Predictive Analytics for Human Resources

1. 1Separate efficiency from effectiveness before you build anything. Time to fill, cost to hire, positions filled and salary vs. benchmark describe HR activity and cost. Competency and speed to competency describe what the hire can do. Report them as different sets so a good efficiency story cannot stand in for results.
2. 2Pull the effectiveness data from its separate systems: 90-day performance ratings, competency assessment results, speed to competency, onboarding feedback (tests and evaluations), engagement survey results, and exit survey data. Budget extra time. In the source example, the data arrived late, some in spreadsheets that needed cleaning and manual Excel work.
3. 3Segment new hires into high, middle and low performance groups using the 90-day performance rating. The source uses a 9-box: high = 6-, 8- or 9-box; middle = the 5-box ('capable'); low = 1-, 2- or 4-box. Keep the same three groups in every later exhibit so the measures can be read against each other.
4. 4Measure competency by group. Report the share of required technical competencies and the share of required business competencies each group holds, aggregated across the period (the source aggregates across all three months of the quarter). Read it as strengths and gaps, not a single score.
5. 5Measure speed to competency by group: the time until each group reaches the minimum performance threshold. Define the threshold before you measure and state it on the exhibit. Report the weeks to threshold for each group, and flag any group that has not reached it by the end of the observation window (90 days in the source).
6. 6Cross-check with onboarding test scores and learner evaluations by group. Ask who gained from training and who already arrived with the skills. High scores with little gain suggest the hire came in competent; low scores with weak feedback point to a training or selection problem.
7. 7Link competency and speed to downstream outcomes using the same groups: engagement at 90 days and one year, exit reasons, turnover at two years, and year-1 productivity. This turns a competency number into a business case.
8. 8Take conclusions to the sponsor as a short story: what the groups look like, how fast each reaches threshold, where the gaps are, and what decision follows (selection, onboarding, or retention).

In the source, a talent analytics lead first presents only efficiency metrics: 75% to 84% of positions filled within the first month, time to fill 25% faster than internal and 44% faster than external benchmarks, salary costs 23% below and total cost to hire 15% below benchmark. The VP of HR asks for results and outcomes. The follow-up report segments the quarter's new hires by 90-day 9-box rating: 60-plus percent are in the middle 'capable' 5-box, 15% to 21% are high potentials, and about 12% are low performers. Competency results: high performers hold 95% of technical and 97% of business competencies, the mid-tier is capable but weaker, and low performers hold only half of what they need. Speed to competency: high performers are fully productive at four weeks, the middle tier needs 12 weeks to reach the minimum performance threshold, and the low group is not competent at 90 days. Test data add that high performers score highest but gain least from training, which suggests they arrived with the skills. Efficiency metrics looked strong while 12% of hires were still not competent at 90 days.

• If efficiency metrics (time to fill, cost to hire) look strong but the VP asks about results, then add competency, speed-to-competency and performance data by group before presenting.
• If a group scores high on the final test but gains little from training, then treat it as a selection strength (they arrived skilled) rather than proof that onboarding works.
• If a group is still below the minimum performance threshold at the end of the observation window (low performers at 90 days in the source), then escalate it as a selection or onboarding problem rather than waiting for it to resolve.
• If the middle tier takes much longer than the high group to reach threshold (12 weeks vs. 4), then target onboarding at that group, since it is the largest segment (60-plus percent).
• If competency scores are aggregated, then break them into technical and business competencies and into groups before drawing conclusions, so gaps are visible.

• Reporting recruiting efficiency as if it were workforce quality. Tell: the deck contains time to fill and cost to hire but no competency or performance measure.
• Using a single average competency or ramp time across all hires. Tell: the average hides the 4-week, 12-week and not-competent-at-90-days groups.
• Leaving the minimum performance threshold undefined. Tell: speed to competency figures cannot be reproduced or defended when the sponsor asks what 'competent' means.
• Crediting training for results driven by incoming skill. Tell: high performers show top test scores with the smallest gain, yet the program claims the credit.
• Underestimating data collection effort. Tell: the data come from several systems, arrive late and need manual Excel cleanup, leaving no time to interpret them before the meeting.

• ▸Observation period: state the hire cohort and window (e.g., one quarter, 90-day review).
• ▸Group definitions: list the performance-rating boxes assigned to high, middle and low groups.
• ▸Minimum performance threshold: write the plain-language definition of 'competent' used for the measure.
• ▸Headcount and % of cohort by group: record the number and share of hires in each group.
• ▸Technical competency % by group: share of required technical competencies held.
• ▸Business competency % by group: share of required business competencies held.
• ▸Weeks to threshold by group: enter weeks, or 'not reached by day 90' where applicable.
• ▸Test score and training gain by group: note final test results and whether the group gained or already knew the material.
• ▸Decision and owner: record the action (selection, onboarding or retention) and who is accountable.

1. 1Anchor the definition in the four needs. Branham's finding is that employees begin to disengage when one of four needs goes unmet: trust, hope, sense of worth, or feeling competent. Treat these as the diagnostic frame, not a satisfaction score.
2. 2Classify engagement as an intangible and a leading indicator. The source says changes in engagement, leadership capability, readiness, and culture are valuable leading indicators. Report it as a predictor of outcomes, not as an HR activity count. Hires, pay actions, and trainings are 'a record of expenditures, not value added.'
3. 3Pick one job group (or a few) rather than the whole company. Organizations are complex and the links differ by group. A bounded group lets you test one pattern at a time.
4. 4Pull the reasons employees gave for leaving, but only if the exit interview process is valid. The source conditions the analysis on this. Check how the interviews are run before you trust the reasons.
5. 5Map each reason to Branham's seven separable, preventable reasons (Exhibit 1.5). Tag each one to the need it violates: trust, hope, worth, or competence. Then look at which needs dominate in which group.
6. 6Run statistical analysis on combinations: reason by tenure, and reason by position. Do not stop at a frequency count of reasons. The value lies in the combinations.
7. 7Test the manager-tenure link. The source reports a case where a manager stays too long in one job and this correlates with high disengagement and with lower quality, productivity, or service in the unit. Check whether the same pattern shows up in your data.
8. 8Track engagement over time against downstream outcomes. The book's exhibits pair engagement by group at 90 days and one year with exit survey results, turnover at two years, and year-1 productivity. Build the same chain: engagement, then turnover, then productivity. Use it to move from describing (what is happening) to predicting (why, and where it leads) to prescribing (what to do).
9. 9Replace speculation with objective analysis. The source's stated purpose is to use analysis 'in lieu of biased, out-of-date speculation' about why people disengage.

