their expected lifetime value before making a counter-offer, rather than reacting on emotion.
• If you cannot link a hire's outcome back to sourcing channel, assessment score, and interview rating, then fix the tracking first. You cannot attribute quality to a decision you did not record.
• If leadership attributes poor output to equipment or layout while turnover is high, then put the turnover and hiring data in front of them first. Piper's leadership resisted that evidence, and it hurt the company.

• Scoring onboarding completion instead of outcomes. The tell: onboarding or training awards are cited as proof of hiring quality while turnover stays high (the Piper pattern).
• Defining quality as performance alone. The tell: top-rated hires are gone within a year, because fit and retention were never measured.
• Benchmarking tenure against old norms or one generation. The tell: you blame millennials for turnover, but tenure has declined across age groups, so the benchmark is wrong.
• Relying on gut feel and 'that's worked in the past'. The tell: no assessment or interview data is stored in a form that can be tested against later performance, so selection cannot be validated or improved.
• Starting measurement at the offer. The tell: you cannot say which channel or first touchpoint produced your best hires, so sourcing spend (US employers spend over $400 billion a year on locating, securing, and holding workers) goes unjudged.

• ▸Cohort and sourcing channel: name the hiring group (role, period) and the channel that produced them.
• ▸Selection signals recorded: list assessment scores and interview ratings stored for each hire.
• ▸Performance indicator and result: state the outcome measure (e.g., time to productivity) and the cohort's result.
• ▸Fit indicator and result: state how fit with culture was measured and the cohort's result.
• ▸Retention window and result: state the window and the share of the cohort still employed at its end.
• ▸Age-cohort tenure benchmark: note the median tenure for the hires' age group you are comparing against.
• ▸Onboarding and post-onboarding support: list early-needs touchpoints and any ongoing development after initial training.
• ▸Weakest link: name whether sourcing, selection, or onboarding best explains the gap.
• ▸Action and owner: state the single change to make and who owns it before the next cohort is scored.

1. 1Fix the unit of analysis as one role at one level, for example Underwriter I, Teller II, or Customer Service III. The source limits ELTV to one role at a time, so a blended 'all customer service' value is the wrong unit.
2. 2Write the definition on the model sheet before building anything: 'a prediction of the net profit attributed to an employee through the employee's tenure in a given role.' It is the customer-lifetime-value definition ('net profit attributed to the entire future relationship') narrowed to tenure in a role.
3. 3Build the three input curves for that role, indexed by tenure. The cost curve covers what the employee costs over time. The performance curve covers the value the employee produces over time. The attrition curve gives the probability the employee has left the role by each point in tenure. The source treats these as metrics/curves, and ELTV is built from them.
4. 4Net cost against performance at each tenure point to get expected net profit if the employee is still in the role. The source says ELTV is a net-profit figure, so keep cost and performance in the same units before netting.
5. 5Weight each tenure point's net profit by the probability the employee is still in the role, using the attrition curve, then sum across tenure. This is the 'probability-weighted by the risk of attrition from the role' step. Because it is risk-weighted, the number is not a best-case projection.
6. 6Do not model promotion or career path. If a person moves, for example from Customer Service III to Internal Sales IV, stop the old calculation and start a new ELTV for the new role with its own cost, performance, and attrition curves.
7. 7Use ELTV as a roll-up number to compare programs or scenarios. Run the same role with and without the intervention, change only the curve the program is expected to move (cost, performance, or attrition), and compare the resulting ELTV figures. The source calls this a direct tie to the bottom line.
8. 8Re-estimate the curves routinely and feed the results into planning. The source envisions dashboards that report cost, performance, attrition, and lifetime value, with the curves used routinely in planning.

The source gives no figures, so this example uses the source's own role structure and shows the logic without invented numbers. Take the role Customer Service III. You build its cost curve, performance curve, and attrition curve across tenure. At each tenure point you net performance against cost and multiply by the probability the person is still in the role. The sum is the ELTV for Customer Service III. Now suppose a program is meant to lower attrition from that role. You hold the cost and performance curves fixed, swap in the lower attrition curve, and recompute. The difference between the two ELTVs is the program's value for that role, expressed in net profit. Suppose a top performer then moves to Internal Sales IV. The Customer Service III calculation ends, and a separate ELTV starts for Internal Sales IV with its own curves. You do not try to predict that move in advance, because the source explicitly excludes promotion-path prediction.

• If an employee changes role or level (for example Customer Service III to Internal Sales IV), then start a new ELTV with new curves. Do not carry the old value forward.
• If you are comparing programs (retention, training, compensation), then compare risk-weighted ELTV for the same role under each scenario, not raw cost or raw performance averages.
• If a proposed model needs to forecast promotions or the employee's 'ultimate journey', then cut it back to the single role. The source scopes ELTV to one role at a time.
• If a scenario changes only one curve (say attrition), then hold the other two constant so the ELTV difference is attributable to that change.
• If someone presents ELTV without an attrition weighting, then treat it as an overstated value, because ELTV is defined as probability-weighted by attrition risk.

