CLC_Selecting_HR_Metrics

1. 1Start from the decision, not the data. Write the business objective (Verizon's example: become a leading global player in telecommunications) and the HR priority beneath it (e.g., Staffing/Leadership). Name the decision the metric must inform, such as where to put retention effort or whether to change recruiting. A metric with no named decision fails this construct.
2. 2Cascade the priority down an inverse tree: strategic priority, then factors that affect it, then detailed measures, then raw data. Verizon's tree runs Staffing/Leadership through factors like Retention, Promotions, Workforce Diversity and Talent Acquisition to measures like Extended Offers, Accepted Offers, New Recruit Yield, Hourly and Salaried Short-Term Terminations, and Right Fit. Stop only when each branch ends in a real number you can pull.
3. 3Translate each raw count into a rate, ratio, composition or index so it carries meaning. Turnover Rate = [Terminations / Average Headcount] x 100. Applicant Ratio = Applicants / Offers Accepted. Average Time-to-Fill = Total Days to Fill / Offers Accepted. Use composition (e.g., HR Expense Breakdown) when the question is where money goes.
4. 4Segment each translated metric to find where the problem sits. The passages list the cut dimensions: organization structure, demographics, employment levels, functions, and employee experience. Function cuts are job family, job title and employment type (regular/temporary). Experience cuts are organization tenure, position tenure, performance ratings and benefits participation. Run the cuts the decision calls for, not all of them.
5. 5Compare segments to the whole and to each other. A segment whose rate departs sharply from the company rate is a candidate intervention target. A pattern that repeats across a population, such as exits clustering at a tenure band or a job family, is a behavior pattern worth acting on.
6. 6Score each candidate metric on GM's decision matrix. Code it X (direct measure) or C (contributory) against each business objective. Add quality points: Subjective 0, Quantifiable 1, Results 1, Qualitative 3, Predictive 6. Then add the feasibility rating: Easy +1, Moderately Difficult -1, Difficult -2. The sum is the comparative rating.
7. 7Weight predictive metrics heavily. A predictive metric 'raises the red flag' as performance starts to deteriorate, which lets management put countermeasures in place. It is also the only quality category worth 6 points. Prefer it whenever the decision is about acting early.
8. 8Keep the metric set stable and review it with line managers. The passages say metrics deliver value when 'selected and used consistently', so changing definitions each cycle destroys the trend you need to see.

Verizon's priority is Staffing/Leadership, under the objective of becoming a leading global player in telecommunications. The tree reaches measures such as Hourly Short-Term Terminations, Salaried Short-Term Terminations, New Recruit Yield and Right Fit. Raw terminations are translated with the turnover formula: [Terminations / Average Headcount] x 100. One company-wide number hides where the loss occurs, so you cut it by the tree's own split (hourly vs. salaried) and then by job family, employment type, position tenure and performance rating. If short-term exits concentrate in one group, that group is the high-impact intervention target. If they cluster at a tenure band, you have a behavior pattern across the population. The recruiting measures (Applicant Ratio, Average Time-to-Fill, New Recruit Yield) then test whether the cause lies upstream in who was hired. As a scoring illustration using GM's point values (not a figure from the source): a quantifiable (1), qualitative (3) metric with data already available (+1) rates 5. The same metric with predictive power (+6) rates 11. A predictive metric needing data that does not yet exist (-2) still rates 9, so you weigh the build cost against the early-warning value.

• If a metric cannot name the decision it informs, drop it, however easy it is to collect.
• If a company-wide rate moves or looks high, segment it before acting. Cut by function, then experience, then demographics, and intervene only where the segment diverges from the whole.
• If two metrics have similar comparative ratings, pick the predictive one. The 6-point weight reflects the value of early countermeasures.
• If a high-value metric rates Difficult to Measure (-2), compare its total with and without the penalty. Build the data if it still ranks high; otherwise take the next-best easy metric.
• If a measure is subjective (rated 0, e.g., employee morale), find a quantifiable proxy before using it to steer a decision.

• Reporting only aggregate rates. The tell: every segment discussion ends with 'turnover is X% overall' and no group is ever named as the target.
• Choosing metrics because the data is already there. The tell: the matrix is dominated by +1 feasibility scores and no predictive metric appears among the finalists.
• Stopping the cascade at the factor level (e.g., 'Retention') without reaching raw data. The tell: no one can say which number to pull or who owns it.
• Slicing by every dimension at once. The tell: dozens of small segments, no stable pattern, and no decision tied to any of them.
• Scoring metrics without linking them to objectives. The tell: blank X/C coding, or a high comparative rating on a metric that contributes to no business objective.