The source describes this scenario as an assumption, not a measured result. You select a job group and pull the reasons its employees gave for leaving, assuming the exit interview process is valid. You then apply statistical analysis to find combinations of reason and tenure, or reason and position. Suppose the analysis reveals a connection between tenure and position. The authors report finding, in some cases, that a manager can stay too long in one job. That correlates with high employee disengagement and with other operating problems such as reduced quality, productivity, or service in the unit. Read through Branham's frame, the next step is to ask which of the four needs (trust, hope, worth, competence) the long-tenured manager's team is missing. This is the descriptive-to-predictive move: you now know what is happening and have a testable explanation for why. The prescriptive step is to decide what to do about it, such as managing manager tenure in that position. The source gives no figures for this case, so any numbers must come from your own data.

• If exit interview data is the main input, then first confirm the interview process is valid. The source makes the analysis conditional on it.
• If a stated exit reason shows up, then cross it with tenure and with position before acting. The source's insight came from the combination, not the single reason.
• If a unit shows disengagement plus falling quality, productivity, or service, then check manager tenure in the role first. The source links a manager staying too long to this pattern.
• If you are reporting an engagement score, then report its predictive link to turnover or productivity, not just the number. The source says 'the important point is not the number but the predictability.'
• If leadership resists the finding, then separate the person from the tool. The source warns that when someone does not want to solve a problem, the tool is not at fault.

• Treating engagement as a satisfaction survey average. The tell: you can state the score but not which of trust, hope, worth, or competence is unmet.
• Reporting HR volumes (people hired, trained, surveyed) as proof of engagement work. The tell: the report shows expenditures and activity but no change in a leading indicator or a business outcome.
• Trusting exit reasons from a weak interview process. The tell: reasons are vague or uniform across groups, and nobody has checked how the interviews were conducted.
• Analyzing single causes in isolation, such as reason alone. The tell: no tenure or position cross-cuts, so patterns like a manager who stayed too long never surface.
• Running a quick fix on obvious blemishes and stopping. The source compares this to giving athletes painkillers while ignoring the wrenched knee. The tell: engagement numbers improve briefly while turnover and productivity problems persist.

• ▸Job group: name the single group or position set you are analyzing.
• ▸Exit interview validity check: note how interviews are conducted and whether you trust the reasons given.
• ▸Top exit reasons: list reasons given, mapped to Branham's seven preventable reasons (Exhibit 1.5).
• ▸Unmet need: for each reason, mark trust, hope, sense of worth, or competence.
• ▸Reason x tenure finding: record what the statistical cross shows.
• ▸Reason x position finding: record what the statistical cross shows.
• ▸Manager tenure in role: note whether a manager has stayed too long and any matching quality, productivity, or service decline.
• ▸Engagement trend: record engagement by group at 90 days and one year, paired with turnover at two years and year-1 productivity.
• ▸Prediction and action: state what the data suggests will happen next and the prescriptive action you will test.

1. 1Fix the rating event before collecting anything. Each employee gets a manager (and sponsor) judgment of performance and potential on a nine-box at two points: 90 days and 365 days. Write down who rates, on what date, and on what scale. Quarterly performance ratings (the source's Exhibit 4.6) sit alongside these as a separate series, so do not merge them.
2. 2Collapse the nine boxes into the three groups the analysis uses: high, capable, low. Write the box-to-group mapping down once and lock it before you look at any outcome data. A mapping chosen after you see results is tuned to the answer.
3. 3Store the ratings in a delimited file with one row per employee per rating event. The source's data-tracking and vertical data-set layouts follow this pattern. Include employee ID, rater, rating date (90 or 365), performance box, potential box, and group. Pair this with a metric-definitions sheet so 'high performer' means the same thing in every report.
4. 4Run the two data-quality checks the source poses: what form is the data in, and is it sufficient? Look for missing 90-day or 365-day ratings, raters who give everyone the same box, and employees with a rating but no outcome record. Resolve these before any comparison.
5. 5Join the group label to outcome metrics by employee ID. The source tracks outcomes by performance group: test-score percentage passing by group and month, engagement at 90 days and one year, turnover at two years, and productivity during year 1. Build one table per outcome, with the three groups as columns.
6. 6Compare groups first, then test prediction. Descriptive: do high, capable and low groups differ on each outcome? Predictive: use correlation and regression, the source's own sequence, to test whether the 90-day rating predicts the 365-day rating and the later outcomes. Report relationships as probabilities, not certainties.
7. 7Place the rating inside a logic model with key performance metrics, so it is one measured step between inputs (hiring, training) and outcomes (productivity, retention). The source frames analytics as measuring the efficiency, effectiveness and outcomes of processes, not the worth of people. Report the rating that way.
8. 8Report the group differences and the 90-to-365-day movement to the executive audience, and name the decision each finding supports (hiring profile, development, retention focus). If the finding changes no decision, drop it from the report.

The source supplies the structure of this analysis but not its numbers, so the example below is a design, with no results claimed. Define three groups (high, capable, low) from nine-box ratings at 90 days and 365 days, then follow the exhibit sequence the source uses for a new-hire population. First, quarterly performance ratings by group. Second, test-score percentage passing by group and month, to see whether high performers separate early. Third, engagement by group at 90 days and one year. Fourth, turnover by group at two years. Fifth, productivity by group during year 1. Read the five tables together. If low-rated hires disengage by one year and leave by year two, the 90-day rating is a usable early signal. If low-rated hires stay as long as high-rated ones, the rating may be noise, or the exit pattern may reflect something else, such as the preventable reasons for disengagement the source lists. Test the 90-day rating as a predictor with correlation and regression, then take the winning metrics into an optimization model. The rating earns its place only if it moves a downstream metric.

• If 90-day and 365-day ratings are missing for a large share of the population, then fix collection before analyzing. The source treats data sufficiency as a gate, not a footnote.
• If a group's outcome differs from the others only at one time point (say, 90-day engagement but not one-year), then report it as a timing effect, not a group trait. Do not generalize it.
• If the 90-day rating correlates with 365-day outcomes, then use it as an early-warning signal. If it does not, then treat it as a descriptive label only and do not build predictive claims on it.
• If a stakeholder resists the findings by recounting past errors, then separate the tool from the person. The source's Fortune 100 CHRO example shows resistance is a people problem, not a flaw in the analysis.
• If a result describes the past, then do not extrapolate it to the future without adding what you know about changing conditions. Descriptive trends alone do not predict.