• Blending several roles or levels into one ELTV. Tell: the cost and performance curves have no single role title attached.
• Counting value the employee would produce only if they stay, with no attrition weighting. Tell: the figure is the same whether the role has high or low turnover.
• Folding expected promotions into the calculation. Tell: the model assumes the person moves up, so tenure in the role is no longer what is being valued.
• Treating the employee only as a cost line. Tell: the model has a cost curve but no performance curve, which ignores the source's 'costs or assets' framing.
• Reporting ELTV once and filing it. Tell: no routine refresh and no use in planning or scenario comparison, so it never ties to the bottom line.

• ▸Role and level: one title only, such as Underwriter I or Customer Service III.
• ▸Tenure window: the span of tenure in this role the curves cover.
• ▸Cost curve: employee cost at each tenure point, with its source.
• ▸Performance curve: value produced at each tenure point, in the same units as cost.
• ▸Attrition curve: probability the employee is still in the role at each tenure point.
• ▸Net profit per tenure point: performance minus cost.
• ▸Risk-weighted ELTV: net profit at each point times the probability of still being in the role, summed.
• ▸Scenario being tested: which one curve the program changes, and the ELTV before and after.
• ▸Role-change trigger: the event (new role or level) that closes this sheet and opens a new one.

1. 1Write the organizational purpose first, in terms of the organization's role, before choosing any metric. The foreword says a goal like 'increase market share by x percent' is a wish, not a purpose. Analytics only has a target once purpose and strategic plan are clear.
2. 2Pick the business outcomes you will be held to from the construct's own list: revenue, operating income, customer satisfaction, cost savings, growth. Choose the few that your purpose and plan actually depend on, and define each as a number with a source system and an owner.
3. 3Map each outcome to the talent pillar most likely to move it. The seven pillars are workforce planning, sourcing, acquisition/hiring, onboarding/culture fit/engagement, performance and employee lifetime value, churn and retention, and wellness/health/safety. Write the hypothesis as one sentence, for example 'retention in group X affects outcome Y.' Treat it as untested until the data supports it.
4. 4Link individual objectives to company objectives (the Chapter 9 approach). Every performance measure you track at the person or team level should trace to one of your chosen business outcomes. Drop measures that trace to none.
5. 5Run the three analytics levels in order. Descriptive: what has happened to the outcome and to the talent metric. Predictive: what must be done to reach the future goal. Prescriptive: how to do it. Do not skip to prescriptive without a descriptive baseline.
6. 6Cost the talent side as well as the outcome side. Use employee lifetime value and cost modeling (employees as asset vs. cost), the retention business case, and outcome measures such as time to productivity. This turns 'talent improved' into a return on workforce investment.
7. 7Report the talent-to-outcome link to senior management in business terms and review it on a fixed cycle (the book's IMPACT cycle). Keep the framework in a dedicated analytics function or center of excellence so the work persists past one project.

The passages give no figures for a business-performance result, and the case studies (STM workforce planning, WSIB workforce safety) are named only in acknowledgments, so no result is quoted here. What the foreword does supply is a usable pattern: the doctor analogy. A patient reports sinus congestion; that is descriptive. The doctor applies knowledge to decide what treatment will cure it; that is predictive. The paper taken to the pharmacy says how to administer it; that is prescriptive. Apply the same sequence to a business outcome. Descriptive: 'customer satisfaction fell in the units where turnover rose.' Predictive: a retention model shows which segments must be retained to restore it. Prescriptive: a specific retention action for those segments, with an owner and a review date. Then measure the outcome you named at the start, not the activity.

• If you cannot state the organization's purpose and the business outcome a talent initiative serves, then stop and define them before building any model.
• If a talent metric traces to none of your chosen outcomes (revenue, operating income, customer satisfaction, cost savings, growth), then drop it from the business-performance report.
• If you only have descriptive data, then report it as a baseline and do not claim cause. Move to predictive modeling before recommending action.
• If a result relies on anecdote or instinct, then label it as such. The endorsements' point is that line management will no longer accept it.
• If the audience is senior management, then present outcome movement and workforce ROI first and pillar-level detail second.

• Treating a target as a purpose. Tell: the goal is 'grow share by x percent' with no statement of why the organization exists or what talent it needs.
• Reporting HR activity as performance. Tell: dashboards show hires, surveys, and programs run, but none of the five outcomes.
• Stopping at descriptive analytics. Tell: you can explain what happened but cannot say what to do or by how much it would matter.
• Counting talent only as cost. Tell: no lifetime value or retention business case, and decisions default to cutting spend.
• Claiming a result without a baseline or an owner. Tell: nobody can name the starting value, the data source, or who answers for the outcome.