• ▸Business objective and HR priority: write the corporate objective and the HR priority beneath it.
• ▸Decision this metric informs: one sentence naming the action and who takes it.
• ▸Metric and formula: name it and write the rate/ratio/composition/index formula.
• ▸Raw data sources: list each raw number at the bottom of the cascade and its owner.
• ▸Segmentation cuts to run: tick function (job family, title, employment type), experience (tenure, position tenure, performance rating, benefits participation), organization, demographics, level.
• ▸Objective coding: mark X (direct) or C (contributory) against each objective.
• ▸Quality points: Subjective 0, Quantifiable 1, Results 1, Qualitative 3, Predictive 6. Enter each category that applies and the quality total.
• ▸Feasibility rating: Easy +1, Moderately Difficult -1, Difficult -2.
• ▸Comparative rating and intervention target: quality total plus feasibility, then name the segment or pattern that would trigger action.

1. 1Start from the business objective, not from the data you already hold. Gather external information, interview line managers and executives, survey HR customers, and collect HR leadership feedback. The aim is to write down what the business requires of HR. Sort those requirements into categories (workforce development, labor relations, customer service, compensation and benefits) to form the HR strategy.
2. 2Build a cascade for each priority: strategic priority, then factors that affect it, then detailed measures, then raw data. Verizon draws this as an inverse tree. Keep going until the bottom layer is real numbers, such as extended offers or hourly short-term terminations. If a branch ends in an abstraction, it is not yet a metric.
3. 3Run the tree through more than one lens. Verizon tests priorities against four perspectives: strategic, operations, customer, and financial. The priority layer is identical across all four, so you can see which priorities have a financial measure and which have none.
4. 4Convert raw data into formulas that carry meaning: rates, ratios, percentages, and indexes. Express each item relative to time, population, or expense. Terminations become terminations over headcount. HR function expense becomes a share of total HR expense, [Function Expense / Total HR Expense] x 100. Survey inputs become an index, such as a Manager Quality Index.
5. 5Segment each formula before you report it. Cut by organization structure, demographics, employment level, function, and employment experience (organization tenure, performance rating, position tenure). Segmentation isolates high-impact targets for intervention. A company-wide average hides them.
6. 6Score every candidate metric on the GM matrix. Code it X (direct measure) or C (contributory measure) against each business objective. Then rate quality: subjective 0, quantifiable 1, results 1, qualitative 3, predictive 6. Then rate feasibility: easy +1, moderately difficult -1, difficult -2. Comparative rating = quality total + feasibility.
7. 7Rank by comparative rating, then check the objective coding. Keep the metrics that cover every objective with at least one direct (X) measure. Use the highest-scoring predictive metrics as early warnings. Use results metrics to state the financial contribution.
8. 8Use the chosen metrics consistently and report them against the objectives. The impact case rests on two outputs: where workforce performance and productivity can improve, and which HR areas need cost control and greater efficiency.

Take Verizon's business objective: position the company as a leading global player in telecommunications. One HR priority under it is Staffing/Leadership, and the tree breaks it into Retention, Promotions, Workforce Diversity, and Talent Acquisition. Talent Acquisition yields Extended Offers, Accepted Offers, and New Recruit Yield. Retention yields Hourly Short-Term Terminations and Salaried Short-Term Terminations, under the factor 'Right Fit'. The raw data are offer counts and termination counts. Step 3 turns the termination count into a rate against headcount. Segmenting by hourly versus salaried shows where the loss concentrates. For scoring, apply GM's scale to the short-term termination rate. These numbers are an illustration built on the passage's scale, not figures from GM or Verizon. It is quantifiable (1) and arguably a results metric (1), and the data already exist (+1), so the comparative rating is 3. Scored against a predictive metric such as strength of employment brand (6), the termination rate ranks lower. A predictive metric flags deterioration early, so it earns more weight. GM's own sample matrix, which compares HR:FTE ratio, new hire turnover, turnover by segment, employee productivity, and skills gap, produces comparative ratings of 6, 6, 4, 2, and 9. The scale separates strong candidates from weak ones. The decision matrix gives HR a defensible reason to report some metrics and drop others.