• Defining groups after seeing outcomes. The tell: the high, capable and low cutoffs shift between report versions.
• Treating the rating as the value of the person rather than a measure of a process. The tell: the output ranks individuals instead of comparing group outcomes against efficiency, effectiveness and outcome metrics.
• Reporting without a data-quality check. The tell: groups differ in size from one table to the next because employees drop out of some tables.
• Jumping to prediction before description. The tell: a regression exists but nobody has first compared the three groups on plain averages by month.
• Analyzing one rating point in isolation. The tell: only the 365-day box is used, so you cannot say whether performance is rising, flat or falling since 90 days.

• ▸Population and hire window: which employees and which dates are in scope.
• ▸Raters: who gives the manager and sponsor ratings, and on what dates (90 and 365 days).
• ▸Box-to-group mapping: which nine-box cells count as high, capable, low. Lock before analysis.
• ▸Data file layout: employee ID, rater, rating date, performance box, potential box, group.
• ▸Data-quality check: percent missing at each rating point, and raters who use only one box.
• ▸Outcome metrics: test pass rate by month, engagement at 90 days and one year, two-year turnover, year 1 productivity.
• ▸Group comparison: results for each outcome across high, capable and low.
• ▸Predictive test: does the 90-day rating predict 365-day rating and later outcomes (correlation or regression result).
• ▸Decision supported: the specific hiring, development or retention action the finding informs, and the owner.

1. 1Pick one job group before you touch any statistics. The source suggests a technical or professional group, long-tenure staff, or operators (assemblers, warehouse pickers, truck drivers) where the job market is tight. State why this group matters to the business.
2. 2Pull the raw fields already in the HR database: date of hire, performance reviews, status changes (promotions, salary increases, jobs held), and date of departure. Turnover is an easy first analysis because this data usually exists already.
3. 3Split leavers by performance review history so high-performer departures are visible as their own line. The source ties turnover to performance and applicant source (its Chapter 3 example), and does not give a cost figure. Treat the high-performer split as your own cost-weighting, not a sourced number.
4. 4Sort departures four ways: by reason, tenure, position, and supervisor. Look for combinations, such as reason and tenure or reason and position, rather than reading each cut alone.
5. 5Check the exit interview process before trusting the reason field. The source's approach assumes you have a valid exit interview process. If you don't, label reasons as unverified.
6. 6Classify stated reasons against Branham's frame: 67 reasons for disengaging, 57 preventable. Personal decisions such as going back to school or the family moving are the unpreventable ones. Preventable reasons trace to four unmet needs: trust, hope, sense of worth, feeling competent. Report the preventable share separately.
7. 7Test for pattern links, such as tenure and position. The source found cases where a manager stays too long in one job, which can correlate with high disengagement and with drops in quality, productivity, or service in the unit. Follow the leaver pattern into those operating measures.
8. 8Connect the retention change to a business outcome (quality, productivity, service). The source says very few attempts do this. Then move from describing what happened, to predicting where it leads, to prescribing the action, and replace speculation with objective analysis.

The source's own scenario runs as follows. You select one job group and sort its departures by reason, tenure, position, and supervisor. Applying statistics to reason-and-tenure and reason-and-position combinations, you find a connection between tenure and position: in some units, the manager has stayed too long in the same job. Those units show high employee disengagement and also reduced quality, productivity, or service. The retention problem is therefore concentrated in specific supervisory situations, not spread across the whole group. You classify the stated exit reasons against Branham's preventable list (unmet trust, hope, sense of worth, or competence) and find which of the four needs the long-tenured managers' units fail to meet. Your prescriptive step is then aimed at those managers' positions, and you track the same operating measures afterward to see whether the change worked. This ties turnover to a business outcome, the connection the source says is rarely made. The source gives no numbers for this scenario, so any thresholds are yours to set.

• If exit interview data is not from a valid, consistent process, then treat reason codes as hypotheses and do not build conclusions on them.
• If departures cluster by supervisor or by tenure-in-position rather than by job group, then investigate the management situation first, not pay or the market.
• If a reason is a personal decision (returning to school, family move), then classify it unpreventable and exclude it from the retention action list.
• If a preventable reason maps to trust, hope, sense of worth, or competence, then target the practice behind that need, not a general engagement program.
• If you can report only a turnover rate, then you are at the descriptive level. Add a link to quality, productivity, or service before proposing action.

• Reporting turnover rate as the whole answer. The tell: the report lists rates by department and never says what the departures did to business results.
• Trusting exit-interview reasons without validating them. The tell: the top reason is vague or socially safe, and no one has checked how interviews are conducted.
• Treating all leavers alike. The tell: high performers and low performers sit in the same count, so the costly losses are invisible.
• Studying one cut at a time. The tell: reason, tenure, position, and supervisor appear on separate slides, so a combination such as a long-tenured manager is never found.
• Extrapolating last year's trend. The tell: the forecast assumes a stable labor market, though the source notes recruitment programs can be made ineffective overnight by market dynamics, competitors, or technology.

• ▸Job group studied: name the group and why it matters to the business.
• ▸Data pulled: confirm hire date, performance reviews, status changes, and departure date are all present.
• ▸Exit interview validity: note how reasons are collected and whether you trust them.
• ▸High-performer flag: define how leavers are tagged using review history, and count them separately.
• ▸Cuts run: tick reason, tenure, position, supervisor, and list the combinations tested.
• ▸Preventable vs. unpreventable split: count leavers in each, using Branham's frame.
• ▸Unmet need (trust, hope, worth, competence): record which need the preventable reasons point to.
• ▸Linked business measure: name the quality, productivity, or service measure you compared against leavers.
• ▸Prescribed action and follow-up: state the change, the owner, and when you re-check the linked measure.