• ▸Organizational purpose: one sentence on the organization's role, not a target.
• ▸Business outcome: choose one of revenue, operating income, customer satisfaction, cost savings, growth.
• ▸Outcome definition and source: the exact measure, system of record, and named owner.
• ▸Talent pillar: which of the seven pillars is hypothesized to move it.
• ▸Linking hypothesis: one sentence tying the talent metric to the outcome.
• ▸Descriptive baseline: current values of both the talent metric and the outcome, with dates.
• ▸Predictive question: what must change in the talent metric to reach the outcome goal.
• ▸Prescriptive action: the specific action, who runs it, and the review date.
• ▸Cost and return view: lifetime value or cost of the talent investment against the outcome change.

1. 1Reframe the problem before measuring it. The source says there is no shortage of qualified people for any one company: 'the talent ocean has not been overfished for the needs of your company.' The problem is that many boats are fishing, and there is a shortage of twenty-first-century acquisition strategies. Write your hard-to-fill roles as 'we lose the competition for this pool' or 'our approach fails', not 'there are no people.'
2. 2Anchor on purpose and a business question first. Per the foreword, recruitment starts with the organization's purpose, goals and strategic plans, not with the employment section of HR. A target like 'raise market share x percent' is only a wish. Write the 3-5 business questions that labor market data must answer, and prioritize them. Otherwise you drift into analysis paralysis and intellectual curiosities.
3. 3Mine the data you already hold before buying anything. Harris says there is a lot of value in the data already in your systems. Start by describing what has happened: where hard-to-fill roles sit, where attrition occurs, which sources have delivered.
4. 4Layer external data on top, in the order Harris lists: publicly available talent data, labor market data, and benchmarked data from other companies. Build a combined view across employee research, employee behavior, web analytics, business reporting, competitive intelligence, and economic and labor market research. The business question ignores whether the data sits in finance, HR or IT.
5. 5Cover every stage of the talent cycle, not only open requisitions. The strategy has to include potential employees working elsewhere, candidates, current employees, and former employees (alumni, including retirees). Each group is a different slice of the market you compete in.
6. 6Move from descriptive to predictive to prescriptive. Descriptive says what has happened. Predictive says what must be done to reach future goals. Prescriptive says how to do it. The authors hold that People Analytics is most useful when forward-looking, so use market conditions to project likely outcomes (for example, where retention or acquisition will get harder), not only to report them.
7. 7Think big, start small. Choose one high-impact area, such as one critical role family. Show demonstrable value to that internal customer, learn from the data, and use it to improve the process. Then widen scope.
8. 8Account for the cost side. Help wanted advertising alone costs employers almost $20 billion per year. Treat sourcing spend as an investment to be targeted by analytics, since talent analytics ensures investment in talent is 'well targeted and delivering the desired results.'

A leadership team says: 'We can't hire because there's a talent shortage.' Applying the source's frame, the analyst first challenges the premise. Qualified people exist; the company is one of many boats fishing the same ocean. Step one is the business question, such as 'Which critical roles must we fill to deliver this year's strategic goals, and how are we losing them?' Step two is descriptive: pull what the company already holds on those roles (sources, hiring outcomes, attrition). Step three adds external data in Harris's order: public talent data, labor market data, and benchmarks from other companies. This is the kind of measurement Jesse Harriott's Monster Employment Index represented, a labor market gauge tracked in more than 30 countries. Step four is predictive: given market conditions, where will filling or keeping these roles get harder? Step five is prescriptive: which sourcing and retention actions to take, including targeting potential employees who work elsewhere and alumni. The team pilots on one role family and uses the results to widen scope. The output changes from 'there's a shortage' to a specific, testable acquisition strategy. (The passages give no figures for this scenario; it applies their framework and invents no data.)

• If leaders attribute hiring pain to 'talent scarcity,' then test the claim against the source's premise: ask whether qualified people exist and you are losing a competition, before accepting a shortage.
• If you cannot name the business question a labor market dataset answers, then do not acquire it. Prioritize questions first to avoid analysis paralysis.
• If internal data on a role has not yet been explored, then mine it before adding external feeds; add public talent, labor market and benchmark data second.
• If a report only describes past market conditions, then extend it to a predictive view and a prescriptive action; backward-looking market reports have the least impact.
• If the scope of the effort is large or the team new to analytics, then pick one high-impact area, prove value there, and expand.

• Treating scarcity as a fact of the market rather than a strategy gap. The tell: every hiring miss is explained by 'the market' and no sourcing method has changed.
• Starting from the tools or the data instead of the purpose. The tell: dashboards of labor market data with no named business question, and goals phrased as wishes ('increase share by x percent').
• Siloed data. The tell: employee data in finance, HR systems and IT never joined with labor market research, competitive intelligence or web analytics, so answers are slow or inconsistent.
• Watching only open requisitions. The tell: no view of potential employees at other firms, alumni or retirees, so the pool is defined as 'whoever applies.'
• Betting on one genius. The tell: the plan hinges on hiring a single data scientist to fix everything; the source says this is wrong, as smart staff is necessary but not sufficient.

