KPI Design: Choosing 12 Metrics That Run the Company

Executive summary

KPI design is the work of deciding which numbers a leadership team will manage the company by, and then making each number precise enough that two people computing it get the same answer. The common failure is not a shortage of metrics. It is a dashboard of forty figures that nobody owns, three definitions of "margin" in circulation, and a monthly meeting that debates whose number is right instead of what to do.

This guide proposes a hard limit of twelve metrics for the company-level set. Twelve is not a magic number. It is small enough that every leader can recite the set from memory and large enough to cover cash, customers, delivery, margin, people, and the forward look. The method has five moves: start from the decisions the leadership team makes, generate candidates, score them against five criteria, write a definition card for each survivor, and attach an owner, a source, a cadence, and a target set by method.

If you only do one thing this week, take the metrics in your current monthly pack and write down, next to each one, the last decision it changed. The ones with no answer are candidates for removal.

When to use this

Use this guide when:

• A new CEO, CFO, or sponsor-appointed operator is setting up management reporting in the first 100 days.
• The monthly pack keeps growing, and meetings run out of time before reaching decisions.
• Two functions report different values for the same metric and the discussion stalls on reconciliation.
• Lenders, a board, or a sponsor have asked for a "scorecard" and you want one that the management team also uses.
• You are about to build or rebuild a dashboard and want to settle what goes on it before anyone opens a BI tool.

Do not start here if the books do not close on a predictable schedule. A KPI set built on numbers that move after publication will lose credibility in the first month. Fix close timing and reconciliations first, then design the set.

Diagnostic questions

Answer these about your current reporting before designing anything new. Most can be answered in a single working session with the finance lead.

1. How many metrics appear in the monthly management pack, and how many appear on the weekly view?
2. For each metric, what is the last decision it changed? Which have changed none in the past two quarters?
3. Is there a written definition (formula, inclusions, exclusions, source system) for each metric, and does it match what is actually calculated?
4. Does every metric have exactly one named owner who can explain a move, not just report it?
5. Which metrics describe results already locked in (revenue booked, margin earned), and which give early warning (backlog, pipeline, overdue invoices, overtime)?
6. Are there metrics that can be improved in the short run in ways that harm the business, such as delaying supplier payments to improve cash?
7. How long after period end is each metric available, and is that fast enough to act on it?
8. Which metrics depend on a manual spreadsheet step, and who maintains it?
9. Do targets exist for each metric, and how were they set?
10. Could a new leadership hire read the pack and understand how the company makes money and where it is at risk?
11. Which numbers do lenders, the board, or the sponsor see, and are they the same ones management uses?

Reading the answers: a high count in question 1 with few decisions in question 2 means you have a reporting problem, not a measurement problem. Gaps in questions 3 and 4 mean you must write definitions and assign owners before adding anything. Weak answers to 5 mean the set is all rear-view mirror.

The framework: from decisions to twelve metrics

Principle 1: A KPI exists to inform a decision

A metric earns a place only if a named person will act differently depending on its value. "Revenue per employee" can be interesting, but if no one will hire, fire, or reprice because of it, it is context, not a KPI. Context belongs in an appendix.

Principle 2: Cover the whole business with a few lenses

A balanced set avoids the trap of ten finance metrics and nothing on delivery. Use these lenses and aim for one to three metrics in each:

Principle 3: Mix leading and lagging indicators

Lagging metrics (margin, EBITDA, cash) tell you what happened. Leading metrics (backlog, pipeline coverage, overdue invoices, overtime hours) tell you what is about to happen while you can still act. A set that is more than two-thirds lagging will always be explaining the past. A rough guide is to make about a third of the set leading, and to pair each lagging outcome with at least one leading driver you can name.

Principle 4: One definition, one owner, one source

Each metric has a definition card (template below) that fixes the formula, the data source, what is included and excluded, and the owner. The definition lives in one place. When the definition changes, the change is dated and the history restated or footnoted. This is what ends meetings that argue about numbers.

Principle 5: Targets come from method

Set targets from sources you can defend: your own best trailing performance, contractual floors (a covenant, a customer service level, a supplier minimum), and named public peers where comparison is meaningful. Do not copy a "best in class" figure from an article and apply it to a different business model.