• If a branch of the cascade ends without raw data, then keep decomposing it. Do not report an abstraction as a metric.
• If two metrics tie on comparative rating, then prefer the one coded X (direct) against a business objective over one coded C (contributory).
• If a metric is high quality but difficult to measure (-2), then discount it, but do not discard it. Check whether a moderately difficult version exists whose data already exist (-1).
• If a company-wide figure looks acceptable, then segment it by structure, demographics, level, function, and experience before concluding anything. Averages hide the high-impact targets.
• If a metric is only subjective (0), then look for a quantifiable proxy (1). If the metric would give early warning, then favor the predictive category (6) over simple counts.

• Choosing metrics because the data are easy to pull. The tell: the metric list has no link to a business objective, and the objective columns of the matrix are empty.
• Reporting raw counts instead of rates and ratios. The tell: dashboards show numbers of terminations or hires with no headcount, time, or expense denominator.
• Reporting one organization-wide average. The tell: leaders ask 'where?' and HR cannot answer without a new data pull.
• Treating the matrix score as the whole decision. The tell: a high-scoring metric (for example one that is easy and quantifiable) is kept even though it is coded C against every objective and no direct measure is covered.
• Changing the metric set every cycle. The tell: no trend lines exist, because the guidance depends on using selected metrics consistently.

• ▸Business objective: state the corporate objective this metric must serve.
• ▸HR priority and cascade path: write priority > factor > detailed measure > raw data.
• ▸Metric name and formula: give the rate, ratio, or index and its denominator.
• ▸Segmentation cuts: list the dimensions you will cut by (organization, demographics, level, function, experience).
• ▸Objective coding: mark X (direct) or C (contributory) for each business objective.
• ▸Quality rating: mark each category met (subjective 0, quantifiable 1, results 1, qualitative 3, predictive 6) and total them.
• ▸Feasibility rating: easy +1, moderately difficult -1, difficult -2.
• ▸Comparative rating: quality total + feasibility.
• ▸Decision and use: adopt or drop, with the business decision this metric informs (productivity gain or cost/efficiency control).

1. 1Start with the business objective, not the metric list. Write the corporate objective as one sentence at the top of the page. Verizon's reads: 'Position Verizon to become a leading global player in telecommunications.' Every metric you keep must trace back to this line.
2. 2Gather the evidence for HR's role from four sources: external information (competitors and the market), interviews with line managers and executives, surveys of HR customers, and HR leadership feedback. Add a review of corporate and business unit objectives. Write down the people requirements these sources produce.
3. 3Group the people requirements into categories such as workforce development, labor relations, customer service, and compensation and benefits. This yields the human capital strategy. Verizon's version has six: Staffing/Leadership, Labor Relations, Diversity, Compensation/Benefits, Workforce Development, and Customer Service. Each carries 1-3 stated priorities, such as 'Define leadership competencies' or 'Build partnership with unions'.
4. 4Build an inverse tree for each priority. Verizon evaluates priorities against four Balanced Scorecard perspectives: strategic, operations, customer, and financial. The first level is the priority, which is identical across all four perspectives. The second level is the factors that affect it. The third is the detailed measures. The bottom level is raw data. Keep splitting until the bottom node is a real number someone could pull.
5. 5Check the tree both ways. Reading down, every leaf must be raw data. Reading up, every leaf must reach a strategic priority and then the business objective. Delete or flag any leaf that links to nothing.
6. 6Convert the surviving raw data into rates, ratios, or compositions (e.g., Turnover Rate = [Terminations / Average Headcount] x 100; Average Time-to-Fill = Total Days to Fill a Position / Offers Accepted). Segment where useful, for example by organization structure, demographics, function, or employee experience.
7. 7Code each candidate against each business objective as a direct measure (X) or a contributory measure (C), as GM's decision matrix does. A metric with no code against any objective is not aligned. Remove it, or return it to the tree to find the missing link.
8. 8Pass the aligned set through quality and feasibility scoring before you commit. Alignment decides what is eligible. Usability and ease of collection decide what is chosen.

Verizon's priority is Staffing/Leadership, which supports the objective of becoming a leading global player in telecommunications. Under the strategic perspective, the factors include Retention, Promotions, Workforce Diversity, and Talent Acquisition. Talent Acquisition breaks into detailed measures such as Extended Offers, Accepted Offers, New Recruit Yield, and 'Right Fit', which is measured through Hourly Short-Term Terminations and Salaried Short-Term Terminations. The bottom of the tree holds the raw data: counts of offers extended, offers accepted, and short-term terminations. Step 3 turns these counts into formulas. Applicants / Offers Accepted gives the applicant ratio, and Total Days to Fill a Position / Offers Accepted gives average time-to-fill. Every number can be traced up through the factor and the priority to the business objective. Raw counts that cannot be traced this way would have no place in the set.