• ▸Business question: the one question this analysis must answer, tied to organizational purpose and goals.
• ▸Role family and why it is high-impact: the one area chosen to start small, and the demonstrable value to its internal customer.
• ▸Scarcity claim vs. evidence: the leaders' stated claim ('shortage of...') and the data that supports or refutes it.
• ▸Internal data inventory: what you already hold (sources, hiring outcomes, attrition) and the system or owner for each.
• ▸External data layers: publicly available talent data, labor market data, and benchmarks from other companies, with source named for each.
• ▸Talent cycle coverage: how you will address potential employees elsewhere, candidates, current employees, and alumni/retirees.
• ▸Descriptive finding: what has happened to date for this role family.
• ▸Predictive view: what the market conditions imply for future acquisition or retention difficulty.
• ▸Prescriptive action and owner: the specific sourcing or retention action, who owns it, and the review date.

1. 1Start from the business question, not the data. Write the question in strategy terms, such as 'How does talent affect execution of our business's strategy?' or 'What capabilities are necessary for our business to succeed?' Business acumen outranks analytics acumen. Reject any project you cannot phrase this way.
2. 2Anchor the question in organizational purpose and strategic plans. The Foreword says everything starts there, and that 'increase market share by x percent' is a wish, not a purpose. Write the purpose first, then pick the talent questions that serve it.
3. 3Size the gap honestly before you ask for money. Deloitte found 75 percent of respondents call talent analytics important, but only 8 percent rate their organization 'strong'. Only 7 percent rate themselves excellent at measuring and improving engagement and retention. Assess your own people, process, technology and sponsorship against that bar.
4. 4Sequence the analytics in three levels. Descriptive covers what has happened. Predictive covers what must be done to reach future goals. Prescriptive covers how to do it. Build the descriptive baseline first, such as hires, exits and engagement. Then build predictive models for selection, onboarding and retention. Then write the prescriptions.
5. 5Pick one lifecycle stage and integrate the data sources that stage needs. The lifecycle runs from workforce planning through sourcing, acquisition, onboarding, engagement, performance, lifetime value and retention. For acquisition, for example, connect application, preinterview assessment and interview data to later performance and tenure, so selection can be predicted rather than guessed.
6. 6Secure executive sponsorship by translating value into language leaders embrace. Stakeholders rarely equate data and insight with HR, and the source warns that being misunderstood usually ends in defunding and dismantling. Show the effect on attracting, engaging, developing and retaining key talent, on optimizing people investments, and on the business impact of HR programs.
7. 7Staff the capability beyond analysts. Smart, knowledgeable staff is necessary but nowhere near enough. Name an owner for each of the five elements: people, process, technology, sponsor and business alignment.
8. 8Report forward-looking results against a business metric, such as the workforce ROI of a talent investment. Use findings to retarget spend, then pick the next lifecycle stage.

The source's own example is the practitioner Berry, who applied fact-based decision making across a 20-plus-year HR career. One question he addressed was the probability that an employee misses work on days contiguous to a two- or three-day scheduled time off. This is a forward-looking question: it predicts an event rather than reporting past absence. To run it, you would pull scheduled time-off records and attendance records into one dataset and estimate absence probability for those adjacent days. Operations could then act on the prediction, for example through coverage planning. Berry's caution on sponsorship applies here. If stakeholders do not understand the science, the effort gets defunded, so you present the result as a business outcome. The Deloitte context shows why this matters: most organizations call talent analytics important, yet only 8 percent consider themselves strong at it.

• If a proposed analysis cannot be phrased as a business question about strategy execution, then drop or reframe it before touching data.
• If your outputs only describe the past, then add a predictive layer and a prescriptive action step before you present them to leaders.
• If stakeholders do not grasp what the analytics does, then explain and demonstrate the value in business language before expanding scope, since misunderstanding leads to defunding.
• If you have talented analysts but no results, then check process, technology, sponsorship and alignment, because staff alone is nowhere near enough.
• If the stated purpose is a number target such as market share, then restate the organization's purpose first, since a target alone is a wish and cannot guide talent strategy.

• Treating hiring smart analysts as the whole capability. Tell: skilled staff, yet no ROI and no data integration or sponsor.
• Leading with technique instead of business questions. Tell: impressive models whose results leaders cannot connect to strategy execution.
• Staying descriptive. Tell: dashboards of past turnover or engagement with no forecast and no recommended action.
• Failing to make the work understood to stakeholders. Tell: leaders say they 'don't get it', then budget is cut and the initiative is dismantled.
• Running on gut feeling and anecdote rather than collected data. Tell: talent decisions rest on interpersonal impressions that nobody can measure or test.