1. 1Define productivity as output per unit of input, using the source's process terms: what the process costs, how long it took, how much output came from a given input, the quality of the output, and how people feel about it. Write one definition per job group before collecting anything. The source says to measure the efficiency, effectiveness and outcomes of processes, not the 'value of people'.
2. 2Pick one output measure and one time or cost base for each job group, such as billable work against hours worked. Record it in a metric-definitions sheet (the book's Exhibit 6.6 covers definitions of metrics). Two groups measured differently cannot be compared.
3. 3Start with descriptive analytics. Produce productivity by group over the first year, as in Exhibit 4.14 (Productivity among Groups during Year 1). Segment by the cuts the source uses for pattern-finding: reason, tenure, position and supervisor. Note which groups sit high or low before explaining anything.
4. 4Link productivity to the talent practices that preceded it. Compare groups by hiring source, testing, assessment, onboarding and training exposure. The source's eyeball method works for small populations. Larger populations need statistical tests, because the amount of data is too much to read by eye.
5. 5Build a data set and rank-order the correlations with productivity, as in Exhibit 6.17. Then fit a regression of productivity on the strongest candidates (Exhibit 6.20, Predictors of Productivity). Treat the output as probabilities and potential impact, not certainties.
6. 6Apply the model forward. Score new hires or candidates for predicted productivity (Exhibit 6.21 does this for five new hires) so selection, training and promotion decisions change before the output is lost.
7. 7Move to prescriptive use. Compare alternative investments, for example which learning investment moves productivity most, and select the option with the best business impact. The source notes this level is rare in HR and that value grows as you move from descriptive to prescriptive.
8. 8Connect productivity to money. Place it beside revenue, profit and total cost of workforce (TCOW), as in the Human Capital Income Statement (HCI$), so leaders see the financial effect of talent programs. Expect the objections 'we could never collect that data' and 'we wouldn't know what to do with it'. Start with the data already in the HR database.

The source's scale argument shows how to size the prize. U.S. corporations spend well over $6 trillion a year on labor. KnowledgeAdvisors reports that organizations that effectively deploy automated talent analytics with predictive analytics see at least a 4 percent productivity improvement. Four percent of that labor spend is approximately $250 billion in bottom-line improvement per year, the source's figure. The same logic works on waste. The U.S. federal government spends about $30 billion a year on training, and research cited in the source says at least half of training spend is wasted, so about $15 billion a year is lost. To size your own case, replace those numbers with your labor spend and your measured productivity gap. The source also names where productivity leaks. Training that trainees do not see as needed is 'almost totally wasted', and so is the productivity lost when people leave their jobs to attend the class. A manager who stays too long in one job can correlate with disengagement and reduced quality, productivity or service in the unit. Those are the first places to look in your own group data.

• If a group's productivity is low, then segment it by reason, tenure, position and supervisor before prescribing anything. The cause may sit with a manager who has stayed too long in one job, not with the employees.
• If the population is small, then compare individual records by eye, as in the Didi, Ken, Leo and Earl review of treatment differences. If it is large, then use correlation and regression, because eyeballing fails.
• If a training investment is proposed to lift productivity, then first check whether trainees feel they need to know the material. Without that, the source expects the spend and the lost work time to be wasted.
• If you need to justify an analytics investment to finance, then express the gain as a percentage productivity improvement on labor spend and tie it to revenue, profit and TCOW, as the HCI$ does.
• If a metric only describes what happened, then do not extrapolate it into a volatile future. Build a predictive model before committing budget on the trend.

• Measuring headcount or 'value of people' instead of process output per input. The tell: reports count people and never state cost, time, output and quality per unit.
• Comparing groups with different productivity definitions. The tell: every group looks like an outlier and nobody can state the numerator and denominator.
• Treating a correlation as the cause and prescribing a blanket fix. The tell: the same development program goes to everyone regardless of assessed need, the 'development blanket' the source criticizes.
• Stopping at descriptive dashboards. The tell: productivity is reported every quarter but no model scores new hires or compares investment options.
• Abandoning the work because the data seems out of reach. The tell: 'We could never collect that data' or 'we wouldn't know what to do with it' ends the project before the HR database has been checked.

• ▸Job group: name the group being analyzed (for example, a technical group or operators).
• ▸Output measure: state the billable or output-generating work counted.
• ▸Input base: state the time or cost denominator (hours worked or TCOW).
• ▸Productivity value, Year 1: record output per input for the group and the comparison group.
• ▸Segments checked: list which of reason, tenure, position and supervisor you cut by, and what you found.
• ▸Talent treatments compared: record hiring source, testing, assessment, onboarding and training exposure for each group.
• ▸Top correlates of productivity: list the rank-ordered variables and their correlation values.
• ▸Predictive model result: record the regression predictors and the predicted productivity of the next hires.
• ▸Prescribed action and financial link: state the chosen investment, its expected effect on productivity, and the revenue, profit or TCOW line it should move.

1. 1Write the definition as a formula before touching data. The construct is the bottom-line value human capital generates, computed from productivity and salary. Name the productivity measure (output per person or group) and the salary measure (the pay data for the same people), and fix the unit of analysis (individual, group, or hire cohort). The passages supply the definition, not a worked formula, so state your own formula and keep it constant across every run.
2. 2Start with the purpose question the source poses: 'Profitability for what purpose?' Tie the construct to the company's go-to-market model and its financial goals (revenue, market share, profitability, EBITDA). A leadership or hiring effort that sells luxury service and one that sells on price will not produce profit the same way, so the profitability you chase must match how your company makes money.
3. 3Place profitability on the analytic value chain: workforce drives operations, operations produce what customers buy, customers' spending produces economic and financial value. Profitability is the end of that chain. List the upstream efficiency measures (e.g., days to fill, cost per hire) and effectiveness measures (e.g., performance ratings, speed to competency, engagement, turnover) that plausibly feed it. Skipping the chain is how you get 'flipping' results: short-term profit with unsolved underlying problems.
4. 4Settle the data questions before modeling: who owns the productivity and salary data, will they share it, what form it is in (delimited file, vertical or cross-tab layout), and whether quality is sufficient. Log each source in a data tracking tool and send a data request using a template. Define each metric in writing so productivity and salary mean the same thing across business units.
5. 5Build a single data set with one row per person or hire. Include productivity, salary, the computed profitability, and the candidate predictors from your efficiency and effectiveness measures (e.g., ratings, competency results, engagement at 90 days and one year, test scores).
6. 6Run correlations among all variables, then rank-order the correlations for productivity and for profitability separately. Compare the two lists: a factor that moves productivity may not move profitability once salary is counted.
7. 7Move from correlation to regression to isolate the predictors of profitability and the predictors of productivity. Use structural equation modeling only if you need to test the paths among variables. Interpret the output, then check it against the causal story from the value chain.
8. 8Apply the model to new hires: predict productivity for the group, then predicted profitability for a smaller set. Report the result as a decision (who to hire, what to develop, what to change) alongside the descriptive baseline. Update the model as the market, labor costs, and skill needs shift.