Step-by-step process

1. List the recurring decisions. Write down the 10 to 15 decisions the leadership team makes repeatedly: where to deploy cash, which customers to prioritize, when to hire, whether to expedite, what to price. Owner: CEO with the leadership team. Output: a decision list.
2. Generate candidate metrics. For each decision, name the one to three numbers that would change the answer. Pull in what lenders and the board already require. Owner: CFO or head of FP&A. Output: a candidate list, typically 20 to 30 items.
3. Score each candidate on five criteria (see the scoring template). Owner: CFO, with each function head scoring their own area. Output: a scored list.
4. Select twelve. Take the highest scores, then check balance across the five lenses and between leading and lagging. Swap a candidate if a lens is empty. Owner: CEO. Output: the proposed twelve and the cut list with reasons.
5. Write a definition card for each. Formula, source, inclusions and exclusions, refresh cadence, owner. Owner: metric owner, reviewed by finance. Output: twelve cards.
6. Test the data. Compute each metric for the last six to twelve months from source systems. Flag any that cannot be computed reliably. Owner: FP&A. Output: a history table and a list of data gaps.
7. Set targets and thresholds by method, with an amber and red trigger that prompts a specific response, not just a color. Owner: metric owner, approved by CEO. Output: target and trigger on each card.
8. Publish and freeze for a quarter. Owner: CEO. Output: a one-page scorecard and a change log.
9. Review the set every quarter. Retire metrics that changed no decision and keep the total at twelve. Owner: leadership team. Output: updated cards.

Templates

Metric definition card (copy one per metric)

Candidate scoring sheet

Score each candidate from 1 (weak) to 5 (strong). Multiply by the weight and sum.

Adjust the weights to your situation. A company under covenant pressure may weight decision relevance and timeliness higher. Keep the weights fixed once scoring starts so the comparison is fair.

One-page scorecard layout

Add a final row group for "what changed since last review" and "decisions requested." The scorecard is a decision document, not a report card.

Worked example (illustrative)

The numbers below are invented to demonstrate the method. They are not benchmarks, and the scores are one hypothetical team's judgments.

A mid-market manufacturer's leadership team lists 15 candidate metrics and scores each on the five criteria with weights of 30, 20, 20, 15, and 15 percent. The weighted score for the top candidate, "13-week cash balance versus minimum cash floor," is computed as 5 × 0.30 + 4 × 0.20 + 5 × 0.20 + 5 × 0.15 + 5 × 0.15 = 1.50 + 0.80 + 1.00 + 0.75 + 0.75 = 4.80.

The top twelve by score are rows 1 to 12. The team then runs the balance check:

• Cash and liquidity: rows 1 and 3 (two metrics).
• Profit and price: rows 6 and 8 (two).
• Customers and revenue: rows 5, 7, 10, 11 (four).
• Delivery and quality: rows 2, 4, 9 (three).
• People and capacity: row 12 (one).

That is 2 + 2 + 4 + 3 + 1 = 12, with every lens covered. Leading indicators in the set: cash versus floor, on-time delivery, quality escapes, backlog, pipeline coverage, overtime, and attrition, which is seven of twelve, so the team judges it slightly forward-looking. Lagging: cash conversion, gross margin, price realization, forecast accuracy, and net revenue retention, which is five. The team notes that "website sessions" scored high on data and timeliness but low on decision relevance, which is the pattern of a metric that is easy to chart and does not change any decision. It goes to the appendix.

Next, the team writes definition cards, names owners, and sets triggers. For gross margin, the target is the best trailing four-quarter value on a like-for-like mix, and the red trigger is a fall below the level the credit agreement implies for covenant headroom.

Metrics and KPIs

These are the meta-metrics that tell you whether the KPI system itself is working. They are not part of the twelve.

Setting targets for the twelve without invented benchmarks. Use three anchors in this order. First, a contractual or physical floor: a covenant threshold, a customer service level in a contract, a minimum cash level you set against the shortest funding lead time. Second, your own best trailing performance on a comparable basis, such as the best four quarters in the last eight. Third, a named public peer, where a listed company reports a comparable figure with a definition you can reconcile to yours. If you cannot name the source, do not use the number. For the meta-metrics above, aim for full definition and owner coverage, and treat any restatement as something to explain at the next review.

Common mistakes

1. Choosing metrics because they are available. The ERP exports certain numbers easily, so they end up on the dashboard. Start from decisions, then work out the data.
2. Letting the set grow. Each function lobbies for its own metric. Hold the line at twelve and require a swap, not an addition.
3. No definition cards. Without written formulas, "gross margin" drifts as people change inclusions. Meetings then argue about the number instead of acting on it.
4. Targets without method. A target pulled from a blog post or a prior sponsor's rule of thumb will not survive the first challenge from a function head. Tie it to a floor, your own history, or a named peer.
5. Gaming exposure. A single metric can be improved at the expense of the business. Pair it with a counter-metric: delivery speed with quality escapes, cash conversion with on-time supplier payment, margin with volume.
6. Owners who report but cannot explain. The owner must be able to say why the number moved and what they will do. If they cannot, the metric is owned by a spreadsheet.
7. Changing the set every month. Trends need a stable set. Freeze for a quarter, then change deliberately and log the change.
8. Colors without consequences. An amber status that requires nothing becomes wallpaper. Each trigger needs a stated response and a person who makes it.