• ▸Business question: write it in strategy-execution terms, for example how talent affects execution of strategy.
• ▸Organizational purpose: state the purpose in terms of the organization's role, not a numeric target.
• ▸Lifecycle stage in scope: choose one, such as sourcing, acquisition, onboarding, engagement, performance or retention.
• ▸Data sources to integrate: list each source, its owner and its gaps.
• ▸Analytics level today: mark descriptive, predictive or prescriptive and note what is missing.
• ▸People: name the analyst and business-acumen owners, and who translates results for leaders.
• ▸Process and technology: describe how data flows from collection to action and which tools support it.
• ▸Executive sponsor and value message: name the sponsor and write the benefit in business language.
• ▸Success metric and next step: set the business or ROI metric and the follow-on action.

1. 1Write the planning question as five explicit dimensions: number, skills, place, time, cost. Each forecast line must state all five, for example 'N people with skill X, in location Y, by quarter Z, at a loaded cost of C'. A headcount-only forecast leaves out four of the five.
2. 2Anchor demand to the business goals, not to last year's headcount. List the current and future goals (growth, new products, new sites, service levels) and translate each into the roles and skills it requires. This keeps the plan tied to the business and not to HR's own history.
3. 3Build the supply side separately from the existing workforce: current headcount by skill and location, plus expected losses and movement. Use the retention and attrition pillar as an input, since the book treats the seven pillars as one connected framework. Do not net supply against demand until both are built independently.
4. 4Compute the gap per skill, place, and time period, not in total. Totals hide a surplus in one skill covering a shortage in another. Rank gaps by business impact and by cost of leaving them open.
5. 5Choose the closing lever for each gap: hire, develop, redeploy, or retain. Use the sourcing, acquisition, development, and retention pillars as the lever menu, and attach a cost to each lever so the 'right cost' dimension is tested, not assumed.
6. 6Replace gut feeling with data at each assumption. The source criticizes talent decisions that rest on 'gut feelings and immeasurable interpersonal relationships'. For each forecast input, record the data source and an owner. Mark any input with no data as an assumption to be tested.
7. 7Run the plan as a cycle, not a one-time report. The book organizes workforce planning around an IMPACT cycle and a dos-and-don'ts list. Fix a review cadence, compare forecast to actual supply and demand, and correct the model each time.
8. 8Hand the output to the other pillars as concrete requirements: roles, volumes, and dates for sourcing and hiring; at-risk skills for retention. Workforce planning is the first of the seven pillars, so its value shows up in the pillars downstream.

The book names a real workforce planning analytics success story from Société de Transport de Montréal (STM), shared by Christophe Paris, Josée Gauvreau, and Cedric Lepine. The supplied passages give no STM figures, so the numbers below are hypothetical and show the method only. Suppose a service goal requires 120 people with skill X at site A by Q3 (demand). Current staff with skill X at site A is 100, and the attrition view expects 8 to leave by Q3, so supply is 92. The gap is 28. A totals view might show the company over-staffed overall and report no problem. The skill-by-site view shows a real shortage. You then cost each lever: redeploy 10 from a surplus site, retain 3 of the 8 at-risk people, and hire 15. You hand the 15-hire requirement, with role, site, and date, to sourcing. At the next review you compare actual supply to the 92 forecast and adjust the attrition assumption.

• If a forecast states only a headcount, then send it back until skills, place, time, and cost are filled in, because the definition requires all five.
• If total supply equals total demand but skill-level gaps exist, then plan at skill-by-location level and treat the aggregate as irrelevant.
• If a gap can be closed by retention or redeployment at lower cost than external hiring, then route it away from sourcing first, since cost is one of the five dimensions.
• If an input rests on opinion rather than data, then label it an assumption, assign an owner to test it, and do not let it drive a large hiring commitment.
• If forecast and actual diverge at the review, then fix the model input before changing the plan, so the cycle improves accuracy over time.

• Planning on headcount alone. The tell: the plan has no skill or location columns and gaps surface only when a team cannot deliver.
• Projecting from history, not from business goals. The tell: next year's plan is last year's plus a percentage, and nobody can name the goal it supports.
• Netting supply against demand in aggregate. The tell: reported balance looks fine while managers still report unfillable roles.
• One-off planning with no cycle. The tell: the forecast is never compared with actuals and the same assumptions appear year after year.
• Planning in isolation from sourcing, acquisition, and retention. The tell: recruiters receive vague requests late, or attrition risk in key skills is missing from the supply view.

• ▸Business goal supported: name the current or future goal that creates this demand.
• ▸Skill / role: state the specific skill or role required.
• ▸Place and time: give the site and the period or date by which supply is needed.
• ▸Demand (number): the people required, with the data source or labeled assumption.
• ▸Current supply: today's headcount for this skill and place.
• ▸Expected change in supply: forecast losses and movements, with the source of the attrition estimate.
• ▸Gap: demand minus projected supply; rank by business impact.
• ▸Closing lever and cost: hire, develop, redeploy, or retain, with the estimated cost of each option considered.
• ▸Owner and review date: who owns the gap and when forecast is compared to actual.