The source's profitability sequence runs through its exhibits, though the passages do not reproduce the figures, so none are invented here. The work begins with a data set for predictive analytics. Correlations among variables are computed (Exhibit 6.16), then rank-ordered once for productivity (6.17) and once for profitability (6.18). A regression isolates the predictor of profitability (6.19) and the predictors of productivity (6.20). The model is then applied forward: predicted productivity scores for five new hires (6.21) and predicted profitability for three new hires (6.22). The lesson of the sequence is that productivity and profitability get separate rankings and separate predictor lists, because profitability nets out salary. A parallel framing comes from the source's attrition case: ask which types of people leave and when, and for what purpose you are reducing attrition. Retention then matters to profitability only where the departing people were generating value above their cost.

• If a predictor ranks high for productivity but low for profitability, then treat it as a cost-sensitive lever and do not recommend it on productivity evidence alone.
• If you cannot name who owns the salary or productivity data, or the two sources use different definitions, then stop and fix the data before modeling.
• If someone asks you to 'improve profitability' without a stated purpose tied to the go-to-market model, then ask 'profitability for what purpose?' before choosing measures.
• If the model shows correlation but you cannot tell a causal story along the value chain, then report it as an association and keep testing before the business acts on it.
• If the only evidence is short-term profit gained by cutting people or selling assets, then do not read it as sustained human-capital profitability; check whether the underlying weaknesses were solved.

• Measuring productivity and calling it profitability. The tell: salary appears nowhere in the calculation, so high-output, high-cost people look as good as high-output, low-cost people.
• Jumping into the value chain at the midpoint, as flippers do, and optimizing a quick-fix metric. The tell: profit rises now, but the underlying strategy, management, or operating system is unchanged and the gain does not hold.
• Installing off-the-shelf models without linking them to the go-to-market model. The tell: the same leadership or hiring profile is applied to very different businesses, such as Neiman Marcus and Walmart.
• Modeling on unchecked data from sources that don't share definitions or quality standards. The tell: units or periods disagree across business units, and correlations shift when one source is dropped.
• Reporting predictions without an action. The tell: the executive report shows predicted profitability for new hires but no decision, owner, or next best move attached.

• ▸Purpose statement: complete 'We need profitability analysis in order to…', tied to the go-to-market model and a named financial goal.
• ▸Profitability formula: write how productivity and salary combine into profitability, and state the unit of analysis.
• ▸Productivity metric and source: name the measure, its definition, and the system or owner holding it.
• ▸Salary metric and source: name the measure, its definition, and the owner; confirm the owner will share it.
• ▸Candidate predictors: list efficiency and effectiveness measures that plausibly feed profitability, with the value-chain link for each.
• ▸Data readiness check: record the form of the data, quality issues found, and the fix for each.
• ▸Analysis steps: mark when done: correlations, rank order for productivity, rank order for profitability, regression, structural model if needed.
• ▸Key predictors found: separate lists for productivity and for profitability, noting any that differ.
• ▸Prediction and decision: state who or what is predicted (e.g., a hire cohort), the predicted result, and the action, owner, and date.

1. 1State the issue as one sentence, then ask the opening question the source asks: 'For what purpose?' Attrition is the passage's example. Reducing unwanted attrition sounds obviously good, but the purpose has to be named before any analysis starts.
2. 2Ask questions in a logical chain from current state to preferred destination. Each question follows from the previous answer, as Socrates did with Thrasymachus: no confrontation, just relevant questions answered one after another until the group reaches the real position.
3. 3Break the issue into its component facts before explaining it. For attrition, the source lists: What is the total attrition rate? Which types of people are leaving? When in their career are they leaving? Why are they leaving? Answer at least some before going further.
4. 4Ask the follow-on questions: once you have the answers, what else do you need to know? What are the effects of attrition? How can you influence it, and for what purpose? Then ask for the business value of knowing.
5. 5Screen every answer for obsolete experience, irrelevancy and bias. Push unimportant issues aside and challenge beliefs that rest on old experience. Verify that each remaining connection is testable.
6. 6Check the three capital categories (human, structural, relational) and ask how a change in one affects the others. Then list the principal internal and external drivers (force analysis), including people, processes, resources, customer demands, regulations, even weather and geography.
7. 7Decide whether analytics is needed at all. If the problem is simple, a supervisor or manager can solve it within the unit. If the questions alone produce the solution, stop. Move to statistical modeling only when multiple causes remain.
8. 8Write the project goal in terms of operations, cost or revenue. A 2% drop in attrition or a 4% rise in engagement means nothing until applied to one of those. Then, and only then, start the statistical work.

The source's own case is attrition. A team starts with Branham's seven reasons for employee disengagement and tries to answer at least some. They ask: What is the total attrition rate? Which types of people are leaving? When in their careers? Why? Then the questioning moves past description: what are the effects of attrition, how can the organization influence it, and for what purpose? The answer is not just to cut the percentage loss. The team asks what the business value of knowing is, because knowing a ratio changed adds nothing until it is tied to improving operations, reducing costs or increasing revenue. A parallel case is leadership development. After an assessment vendor shows current leader capability, the next question is Drucker's 'Leadership for what purpose?' That forces a study of what the future market will demand before building a model to test. In both cases the questions define the problem and the target before any statistics are run.

• If the problem is simple and has one cause, then let the unit's supervisor or manager solve it; bring in analytics only when multiple causes exist.
• If the questions have already produced the answer, then skip the statistical exercise.
• If a metric movement (attrition down 2%, engagement up 4%) cannot be tied to operations, cost or revenue, then keep asking 'for what purpose?' before modeling.
• If a stakeholder's belief rests on past experience, an irrelevancy or a bias, then ask for a verifiable, testable connection or set it aside.
• If a change in one capital category (human, structural, relational) is under discussion, then ask what it does to the other two before settling the problem statement.

• Running the statistics first, like driving off without a travel plan. The tell: no one can state the destination or the purpose in one sentence.
• Treating a metric change as the goal. The tell: the project target is 'reduce attrition' or 'raise engagement' with no link to operations, cost or revenue.
• Assessing capability without asking what it is for. The tell: a leadership assessment is done but no one has defined the future leadership model or what the market will demand.
• Letting obsolete experience, irrelevancies or bias stay in the frame. The tell: the group argues over unimportant issues and cannot agree on the essential conditions.
• Confronting rather than questioning. The tell: a stakeholder digs in on a stated position instead of reasoning through logical, relevant questions to a shared view.