• If a candidate metric cannot be placed on a branch of the tree below a strategic priority, then drop it or rebuild the tree. Do not add it because the data is easy to get.
• If a branch stops at an abstract measure (e.g., 'Right Fit' or 'employee morale') with no raw data beneath it, then keep splitting until you reach countable items. Subjective terms score 0 at GM.
• If a metric is coded X (direct) against an objective, then it takes priority over C (contributory) metrics for that objective. Use C metrics to explain movement in the X measure.
• If the same priority layer appears across all four Balanced Scorecard perspectives, then build separate trees per perspective. Do not reuse one tree's measures for another.
• If a metric is strongly aligned but scores low on feasibility (e.g., -2 at GM, where the metric and its data may not exist), then do not discard it. Keep it on the list with a data-build plan, and rely on the easy-to-measure aligned metrics in the meantime.

• Starting from available HR data and reverse-justifying it to strategy. The tell: the metric list matches the HRIS report menu, and the objective column is filled in last.
• Skipping the requirements gathering, so HR priorities reflect HR's own view. The tell: no line manager or executive input is documented, and priorities look like HR function labels with no business objective behind them.
• Stopping the cascade at a broad measure. The tell: a leaf like 'diversity' or 'retention' with no formula or raw data beneath it, so nobody can say what to pull.
• Treating alignment as pass/fail and ignoring usability. The tell: a long list of aligned metrics with no quality or feasibility ratings, so they cannot be ranked or compared.
• Having one person assign the codes and ratings with no review. The tell: X/C coding and ratings that no stakeholder has validated, so the matrix looks precise but reflects one person's judgment.

• ▸Business objective: write the corporate objective in one sentence.
• ▸Evidence sources used: list the external, line-manager, HR-customer, and HR-leadership inputs behind HR's priorities.
• ▸HR strategic priority: name one priority (e.g., Staffing/Leadership) and its stated goals.
• ▸Balanced Scorecard perspective: strategic, operations, customer, or financial.
• ▸Factors: list what affects the priority (e.g., retention, promotions, talent acquisition).
• ▸Detailed measures: break each factor into constituent measures.
• ▸Raw data (leaf): name the countable item and where it is pulled from.
• ▸Formula type and expression: rate, ratio, or composition, plus the formula and any segmentation dimension.
• ▸Alignment code: mark X (direct) or C (contributory) per business objective; leave blank if untraceable.
• ▸Quality and feasibility rating, comparative total, and keep/hold/drop decision: score using the GM categories and record the outcome.

1. 1Fix the top of the tree first. List the HR strategic priorities that come out of Step 1 (Verizon's: Staffing/Leadership, Labor Relations, Diversity, Compensation/Benefits, Workforce Development, Customer Service). Each one hangs from a stated business objective. Verizon's is 'Position Verizon to become a leading global player in telecommunications.'
2. 2Choose the perspectives you will test the priorities against. Verizon uses four: strategic, operations, customer, and financial. The priority layer is identical across all four, so build it once and copy it onto each perspective's tree.
3. 3Draw one inverse tree per perspective with four labeled levels: Strategic Priorities, Factors, Detailed Measures, 'Raw Data'. Work top-down and ask at each level, 'What affects this?' Factors are the things that affect a priority. Detailed measures are the constituent parts of each factor.
4. 4Keep breaking down until a node is a real number someone can pull from a system. In Verizon's strategic-perspective tree, Staffing/Leadership runs to Talent Acquisition, then Extended Offers, Accepted Offers, and New Recruit Yield, and also to Right Fit, which carries Hourly and Salaried Short-Term Terminations. Stop only at that raw-data level.
5. 5Translate each raw-data leaf into a formula type: Rate (Terminations / Average Headcount x 100), Ratio (Total Days to Fill a Position / Offers Accepted), Composition (part / whole x 100), or Index (survey-weighted score). Choose the type by the question you need answered: volume, cost, time, satisfaction, or quality.
6. 6Attach segmentation dimensions to each measure: organization structure, demographics, employment levels, function, and employment experience. For example, split short-term terminations into hourly and salaried, or by tenure, so the result points to a specific population to act on.
7. 7Trace every surviving measure back up the branch to its priority and business objective, and write that path down. A measure with no path up is an orphan; drop it.
8. 8Send the candidate list to Step 4 evaluation (strategic alignment, quality/usability, feasibility). The tree generates candidates; the evaluation selects them.