1. 1Start from demand, not channels. Sourcing is the second pillar, after Workforce Planning Analytics and before Acquisition/Hiring. Pull the roles and volumes the workforce plan says you need, and judge each channel by whether it fills those roles, not by how much traffic it generates.
2. 2Inventory every channel you currently use: referrals, job boards, social media, college recruiting trips, open houses, advertising, and any Big Data sources. The foreword calls scanning job boards, campus trips, open houses and advertising obsolete as a differentiator, because everyone uses them and results come out similar rather than unique. Mark each channel as 'commodity' (all competitors use it) or 'proprietary' (data or access others lack).
3. 3Tag the source on every candidate record at first touch and keep the tag through application, hire and later outcomes. Without a persistent source tag you can only report volume, never source-of-hire performance.
4. 4Measure each channel on two sides: spend and downstream result. Spend is recruitment cost attributed to the channel. Result goes beyond hires to quality of hire (the book credits ERE Media's Rob Macintosh for input on this) and, where possible, retention and performance. Tying talent to business outcomes is the Preface's stated opportunity.
5. 5Add a mobile view. Chapter 5 treats mobile's impact on sourcing as its own topic, so split every channel metric by device or access path and check whether candidates reach you through mobile.
6. 6Test one Big Data source against a commodity channel. Chapter 5 positions Big Data and advanced technology as the way out of the 'similar results' problem. Run a bounded pilot with the same role, the same window and the same tagging, then compare the results to your baseline channel.
7. 7Run the full cycle using the IMPACT framework the chapter closes with ('Putting the IMPACT Cycle into Action'). The supplied text does not spell out its steps, so use the book's own version. In practice: baseline, change, measure, reallocate, repeat.
8. 8Replace gut-feel channel decisions with a written reallocation rule. The Preface says 'that's worked in the past' is no longer enough. Record each budget shift, the data behind it, and the date you will re-check it.

The source passages supply a diagnosis rather than a data set, so this walkthrough is illustrative and uses no invented figures. The foreword names the problem: job boards, campus trips, open houses and advertising give no competitive advantage, because every competitor uses them and the outcomes look alike. Take a team whose plan calls for one hard-to-fill role family. Step 1: tag all candidates for that role by source (referral, job board, social, campus, a Big Data pilot). Step 2: record spend per source, then trace each hire's quality of hire and retention. Step 3: label job boards and campus trips 'commodity' and check whether they merely match what competitors get. Step 4: if the Big Data pilot or referrals show better quality per dollar than the boards, shift budget toward them. If they do not, keep the baseline and change the pilot design. Step 5: re-measure next cycle. The decision rests on tagged data, not on what has worked in the past.

• If a channel is used by every competitor and its tagged results are no better than average, treat it as a cost to minimize, not an advantage to defend.
• If a channel produces volume but its hires lack quality-of-hire or retention evidence, do not scale its budget until you can trace those outcomes.
• If a new Big Data or technology source beats your baseline on quality per dollar over a comparable role and window, move budget to it. If not, keep the baseline and redesign the pilot.
• If the workforce plan does not call for a role family, do not fund sourcing for it, whatever the channel metrics show.
• If a recommendation rests only on 'it has worked before', demand the source-tagged data before approving the spend.

• Reporting volume instead of source-of-hire performance. The tell: dashboards show applicants per channel but no hires, quality or retention by channel.
• Relying on conventional channels everyone uses. The tell: your results look like your competitors' and no channel is distinctive, which is the foreword's 'similar rather than unique' problem.
• Missing or lost source tags. The tell: a large 'unknown' or 'direct' bucket, and hires you cannot trace back to a first touch.
• Ignoring mobile as a separate path. The tell: channel metrics are never split by device, though Chapter 5 gives mobile its own section.
• Making channel decisions on gut feel and interpersonal relationships. The tell: budget moves are explained by anecdote ('that's worked in the past') with no recorded data or review date.

• ▸Role family and workforce-plan demand: state the roles and volume the workforce plan requires.
• ▸Channel: name it (referral, job board, social media, campus, advertising, Big Data source, other).
• ▸Commodity or proprietary: mark whether competitors widely use it and whether your results differ from theirs.
• ▸Spend: total recruitment cost attributed to this channel for the period.
• ▸Candidates and hires by source tag: counts at first touch and at hire, plus the share of untagged records.
• ▸Quality-of-hire evidence: the outcome measure you use (performance, retention, other) and the result for this channel's hires.
• ▸Mobile share: portion of this channel's candidates who arrive via mobile.
• ▸Decision: increase, hold, cut or pilot, with the data behind it.
• ▸Re-check date: when the decision will be reviewed in the next IMPACT cycle.