• ▸Issue statement: one sentence describing the problem as currently stated.
• ▸Purpose: answer 'for what purpose?' and name the operations, cost or revenue outcome it serves.
• ▸Descriptive facts: the answers to how much, who, when and why (e.g., attrition rate, types leaving, career timing, reasons).
• ▸Follow-on questions: effects, how it can be influenced, and the business value of knowing.
• ▸Capital interactions: note human, structural and relational capital effects.
• ▸Internal and external drivers: list people, processes, resources, customer demands, regulations and other forces.
• ▸Beliefs and biases screened: record each belief rejected as obsolete, irrelevant or biased, and why.
• ▸Testable connections: list the verifiable relationships left standing.
• ▸Analytics decision: record whether the problem is simple, solved by questions alone, or multicausal and needing modeling.

1. 1Start from the end, not the data. Before touching any system, write the business question the analysis must answer. The source's Chapter 5 opens with 'Begin with the End in Mind' and 'Go Back to the Beginning', and Chapter 5 says analytics starts with logical questions and statistics come later. The question decides which data you need.
2. 2Inventory where each needed data element lives. Chapter 5 frames this as 'where is it'. List every source for the question, including HR systems and the finance, marketing, and other functional data an operating question will pull in. Record system, format, and what it holds.
3. 3Name an owner for each source and test willingness to share. The chapter asks 'Who owns data, and will they share it?' as a separate question from where it is. Get the owner's agreement before you design the analysis around their data.
4. 4State what you will do with the data. For each element, write the specific use (efficiency, effectiveness, or business outcome measure). Drop elements with no use rather than cleaning them.
5. 5Document the form of each element. Note spreadsheet, database, or unstructured. The source says roughly 80% of the world's data is unstructured. Note which fields need conversion before they can be joined.
6. 6Set a standard definition for every metric. The source blames the analytics gap on a missing 'history of data collection, standard definitions and ongoing monitoring' (it estimates probably 75% of HR departments lack useful basic metrics). Use the Center for Talent Reporting's Standard Metrics Definitions, which the book's appendix lists, as the starting reference.
7. 7Test quality on the data you will use. The chapter asks whether the data is valid, reliable, and sufficient. The index flags the specific problems: keypunching errors, misaligned data, and missing data. Check each element for all three before analysis.
8. 8Build the architecture to receive the numbers, then start collecting and monitoring on a schedule. The source's answer to missing history is 'learning as you go': start with a basic infrastructure and refine as you learn.

The source describes an analytics unit built from scratch. Once the unit knew what data it needed, 'the HR database architecture had to be established.' Because operating issues were going to be addressed, the unit needed data from finance, marketing, and other functions, so systems analysts and coders were put to work on a 'flexible, mutually beneficial architecture.' New technology tools were needed, and the director built the money for them into an operating budget that top management approved. Reports were also redesigned: their rationale, format, and content were renegotiated, and a testable set was cleared with their readers. Only after this infrastructure was in place did the unit move into full production and agree with the C level on what reports would appear and when. Trial and error then settled the data collection methodology. Two years after inception the unit was fully functioning. The lesson is that data access, architecture, funding, and agreed reports came before analysis, and they took negotiation, not just technical work.

• If you cannot name the business question, then do not start collecting or cleaning data. The question determines which data matters.
• If the data lives in finance, marketing, or another function, then secure the owner's agreement to share before you commit to an analysis that depends on it.
• If a metric has no written standard definition, then adopt one (for example from the Center for Talent Reporting's Standard Metrics Definitions) before you trend it.
• If you find keypunching errors, misaligned records, or missing values in a field, then fix or exclude them before modeling. Do not model on a field you have not tested for validity and reliability.
• If you have little data history, then start with a quick-win problem and build the infrastructure as you learn. Do not wait for a perfect data foundation.

• Jumping to statistics before defining the question. The tell: you have many metrics and no decision they inform.
• Assuming HR data is analysis-ready. The tell: no standard definitions, no collection history, and no ongoing monitoring (the condition the source says probably 75% of HR departments are in).
• Designing the analysis around data you do not own. The tell: the owner has not agreed to share, or the extract keeps being delayed.
• Skipping quality checks. The tell: keypunching errors, misaligned data, or missing data surface after results are already presented.
• Ignoring data form. The tell: you cannot join records because key information sits in unstructured sources or incompatible formats.

• ▸Business question: write the one decision this analysis must support.
• ▸Data elements needed: list each field required to answer it.
• ▸Location: system or function where each element lives (HR, finance, marketing, other).
• ▸Owner: named person or department responsible for each source.
• ▸Willingness to share: yes, no, or pending, with the date confirmed.
• ▸Intended use: efficiency, effectiveness, or business outcome measure.
• ▸Form: spreadsheet, database, or unstructured, plus any conversion needed.
• ▸Definition: the standard definition used (cite source, e.g., Center for Talent Reporting).
• ▸Quality test result: findings on keypunching errors, misaligned data, and missing data; verdict on validity, reliability, and sufficiency.

1. 1Start from the executive's own question, not from your method. The source frames strategy as top executives asking 'How do we make money?' (for not-for-profits, 'How do we service our constituents?'). Write one sentence connecting your analytics project to that question before any sponsor meeting.
2. 2Choose the value path that matches the sponsorship you actually have. The source names three: solve a problem, form an analytics unit, develop an analytics culture. The culture path needs the full support of C-level team members who practice analytics among themselves and insist subordinates do likewise. Without that, the source advises starting with the first two paths.
3. 3Sell a quick hit first. A low-hanging-fruit problem adds value, builds credibility, and gives a first step toward larger value (for example, identifying high-performance variables that set up a training curriculum). Frame the request to the sponsor as this one bounded problem, not 'fund analytics'.
4. 4Put the cost of inaction in dollars and in competitive terms. The source's model pitch is a 4 percent productivity gain on a labor spend above $6 trillion (about $250 billion a year) and the claim that at least half of training spend is wasted. It also asks pointed questions: 'What chief executive wants to continue making more bad hires than competitors?' Build your own version from your company's labor spend, hiring and training numbers.
5. 5Recruit a named sponsor or champion. Chapter 3 of the source treats 'Recruiting a Sponsor or Champion' and 'Making the Sale' as separate tasks from 'Dealing with the C Level'. Ask the sponsor for three specific things: resources, data access, and visible backing when you present findings.
6. 6Settle data access during the sale, not after it. The source asks 'Who owns data, and will they share it?' as a core question. Get the sponsor to commit to opening the data owners before the project starts.
7. 7Be honest about your data starting point. The source estimates about 75% of HR departments lack useful basic metrics, and its answer is 'learning as you go'. Present the project as building an infrastructure for the numbers, so the sponsor does not expect predictions you cannot yet produce.
8. 8Sell the change, not the analysis. The source says the best analytic outputs are useless if the organization cannot change to use them, and that salesmanship and change management are imperative skills for analytic units. Before delivering results, agree with the sponsor on who will act on them and what decision they will change.