Take Verizon's strategic-perspective tree for the Staffing/Leadership priority, which supports the objective of becoming a leading global player in telecommunications. Level 2 (factors) includes Talent Acquisition and Right Fit. Under Talent Acquisition, the detailed measures are Extended Offers, Accepted Offers, and New Recruit Yield. Under Right Fit, they are Hourly Short-Term Terminations and Salaried Short-Term Terminations. The raw data are the counts behind each of these. Step 3 then converts the counts into formulas. Applicant Ratio is Applicants / Offers Accepted. Average Time-to-Fill is Total Days to Fill a Position / Offers Accepted. Turnover Rate is Terminations / Average Headcount x 100. Segmenting short-term terminations by hourly versus salaried shows which population is failing to fit. The result is a chain from the abstract priority 'invest in leadership growth' down to counts a team can actually collect.

• If a branch ends in a concept ('engagement,' 'service capability') and not a countable item, keep decomposing. A tree is finished only when its leaves are raw data.
• If a raw number is hard to interpret alone, convert it: use a rate for proportion of a whole, a ratio for one number relative to another, composition for a breakdown, and an index to combine disparate survey data into one number.
• If a result looks flat or average, segment it by dimension (organization structure, demographics, employment level, function, experience) before drawing conclusions. Differences between populations isolate high-impact targets for intervention.
• If a priority appears in the tree but you cannot name a business objective above it, remove it or go back to Step 1. Alignment runs through the tree, not around it.
• If two measures answer the same question, keep the one with higher alignment, usability, and ease of collection (the Step 4 criteria).

• Stopping at the factor or detailed-measure level. The tell: leaves are labels like 'Retention' or 'Talent Acquisition' that no one can pull from a system.
• Building the tree from available data upward instead of from priorities downward. The tell: the metrics list mirrors the HRIS reports and cannot be traced to a business objective.
• Presenting raw counts without formulas. The tell: a dashboard of totals (terminations, offers) with no rate, ratio, or time context, so no one can say whether a number is good.
• Skipping segmentation. The tell: one organization-wide figure, such as a single turnover rate, hides the hourly versus salaried difference, so no action target emerges.
• Treating the cascade as the end of the process. The tell: every tree leaf becomes a metric, producing a long list with no evaluation of alignment, usability, or feasibility.

• ▸Business objective: write the corporate objective this tree supports.
• ▸HR strategic priority: name one priority (e.g., Staffing/Leadership) from Step 1.
• ▸Perspective: strategic, operations, customer, or financial.
• ▸Factors: list what affects the priority (e.g., Talent Acquisition, Right Fit).
• ▸Detailed measures: break each factor into its constituent parts (e.g., Accepted Offers, Short-Term Terminations).
• ▸Raw data leaf: the specific countable item and its source system.
• ▸Formula type and formula: Rate, ratio, composition, or index; write the calculation.
• ▸Segmentation dimensions: choose from organization structure, demographics, employment level, function, experience.
• ▸Handoff to Step 4: note alignment, usability, and ease-of-collection concerns for evaluation.

1. 1Start from the measurement area, not the data. Take the measurement areas that Step 2 derived from the cascade (business goal to HR objective to detailed measurement item). Write down which raw data elements feed each one before building any formula.
2. 2Pick the output type for each raw data element. The passages name four forms: a rate, a ratio, a composition (a share of a whole), and an index. Choose by the question: volume, cost, time, satisfaction, or quality.
3. 3Build compositions with the passage's pattern: [part / total] x 100. For Expense Breakdown, the formula is [HR Function Expense (Benefits Administration, Payroll, Staffing) / Total HR Expense] x 100. Compute one line per HR function so the lines sum to 100%.
4. 4Build an index only when several disparate inputs must become one number. An index is a weighted combination of disparate data, expressed relative to a scale or anchor. Write down the inputs, the weights, and the anchor before calculating. The passages' examples are the Manager Quality Index (employee perceptions of manager quality, from surveys) and New Hire Performance Satisfaction (how well recent hires perform in their position, from surveys).
5. 5Choose segmentation cuts from the five dimensions in Table 2: Organization Structure (financial reporting, supervisory reporting, HR responsibility, geographic location), Demographics (age, gender, ethnicity, education level), Employment Levels (exempt/non-exempt, hourly/salaried, EEO job category, manager level, compensation band, annual salary), Function (job family, job title, regular/temporary), and Employment Experience (organization tenure, performance ratings, benefits participation, position tenure). Pick the cuts that match the decision the metric supports, not every available cut.
6. 6Run the metric at the total level first, then re-run it by each chosen segment. Look for populations or business units where the result departs from the total. Those gaps are the candidates for high-impact intervention targets and for behavior patterns across the workforce.
7. 7Pass each translated metric into Step 4. Score it on alignment with strategy, usability, and ease of collection, and keep the metrics that give the most insight into business objectives.
8. 8Record the definition once, including formula, data source, segment list, and owner, so the metric is calculated the same way every period. The source stresses that metrics work when they are selected and used consistently.