1. 1Start from purpose, not the requisition. The foreword says recruitment starts with the organization's purpose, goals and strategic plans, and that a target like 'increase market share by x percent' is a wish, not a purpose. Write one sentence on the role this talent plays in the organization's purpose, then name the hiring outcome you want to predict. The definition gives two: performs well and stays (loyalty).
2. 2Run the three analytics levels in order for one role. Descriptive: what has happened (who applied, who was hired, who left or underperformed). Predictive: what must be done to reach the future goal. Prescriptive: how to do it (the actual selection rule or process change). The foreword's doctor analogy fixes the sequence: symptom, diagnosis and treatment, then the written prescription. Do not skip to a scoring model without the descriptive baseline.
3. 3Check the upstream pillars before you model. The framework places Workforce Planning and Sourcing before Acquisition/Hiring. Confirm the role is a planned need and that sourcing delivers the candidate pool you intend to score. The foreword's warning is that job boards, campus trips and open houses are used by everyone and produce similar results, so a scoring model on top of the same undifferentiated pool gains little.
4. 4Map the funnel by the chapter's phases: application phase, preinterview assessment, interviews, then the final selection decision. For each phase, list the data captured and the decision made. The aim is to see where candidates are screened out and on what basis. Where the basis is gut feeling or interpersonal relationships (the preface's term for traditional talent decisions), mark it as a candidate for replacement or testing.
5. 5Treat each phase as a predictor to validate. Record preinterview assessment results and structured interview ratings for every candidate, then compare them to the later outcomes you chose in step 1: performance and retention. Keep the predictors that relate to outcomes and drop those that don't. The chapter's framing, interviews as 'separating the wheat from the chaff', makes the interview a filter that has to earn its place.
6. 6Combine the surviving predictors into one selection score. The chapter closes on 'Putting It All Together: Predictive Analytics for Selection.' Use the score to rank and shortlist candidates and keep the hiring manager in the decision. Write down the prescriptive rule (who advances, who gets a further assessment, who is declined) before the model goes live.
7. 7Close the loop through the next pillars. Hand the hire into Onboarding, Culture Fit and Engagement, then track performance and Employee Churn/Retention. Feed those outcomes back as the labels for the next model version. Without this step the loyalty and performance predictions cannot be checked.
8. 8Frame results for the executives who are asking. The preface lists their questions: right talent mix, best future leaders, who is at risk of leaving. Report the hiring model's results in those terms, as bad hires avoided and acquisition cost, which are the stated goals of the construct.

The source passages give no numeric hiring case, so this applies the foreword's doctor analogy to a hire, with no figures invented. A hiring manager says the sales team keeps losing new hires. That is the descriptive layer, like the patient reporting sinus congestion. The analyst then does the predictive work, as the doctor does when determining the treatment. They compare past hires' application data, preinterview assessment results and interview ratings with who stayed and performed, and keep the signals that relate to those outcomes. The prescriptive layer is the written instruction: advance only candidates above the chosen score, send borderline candidates to one more assessment, and route all new hires into onboarding tracked against retention. The analyst also checks the sourcing pillar. If every candidate comes from the same job-board channel everyone else uses, the foreword's point applies: similar pool, similar results. The score then helps pick among candidates but does not create an advantage on its own.

• If you cannot state the outcome (performance, retention or both) in one sentence, then stop; the model has nothing to predict, and the foreword says goals phrased as a market-share target are wishes, not purpose.
• If a funnel phase rests only on gut feeling or personal rapport, then add a structured measure to it and test it against outcomes before trusting it.
• If the candidate pool comes from channels every competitor uses, then fix sourcing first; a selection score on an undifferentiated pool yields similar results to everyone else's.
• If a predictor does not relate to later performance or retention, then drop it from the score, however standard the step is in your process.
• If you only have descriptive reporting, then do not call it prediction; move to the predictive layer, then write the prescriptive rule that says what to do about it.

• Modeling without a purpose: the team builds a score with no defined outcome. Tell: no one can say what 'good hire' means, and the model is judged by how many candidates it screens out.
• Keeping the interview as untested ritual: ratings are collected but never compared with performance or retention. Tell: interviewers disagree on the same candidates and no one has checked who proved right.
• Stopping at description: dashboards show time and sources of hire but no prediction or action rule. Tell: reports are reviewed and nothing in the process changes.
• Optimizing the hire and ignoring the downstream: the model is never fed onboarding, performance or churn outcomes. Tell: the score is not updated after launch and loyalty claims are untested.
• Scoring a commodity pool: the sourcing pillar is untouched, and every competitor draws from the same boards and events. Tell: your shortlists look like everyone else's and quality does not improve.

• ▸Role and purpose: one sentence on the role this talent plays in the organization's purpose.
• ▸Target outcomes: define 'performs well' and 'stays' (loyalty) in measurable terms you will track.
• ▸Descriptive baseline: what you already know about past hires, applicants, sources and leavers for this role.
• ▸Funnel phases and data: for application, preinterview assessment, interview and selection, list the data captured and who decides.
• ▸Gut-based steps: mark each phase that relies on gut feeling or rapport, with the replacement measure to test.
• ▸Predictors kept or dropped: list each predictor, its relation to the outcomes, and keep or drop.
• ▸Prescriptive rule: who advances, who gets a further assessment, who is declined, and who makes the final call.
• ▸Sourcing check: where candidates come from and whether the channel gives any edge over competitors.
• ▸Feedback loop: onboarding, performance and churn outcomes to be fed back, with the review date.