The source's own sales argument shows the structure. KnowledgeAdvisors states that organizations deploying automated talent analytics with embedded predictive analytics see at least a 4 percent productivity improvement. It applies that to a labor spend of more than $6 trillion, which gives roughly $250 billion a year (the straight math is $240 billion, so state it as 'approximately'). It then notes that most companies trade well above four times earnings, so the shareholder value is on the order of a trillion dollars. For the public sector, it cites the claim that half of training spend is wasted, and applies it to about $30 billion of federal training spend to reach about $15 billion wasted. It closes with questions a CEO would struggle to answer, such as why shareholders should not expect a quality control system for onboarding or the corporate university. To use this internally, replace the macro figures with your own labor spend, training budget, and bad-hire cost. Keep the same order: size of waste, a competitor comparison, a pointed question, then a small first ask.

• If you lack C-level executives who use analytics themselves, then do not propose a culture change. Pursue a solve-a-problem quick hit or a small analytics unit.
• If the sponsor asks 'where's the beef?', then lead with the dollar outcome and the decision it changes. Do not lead with statistical method. The source notes that non-statisticians feel overwhelmed by analytics talk.
• If the data sits with owners who have not agreed to share it, then pause the analysis and escalate to your sponsor. Do not build around the gap.
• If your basic metrics and standard definitions do not exist, then sell a staged build ('learning as you go') with the infrastructure first. Do not promise prediction.
• If the findings require someone to change how they work, then name that person and the new action before the readout. Output alone does not create change.

• Selling the technology or the statistics. The tell: the sponsor's reaction is polite interest and no commitment of budget or data access.
• Aiming at culture change without top-level practice. The tell: executives endorse analytics in meetings but still decide by habit and do not ask their subordinates for data.
• Over-scoping a first project. The source's Fortune 20 client spent almost two years on a database never designed for analytics and had to go back and renegotiate a limited Version 1.0. The tell: a long timeline with no early result to show.
• Using inflated macro numbers without grounding them in your own company. The tell: the sponsor asks 'what does this mean for us?' and you have no figure.
• Treating sponsorship as a one-time approval. The tell: the sponsor signed off at kickoff but is absent when data owners resist or when results require organizational change.

• ▸Executive's core question: restate 'How do we make money?' (or serve constituents) in the sponsor's own words.
• ▸Value path: solve a problem, form an analytics unit, or develop a culture. Note whether the sponsor and peers practice analytics themselves.
• ▸Quick-hit problem: one bounded problem you can deliver on soon.
• ▸Cost of inaction: your company's labor spend, training spend, or bad-hire cost, with the estimated waste in dollars.
• ▸Competitive question: one pointed question for the executive, modeled on the 'bad hires versus competitors' question.
• ▸Named sponsor and ask: who the champion is, and the specific resources, data access, and visible backing requested.
• ▸Data owners and sharing status: who owns each data source and whether they have agreed to share.
• ▸Metrics readiness: which basic metrics and definitions exist and which must be built first.
• ▸Change owner: who will act on the findings and which decision or behavior changes.

1. 1Start with the question, not the statistic. The source says analytics starts with logical questions and statistics come later. Write one sentence naming the business problem the investment must solve (for example, new hires reach competency too slowly, or leaders are ineffective). Name who owns that problem.
2. 2Anchor the problem in how the company makes money. The analytic value chain starts with the executives' question, 'How do we make money?' Link each recruiting, onboarding, training, coaching, or development line to a strategic goal in that answer. If you can't link it, it is not a priority investment.
3. 3Assess the gap before you spend. Define the skill or capability the business needs, and measure where the workforce stands against it. The source treats assessment as the first step in its go-to-market models and in leadership development. Record the gap in a form you can re-measure: competency results, quarterly performance ratings, test scores.
4. 4Build a logic model for each program. Chain the resources invested to the activity, then to the learning or behavior change, then to the business result. The book's exhibits show a blank and a completed logic model. Each link needs a named measure.
5. 5Measure at three levels using the TDRP structure of efficiency, effectiveness, and outcomes. Efficiency covers open positions and positions filled per month, time to fill against benchmarks, and salary cost and total cost to hire. Effectiveness covers competency results, speed to competency, and the percentage passing tests by performance group. Outcomes cover engagement at 90 days and one year, exit survey results, turnover at two years, and first-year productivity by group.
6. 6Check the data before modeling. Find out where it lives, who owns it, whether they will share it, what form it is in, and whether quality is sufficient. Use a data request template and a metric definitions sheet so every group counts the same thing.
7. 7Move up the analytic ladder in order: descriptive (what happened), predictive (what will happen), prescriptive (which option pays most). Prescriptive modeling of alternative learning investments is rare in HR. Use descriptive reporting to establish baselines, then relate effectiveness measures to outcomes with correlation and regression.
8. 8Reallocate and re-test. Shift money toward the programs whose measured effectiveness predicts outcomes, and cut or redesign those that don't. Refresh the assessment on a schedule. The source warns that last year's recruitment program can be made ineffective overnight, and that skill requirements age quickly.

The source's clearest real figures come from the Foreword. Research cited there says at least half of every dollar spent on training is wasted. The U.S. federal government spends about $30 billion a year on training, so roughly $15 billion a year is wasted. The same Foreword says some federal agencies now use automated analytic solutions and are seeing 'dramatic reductions in waste and significant improvements in productivity.' It also cites a vendor claim (KnowledgeAdvisors) that effective deployment of talent analytics yields at least a 4 percent productivity improvement. Treat both numbers as claims to test, not guarantees. The method that follows is to split the spend by program, attach an assessed skill gap to each, and measure efficiency (cost, time), effectiveness (competency, test passing) and outcomes (performance, turnover). Then rank the programs by measured return. The programs with no assessed gap and no outcome link are your likely half of wasted dollars, and they are the first candidates to cut or redesign.

• If you cannot state the assessed skill gap a program closes, don't fund it or expand it until the gap is measured.
• If you only have efficiency data (time to fill, cost to hire), add effectiveness and outcome measures before judging the investment. Cheap and fast are not the same as capable.
• If a program's logic model has a link with no measure, fix that link before spending more, because you cannot prove what you cannot see.
• If market conditions, technology, or regulations have shifted, re-assess the gap before renewing last year's recruiting or training plan.
• If data ownership or quality is unresolved, settle it before running regression or prediction. Otherwise you will model noise.