Take the Expense Breakdown composition from the source. Assume HR wants to know where its spend goes. For each HR function (Benefits Administration, Payroll, Staffing), divide that function's expense by Total HR Expense and multiply by 100. The three shares show how total HR spend is distributed, which flags where cost control and efficiency work belongs. The same logic applies to survey-based indices. The Manager Quality Index reflects employees' perceptions of manager quality from survey results. Reported only as one company-wide number, it hides variation. Cutting it by Organization Structure (supervisory reporting, geographic location) and by Employment Experience (organization tenure) shows whether low scores concentrate in particular units or among particular tenure groups. Those concentrations are the places the source says to isolate as high-impact intervention targets. The company-wide figure alone could not do that.

• If the question is how a whole divides (e.g., HR spend by function), use a composition: part / total x 100. If it asks for one score from several disparate inputs, build a weighted index with a stated anchor.
• If the metric is a survey perception (manager quality, new hire performance satisfaction), report it as an index and segment it before acting on it.
• If a company-wide result looks acceptable, segment it anyway by organization structure or experience. Variation across populations is where intervention targets appear.
• If a segment cut does not map to a decision someone can take, drop it. Choose from the five dimensions based on the action the result would trigger.
• If a translated metric scores poorly on strategy alignment, usability, or ease of collection, do not keep it, however easy the formula is to compute.

• Reporting only the aggregate. The tell: leaders say 'the number is fine' while a unit or tenure group is struggling, and no one can say where the problem sits.
• Building an index with no stated weights or anchor. The tell: nobody can explain what a score of, say, 72 means or why it moved.
• Starting from available data instead of measurement areas derived from strategy. The tell: formulas exist with no link to a business goal or HR objective.
• Segmenting by every dimension at once. The tell: dashboards with many cuts, each with too little data to read, and no named action attached to any of them.
• Inconsistent definitions across periods or units. The tell: a composition's parts do not sum to 100%, or the same metric name yields different results across units.

• ▸Measurement area: name the area and the HR objective it traces back to in the cascade.
• ▸Raw data elements: list each source data point and where it comes from.
• ▸Output type: rate, ratio, composition, or index.
• ▸Formula: write it out, e.g., [HR Function Expense / Total HR Expense] x 100; for an index, list inputs, weights, and anchor.
• ▸Information revealed: volume, cost, time, satisfaction, or quality.
• ▸Segmentation dimensions chosen: Organization Structure, Demographics, Employment Levels, Function, or Employment Experience, with the specific cuts.
• ▸Decision the segmented result informs: name the intervention or action.
• ▸Step 4 scores: alignment with strategy, usability, ease of collection.
• ▸Owner and review cadence: who calculates it and how often.

1. 1Build the candidate list from the cascading tree before you score anything. Each metric should be a detailed measure that traces from a strategic priority through its factors down to raw data, and ideally already expressed as a rate, ratio, composition or index. Scoring vague topics like 'engagement' produces meaningless totals.
2. 2Set up the decision matrix with one column per candidate metric. Add one row per business objective, then one row per quality category (Subjective, Quantifiable, Results, Qualitative, Predictive), then a feasibility row and a comparative-rating row.
3. 3Code alignment first. For each metric and each business objective, mark X if the metric directly measures achievement of the objective, C if it only contributes, and leave it blank if there is no link. A metric with no X or C against any objective is dropped before quality scoring.
4. 4Rate quality using GM's definitions and values: Subjective 0 (cannot be expressed as a number), Quantifiable 1 (a numeric or visible trend goal can be set), Results 1 (directly affects financial performance), Qualitative 3 (reflects quality of performance), Predictive 6 (raises the red flag as performance deteriorates and supports problem solving). The categories stack, so a metric can score in several and you add them.
5. 5Rate feasibility with one value per metric: Easy to Measure +1 (metric or data already exists), Moderately Difficult -1 (metric doesn't exist; data exists but isn't readily available), Difficult -2 (metric doesn't exist; data may not exist).
6. 6Compute the comparative rating as quality total + feasibility rating. Rank metrics from highest to lowest. The number is a simple way to compare the usability of all candidates; it does not replace judgment about strategy.
7. 7Assign one owner on the HR planning team to draft the categories and values. The values should be weighted to what your organization values (GM weights Predictive at 6 because predictive metrics let management act proactively). Circulate the draft so the metrics planning team can challenge categories and ratings before the ranking is used.
8. 8Review the ranking against the alignment coding. Adopt top-ranked metrics that carry X marks against priority objectives. Treat high-scoring metrics with only C marks as supporting measures, not primary ones.