AI prompt assistance

AI can speed up the drafting and the checking of definitions. It cannot know your business, and it will confidently invent formulas that look plausible. Give it your own data and verify every output against the ledger or source system before use.

1. Draft candidate metrics from decisions.

Here is a list of the recurring decisions our leadership team makes (pasted). For each decision, propose one to three metrics that would change the answer, say whether each is leading or lagging, and name the system of record that would normally hold the data. Do not propose more than three per decision.

Check: delete any suggestion where you cannot name the person who would act on it and the system the number comes from.

2. Stress-test a definition card.

Here is a metric definition card (pasted). Identify ambiguities in the formula, inclusions, or cut-off rule that would allow two analysts to compute different values from the same data. List each ambiguity and propose a precise wording. Also suggest one counter-metric that would reveal gaming.

Check: compute the metric both ways on one real month and confirm the difference you were warned about actually appears.

3. Spot gaming and perverse incentives.

Given this set of twelve metrics and their owners (pasted), list ways an owner could improve their metric while harming the business, and which other metric in the set would reveal it. Flag any metric with no counter-metric.

Check: review each flagged case with the owner. Treat the output as a prompt for discussion, not a finding.

4. Draft the scorecard commentary.

Using this month's values and targets (pasted), write a half-page summary for the CEO: metrics outside trigger, change since last month, and decisions requested. Do not add any figure that is not in the pasted data.

Check: reconcile every figure in the summary to the scorecard before it circulates.

Implementation guide

Governance: the CEO owns the set, the CFO owns definitions and data integrity, each function head owns the metrics in their area. The leadership team reviews the set quarterly. The board or sponsor sees the same twelve, not a separate view.

What to expect: the first scorecard will expose data problems. That is a result, not a failure. Expect two or three metrics to need a manual workaround for a quarter while source-system fixes are scheduled, and record them in the manual-step share. Resist adding metrics to fill the gaps. For the dashboard that displays the set, see the guide on executive dashboard design, and for the monthly pack that carries it, see the management reporting playbook.

Downloadable assets

The metric definition card, the candidate scoring sheet, and the one-page scorecard layout above are designed to be copied into a spreadsheet as they are. No calculator is currently planned for this topic in the operator library asset set.

Related resources

• CEO toolkit: the hub for operating-system guides written for CEOs of mid-market companies.
• Financial KPIs for executives: the finance metrics that often anchor the cash and profit lenses.
• Executive dashboard design: how to present the twelve so decisions come faster.
• Operating cadence: weekly, monthly, quarterly rhythms: where each metric is reviewed and by whom.
• Management reporting playbook: the monthly pack that carries the scorecard.
• Cash flow operating system playbook: the cash lens in depth.

Frequently asked questions

How many KPIs should a company have?

A company-level set works best at around twelve, which is enough to cover cash, profit, customers, delivery, and people but small enough that every leader can recall it. Each function can keep its own supporting metrics below that level. The limit is a discipline, not a rule: the test is whether each metric changed a decision this quarter.

What is the difference between a KPI and a metric?

A metric is any measurement. A KPI is a metric the leadership team has chosen to manage the business by, with a named owner, a written definition, and a target. Many metrics are useful context, but only a few deserve KPI status because someone will act differently when the value changes.

How do you set KPI targets without industry benchmarks?

Anchor to a contractual or physical floor, then to your own best trailing performance on a comparable basis, then to a named public peer whose definition you can reconcile to yours. If you cannot name the source of a target, do not use it. Targets taken from generic articles rarely fit your model and customer mix.

How often should KPIs be reviewed and changed?

Review values on the cadence that matches how fast each metric can be acted on: weekly for operating metrics, monthly for financial ones. Review the set itself quarterly, and freeze it between reviews so trends are comparable. When you change a definition, date the change and restate or footnote the history.

Should leading or lagging indicators come first?

Neither comes first. Lagging indicators confirm results, and leading indicators give time to act. Pair each lagging outcome with at least one leading driver you can name, and aim for roughly a third to a half of the set to be forward-looking. A set that is all lagging only explains the past.

How this works in a System of Performance

A KPI set only matters if the history behind each number stays connected: who owned it, what the definition was at the time, what trigger fired, what was decided, and what happened next. In most companies that trail is spread across spreadsheets, slide decks, and memory. A Performis routine can take on a recurring job like this, with the scope and fit agreed with you: it works from the records your team supplies, prepares the work for your named reviewer, and keeps each approval on record. If you want to see how reliably your own twelve numbers can be computed today, bring it to us as a launch partner.