• Reporting activity instead of analysis. The tell: dashboards show courses delivered and hires made, with no link to performance or turnover. The source separates reporting from analysis.
• Funding programs without an assessed gap. The tell: no baseline competency or performance data exists before launch, so improvement cannot be shown.
• Relying on anecdote, or on extrapolating past trends. The tell: justification sounds like 'people liked it' or 'it worked last year,' even though the market has changed.
• Jumping to statistics before the question is defined. The tell: a team asks 'what should we measure?' before it can say what decision the data will inform.
• Ignoring data ownership and quality. The tell: metrics are defined differently by each group, and data requests stall or return mismatched formats.

• ▸Business problem and owner: one sentence naming the problem and the executive who owns it.
• ▸Strategic link: how this investment connects to how the company makes money.
• ▸Program and resources invested: recruiting, onboarding, training, coaching, or development, with budget and staff time.
• ▸Assessed skill gap and baseline: the capability gap and its current measured level (competency result, rating, test score).
• ▸Efficiency measures: for example time to fill, cost to hire, positions filled per month, with benchmark.
• ▸Effectiveness measures: for example speed to competency, percent passing tests, competency results, by group.
• ▸Outcome measures: engagement at 90 days and one year, turnover at two years, first-year productivity, exit survey results.
• ▸Data owner, source, and quality check: who holds each measure, whether they will share it, its format, and a validity note.
• ▸Decision and review date: expand, redesign, or cut, with the date you will re-assess the gap.

1. 1Define the three efficiency numbers before pulling data: time to fill (days), cost to hire, and salary cost of the position. Write each definition down, in the style of the Center for Talent Reporting's Standard Metrics Definitions the book cites, so every business unit counts the same way. Exhibit 6.6 (Definitions of Metrics) exists for this reason.
2. 2Send a data request with named fields and owners instead of an open ask. In the book's case, data arrived late and had to be processed manually in Excel. Use a data request template and a tracking tool so you know where each field lives, who owns it, and what form it is in.
3. 3Build the volume view first: open positions and positions filled per month. This is the baseline pair behind every other number, and it shows whether you are keeping pace with demand.
4. 4Compare time to fill against benchmarks, then salary costs and total cost to hire against benchmarks. A raw average with no external comparator tells you nothing about whether you are efficient.
5. 5Split the same three measures by business unit per quarter: open positions, average days to fill, average cost per hire. Efficiency problems usually sit in one unit, and a company-wide average hides them.
6. 6Place efficiency inside a logic model (inputs, activities, outputs, outcomes). Recruiter time and effort, supporting technology, offered salaries, and benefits go under inputs. Then ask the question efficiency cannot answer: how do we know we are hiring good people? If you have no new-hire quality metric, add one.
7. 7Link each hiring source and selection method to performance, potential ratings, and retention, as in the Staffing Process Analysis. Eyeball the treatments (testing, assessment, onboarding) before attempting statistics.
8. 8Report the direction and predictability of the numbers, not the counts. The book states that the number hired or paid is a record of expenditure, not value added. Present efficiency alongside effectiveness (e.g., 90-day performance) and outcomes (e.g., two-year turnover).

The book's VP walks you through a logic model for onboarding and retention of new hires. You list the inputs: recruiter time and effort, supporting technology, the salaries offered, and the benefits package. These are exactly the efficiency data you already hold. She then asks, 'How do we know if we are hiring good people?' and you answer frankly, 'We don't know.' You had no new-hire quality metric. Her real concern is an outcome: at least 30% of campus hires leave voluntarily within two years, above the industry benchmark. You add that to the scorecard and brainstorm causes: marketable skills from training, better work/life balance elsewhere, feeling undervalued, being underpaid. You then compile effectiveness data: 90-day performance ratings, competency results, speed to competency, engagement, and exit surveys. The Staffing Process Analysis shows why speed and cost alone mislead. Didi, Ken, and Leo were all newspaper hires. Didi had testing, assessment, and onboarding, was rated well, and stayed. Ken and Leo had testing only, rated below average, and left. Earl scored low and stayed, so you ask how many Earls you have. Cheap, fast newspaper hiring looks efficient until you see what happens after the hire.

• If time to fill and cost to hire look good against benchmarks but you cannot show new-hire quality, then add a quality metric before claiming efficiency.
• If one business unit's days to fill or cost per hire departs from benchmark, then investigate that unit's process (inputs and throughputs) rather than acting on the company average.
• If a source produces mixed results (as newspaper hires did), then compare the treatments the strong and weak hires received (assessment, onboarding) before cutting or favoring the source.
• If a downstream outcome such as two-year turnover exceeds benchmark, then do not push for faster or cheaper hiring. Work backward from the outcome to the likely causes and test them with engagement and exit data.
• If asked to report hiring volumes, then pair them with a trend or predictive relationship, because volumes alone only support cutting staff when business slows.

• Treating hires made and dollars spent as value delivered. The tell: your report contains counts and costs but no link to performance, engagement, or turnover.
• Reporting averages without benchmarks or business-unit splits. The tell: no one can say whether 'good' is good, or which unit is dragging the number.
• Starting from what to measure instead of the logical question. The tell: you collect everything recruiting tracks and still cannot answer the VP's 'why' question.
• Fixing a few visible blemishes and stopping, the way flippers treat obvious issues while central flaws stay hidden. The tell: time to fill improves while quality or retention stays flat or worsens.
• Ignoring outliers and manual data hygiene. The tell: late, hand-built Excel pulls, and people like Earl (low scores, still employed) going unexamined.

• ▸Business unit and quarter: name the unit and period so averages are not blended.
• ▸Open positions / positions filled (monthly): enter both counts for each month in the quarter.
• ▸Average days to fill vs. benchmark: record your average, the benchmark, and the gap.
• ▸Salary cost for the positions vs. benchmark: enter the salary cost and its benchmark comparison.
• ▸Total cost to hire vs. benchmark: enter the total cost per hire and its benchmark comparison.
• ▸Hiring source and selection treatments: list the source (e.g., newspaper) and what each hire received (testing, assessment, onboarding).
• ▸Effectiveness link: record 90-day performance/potential rating and speed to competency for the same hires.
• ▸Outcome link: record retention or turnover at the agreed horizon (e.g., two years) against benchmark.
• ▸New-hire quality metric: state the measure, or mark 'none' and name who will define it.
• ▸Action and owner: write the one process input or throughput you will change and who owns it.