1. 1Define the culture you are onboarding into before you design anything. Deloitte's 3,300-leader survey found only 12 percent of organizations have a program to define and build a strong culture, so write down the specific values and behaviors a new hire is expected to adopt. Without this list there is nothing to measure fit against.
2. 2Map the onboarding process as stages, following the chapter's sequence: Organizational Culture, Onboarding Process, Stages of Onboarding, Early Employee Needs. For each stage, record who owns it (manager, HR, mentor), what the hire receives (introduction, training, mentoring, integration), and the date it happens.
3. 3List the early needs of a new hire and attach each one to a stage. Treat any need with no owner or no date as a gap, because the chapter treats early employee needs as the input to the framework.
4. 4Choose outcome measures before launch. The chapter names time to productivity 'and other outcome measures'. Pair it with engagement and retention, which match the definition's loyalty and engagement goals. Fix how each is calculated and when it is read, for example at the end of each stage.
5. 5Standardize the process so it can be analyzed. The Schmidt and Hunter and Google findings show the principle: the more you standardize (preset questions, consistent assessment), the more analytics can tell you which inputs predict success. Apply it to onboarding by giving every hire the same checkpoints, the same check-in questions, and the same recording format.
6. 6Close the measurement gap on engagement. Deloitte found 60 percent of leaders lack an adequate program to measure and improve engagement, and only 7 percent rate themselves excellent. Put a recurring engagement read inside the first months and tie each response to the hire's record.
7. 7Join onboarding data to hiring and outcome data, then build the predictive model the chapter calls for ('Create an Onboarding Predictive Model'). Test which onboarding inputs (mentor assigned, training completed on schedule, early check-in answers) relate to time to productivity and retention. Keep the inputs that predict and drop the ones that do not.
8. 8Review and refine on a fixed cadence. Compare new-hire cohorts against long-tenured employees, the comparison one company in the passages used to target talent for long-term engagement. Feed the findings back into the stage design and into sourcing and hiring.

The passages give a real benchmark and a real method, but no onboarding case with numbers, so this example combines them and uses no invented figures. Start from the benchmark: Deloitte's survey of more than 3,300 HR and business leaders in 106 countries found only 12 percent have a program to define and build a strong culture, 60 percent lack an adequate program to measure and improve engagement, and just 7 percent rate themselves excellent at measuring and improving engagement and retention. An organization in the majority writes down its cultural values and maps its onboarding stages, then sets time to productivity as the primary outcome. Next it applies the Google lesson from hiring. Laszlo Bock's team found interviews were 'a complete random mess' when each interviewer asked different questions, and structured behavioral interviews worked because everyone was assessed consistently. The organization gives every new hire the same stage checkpoints and the same engagement questions. After a few cohorts, it checks which inputs line up with faster productivity and longer retention. It also compares new hires with long-tenured employees, as the passage's company did. The result is a model of which onboarding actions matter, built on consistent data rather than anecdote.

• If you cannot state your culture in specific behaviors, then define it before designing onboarding, because fit cannot be measured against an undefined target.
• If different managers run onboarding differently, then standardize the stages and checkpoints first; inconsistent inputs make any later predictive analysis meaningless, as with Google's unstructured interviews.
• If a new-hire need has no owner and no date, then treat it as a gap and assign both before adding new programs.
• If you track only a survey score, then add outcome measures (time to productivity, retention) and test whether the onboarding inputs predict them.
• If an input shows no relationship with outcomes across cohorts, then cut or redesign it rather than keeping it for tradition.

• Onboarding as a one-day orientation: the tell is that you cannot name the stages, owners, or dates for the first months.
• Culture left undefined: the tell is that managers give conflicting answers to 'what do we value here?' and no program exists to build it, as in 88 percent of Deloitte's respondents.
• No engagement measurement: the tell is that you learn about disengaged new hires only when they resign, and the investment in training and relationships walks out the door.
• Analysts without business translation: the tell is raw data and an unreadable graph delivered after weeks, as in the company that hired a statistician and got output the business could not use. Define the questions with line managers first.
• Inconsistent process data: the tell is that every manager records onboarding differently, so cohorts cannot be compared and no model can be built.

• ▸Culture statement: list the specific values and behaviors a new hire is expected to adopt.
• ▸Stage name and date window: label each onboarding stage and its start and end dates.
• ▸Stage owner: name the manager, HR contact, or mentor accountable for the stage.
• ▸Early needs addressed: list the new-hire needs this stage must meet.
• ▸Standard checkpoint question: write the one question asked of every hire at this stage.
• ▸Primary outcome measure: state time to productivity and how and when it is calculated.
• ▸Engagement and retention read: record the engagement indicator and the retention status at each checkpoint.
• ▸Predictor hypothesis: name the onboarding input you expect to move the outcome (for example, mentor assignment).
• ▸Cohort comparison result: record what the data showed against prior cohorts or long-tenured employees, and the change you will make.