GM's sample decision matrix scores five candidates: HR:FTE Ratio, New Hire Turnover, Turnover by Employee Segment, Employee Productivity and Skills Gap. Feasibility ratings are 1, 1, -1, 1, -1, and comparative ratings are 6, 6, 4, 2, 9. Subtracting feasibility from each comparative rating gives quality totals of 5, 5, 5, 1 and 10, which is what the quality rows imply: every metric earns 1 for Quantifiable, three earn 1 for Results, and four earn 3 for Qualitative. One earns the single Predictive 6. Skills Gap therefore ranks first at 9 even though it is hard to measure (-1), because its Predictive 6 outweighs the feasibility penalty. HR:FTE Ratio and New Hire Turnover tie at 6 with easy data. Turnover by Employee Segment drops to 4 because of the -1 feasibility. Employee Productivity ranks last at 2, since it is quantifiable only. The per-category assignment for each metric is inferred from the totals; the source prints only the row values.

• If a metric scores Predictive (6), expect it to outrank easy-to-collect metrics even with a -1 or -2 feasibility; GM treats predictive metrics as considerably more valuable, so accept the data-collection work.
• If two metrics tie on comparative rating, prefer the one coded X (direct) against a priority objective over one coded C (contributory).
• If a metric is Subjective (0) and has no quantifiable form, then either redefine it as a number or rate it low; it can't carry a numeric goal or trend.
• If a metric's feasibility is -2 (data may not exist), then don't adopt it on score alone; check whether a Moderately Difficult proxy built from existing data gets the same insight.
• If stakeholders dispute the ranking, challenge the category values (for example, whether Predictive should be 6) before challenging individual ratings, because the weights drive the outcome.

• Scoring without alignment coding: the matrix ranks 'usable' metrics that tie to no business objective. Tell: top-ranked metrics have empty X/C cells.
• Treating the comparative rating as objective truth. The values are weighted to GM's priorities and set by one planner. Tell: nobody on the planning team has reviewed or questioned the category values.
• Letting feasibility dominate by always choosing easy data. Tell: the shortlist is all +1 metrics with no Predictive or Qualitative scores, so it only reports the past.
• Rating topics rather than defined metrics. Tell: ratings differ between reviewers because the metric has no formula or raw-data source.
• Mixing up the two quality/feasibility scales by adding feasibility into the quality total twice, or ignoring the negative values. Tell: comparative ratings don't equal quality total plus the single feasibility value.

• ▸Metric name and formula: write the defined metric (for example, [Terminations / Average Headcount] x 100), not a topic.
• ▸Strategic priority and raw-data source: note the branch of the cascading tree and where the data originates.
• ▸Objective alignment: for each business objective, enter X (direct), C (contributory) or blank.
• ▸Quality ratings: enter Subjective 0, Quantifiable 1, Results 1, Qualitative 3, Predictive 6 for every category that applies.
• ▸Quality total: sum the quality ratings.
• ▸Feasibility rating: enter +1 (Easy), -1 (Moderately Difficult) or -2 (Difficult).
• ▸Comparative rating: quality total + feasibility rating.
• ▸Rank and decision: record rank order and adopt, hold or drop.
• ▸Reviewer feedback: note planning-team challenges to categories or ratings and the resulting changes.

1. 1Start from the cascade, not from a wish list. Build the tree: strategic priority, then factors, then detailed measures, then raw data. At Verizon, Customer Service breaks into Staffing/Leadership, Retention, and so on, down to Extended Offers, Accepted Offers, and Hourly Short-Term Terminations. Every candidate metric must trace to a raw-data leaf.
2. 2Code each candidate against the business objectives. Mark it X if it directly measures the objective and C if it only contributes. A candidate with no X or C against any objective leaves the matrix before scoring.
3. 3Express the raw data as a rate, ratio, composition, or index (for example, Turnover Rate = Terminations / Average Headcount x 100). Score the formula, not the raw count, because the formula is what carries the insight.
4. 4Score the five quality categories with GM's values: Subjective 0, Quantifiable 1, Results 1, Qualitative 3, Predictive 6. Give credit only when the definition is met. Predictive means the metric 'raises the red flag' while performance is starting to deteriorate, so management can still apply countermeasures.
5. 5Score feasibility with one value: Easy to Measure +1 (the data exists now), Moderately Difficult -1 (the data exists but is not readily available), Difficult -2 (the data may not exist).
6. 6Compute the comparative rating as the quality total plus the feasibility rating. Rank all candidates in one matrix so the totals can be compared side by side.
7. 7Review the ranking against the objective coding. Make sure each objective is covered by at least one X metric and that the set includes predictive metrics, not only after-the-fact ones.
8. 8Plan segmentation for the survivors before launch. Choose the cuts (organization structure, demographics, employment level, function, employment experience) that isolate high-impact intervention targets, such as Turnover by Employee Segment.

The GM decision matrix compares five candidates: HR:FTE Ratio, New Hire Turnover, Turnover by Employee Segment, Employee Productivity, and Skills Gap. Their comparative ratings are 6, 6, 4, 2, and 9. The highest total, 9, is the only one that includes the 6-point predictive credit. It is arithmetically consistent with Quantifiable 1 + Qualitative 3 + Predictive 6 + Moderately Difficult -1. The passage's column layout is garbled, so treat that mapping as an inference. The lesson holds either way: a hard-to-collect predictive metric can still outscore easy metrics. Here is the same arithmetic on the passage's own category examples. The feasibility ratings are assumed for illustration. 'Hours of training' is quantifiable (1) and assumed easy (+1), so it scores 2. 'Utilization of training' is quantifiable (1), qualitative (3), and assumed easy (+1), so it scores 5. 'Strength of employment brand' is quantifiable (1), qualitative (3), and predictive (6), but assumed moderately difficult (-1), so it scores 9. 'Employee morale' is subjective, so it earns 0 until someone defines a numeric version. Employment brand ranks first despite the collection effort, because it flags deterioration early.

• If a metric is predictive, keep it even when it is Difficult to Measure. At -2 it still nets +4 from the predictive credit, and a predictive metric that is also quantifiable and qualitative can score 8 net, above an easy non-predictive metric's best possible 6.
• If a metric scores 0 as Subjective (for example, employee morale), then define a numeric version, such as an index built from survey results, and rescore it. Otherwise drop it.
• If two metrics tie on comparative rating, then keep the one coded X (direct) against an objective and the one with the predictive credit.
• If a metric has no X or C against any business objective, then remove it regardless of its score, because the rating measures usability, not strategic fit.
• If the surviving set contains no predictive metric, then go back down the tree and look for a leading indicator for the priority, such as employment brand strength for Talent Acquisition, before approving the set.

• Choosing by ease of collection. The tell is a dashboard full of +1 feasibility metrics, with no qualitative or predictive credit, that only reports what already happened.
• Treating 'qualitative' as 'subjective'. The tell is a metric dismissed or scored 0 because it involves surveys, even though qualitative means it reflects the quality of performance (worth 3) and can be numeric.
• Scoring without the objective coding. The tell is a high-scoring metric, such as a 9, that no business objective asks for.
• Skipping the translation step and reporting raw counts. The tell is metrics like 'number of terminations' with no headcount denominator, so no one can say whether the number is good or bad.
• Approving a Difficult-to-Measure metric with no collection plan. The tell is a metric that scores well on paper but has no data owner or start date, so it never reaches the scorecard.

• ▸Candidate metric and formula: write the name and the rate, ratio, composition, or index formula.
• ▸Source leaf in the cascade: record the priority, factor, and raw-data item it traces to.
• ▸Objective coding: enter X (direct) or C (contributory) for each business objective; leave blank if none.
• ▸Quality points: enter Subjective 0, Quantifiable 1, Results 1, Qualitative 3, Predictive 6, and record the evidence for each credit claimed.
• ▸Predictive test: state in one line what deterioration this metric flags and what countermeasure it would trigger.
• ▸Feasibility rating: enter Easy +1, Moderately Difficult -1, or Difficult -2, and name where the data lives.
• ▸Comparative rating: add the quality points and the feasibility rating.
• ▸Segmentation cuts: list the dimensions (organization, demographics, level, function, experience) used to analyze it.
• ▸Decision: write keep, rework, or drop, with a one-line reason and a data owner.