Skip to content

KPI reporting

KPI Reporting Framework: Choosing Metrics That Survive Contact With the Business

Almost every company I meet has too many KPIs and too little agreement about them. Forty metrics is not rigour; it is a refusal to choose. A KPI reporting framework is the set of rules that decides what gets measured, who owns it, what good looks like, and who reads it when. This is the one I use.

By Ashish Kumar Agnihotri·Last reviewed

The forty-metric problem is worth naming precisely, because it looks like diligence. A leadership team wants control, so each function proposes the numbers it tracks. Nobody wants to be the person whose area is unmeasured. The list grows, gets pasted into a monthly pack, and is presented in full at a review where there is time to discuss perhaps five items. What follows is predictable. Attention distributes itself by whoever speaks loudest rather than by what matters. Metrics with no threshold sit there for quarters without anyone noticing they have drifted. And because everything is measured, nothing is prioritised. Choosing fewer metrics is an act of management, not of simplification.

A framework prevents that drift by making selection a rule-governed process rather than a negotiation. It answers five questions in order. What outcomes does this business have to get right — usually four or five, no more. For each outcome, what one lagging metric proves it, and what one leading metric predicts it. Who owns each metric, meaning who can change the number. What is the threshold that separates acceptable from not, agreed before anyone sees the data. And who reads what, at what interval: the board sees the smallest set, leadership sees five to nine on a weekly screen, teams see their own operating detail daily. Everything else in KPI reporting is craft applied to those five answers.

I have built and rebuilt this kind of structure for most of nineteen years in operations. Most recently as Senior Director, Business Excellence at Publicis Groupe, where quality and delivery spanned 500+ clients, teams of 2,000+ and more than USD 750 million in annual media spend for brands including Disney, Samsung, Adobe and P&G. Earlier, building HR scorecards and business intelligence across 23 business units at Raymond, and reporting and audits covering 4,500+ retail stores across 19 telecom circles for Vodafone. Different sectors, the same failure modes. I now work independently with companies between fifty and five hundred people in India, the USA, the UK and Europe.

In depth

What you need to know about KPI reporting framework.

What a KPI reporting framework actually is

It is not a list of metrics. It is the set of rules by which metrics are chosen, defined, owned, thresholded, reviewed and retired. A company with a good framework and a mediocre first set of KPIs will converge on the right numbers within two quarters. A company with an excellent metric list and no framework will drift back to forty metrics within a year, because there is no rule that stops anyone adding one. The framework is what survives personnel changes. It should be written down in perhaps two pages: the outcomes the business is managing, the selection tests, the definition standard, the ownership rule, the threshold-setting method, and the reporting cadence with its audiences. If it takes more than two pages, it will not be followed.

The five tests a metric must pass

I put every candidate through the same tests, and I am unsentimental about the failures. Decision: does a bad reading change what someone does next week? Ownership: can one named person move this number, or is it the average of forces nobody controls? Definition: can two people compute it independently and get the same answer? Timeliness: does it arrive while the situation is still changeable, or only after the quarter has closed? Resistance: if someone optimised this metric aggressively and cynically, would the business be better or worse off? That last test kills more metrics than the other four combined. Anything failing two tests is out. Anything failing the decision test is out immediately, however satisfying it is to know.

Leading versus lagging, and the pairing rule

Lagging indicators measure what happened: delivered quality, on-time completion, revenue, cost per unit, escaped defects reaching a client. They are what the board is accountable for and they are unarguable, but by the time they move, the causes are weeks old. Leading indicators measure the conditions that produce those outcomes: queue ageing, first-pass yield at an early process step, capacity against booked demand, training completion before a launch, sampling coverage on a new account. The rule I use is one of each per outcome. Every lagging metric that leadership cares about gets exactly one paired leading metric that someone can act on now. A framework of only lagging indicators produces an eloquent post-mortem culture. One of only leading indicators eventually loses the plot about what the business is for.

Choosing five to nine, and retiring the rest

Five to nine metrics is the size I hold to for a leadership scorecard, and it is deliberate. It is roughly what a group of executives can hold in mind, argue about properly, and act on within a single week. Getting from forty to nine is a subtraction exercise done in one room with the leadership team, using the five tests, and it is uncomfortable — every metric has a sponsor. Two things make it easier. First, distinguish leadership metrics from operating metrics: a function may legitimately track twenty numbers; the question is only which ones reach the leadership screen. Second, retire rather than delete. Move demoted metrics to a documented reserve list with a note on why they were dropped, so the decision is revisitable and nobody feels erased.

Setting targets that mean something

An unthresholded metric is decoration. But targets set by ambition alone are equally useless, because a number that is missed every week stops being information. I set them in this order. Establish the baseline honestly, over enough history to see the variation, not a single flattering month. Distinguish the acceptable threshold — below which the business is materially harmed — from the improvement target, which is where you are trying to get to by a stated date. Make the threshold binary and visible on the screen. Then set the improvement target with a named owner, a date, and a stated mechanism: not merely quality to 99 per cent, but which change in the process produces it. When I took an enterprise media operation's quality from 95 per cent to 99 per cent across 2,000+ campaigns and 450 clients, the target was only useful because the mechanism was named alongside it.

Definitions: the invisible half of the framework

More KPI reporting fails on definitions than on selection. On-time delivery sounds unambiguous until you ask whether the clock starts at brief receipt or at approved brief, whether client delays pause it, and whether partial delivery counts. Each of those choices is defensible; what is fatal is that different teams silently made different ones. Every metric in the framework needs a written entry: plain-language description, numerator, denominator, inclusions, exclusions, source system of record, refresh frequency, owner, and the date the definition last changed. That last field matters more than it looks — when a metric jumps, the first question should be whether the business changed or the definition did. Definitions belong to the business, not to the data team, which is the point where KPI reporting and data governance become the same conversation.

The reporting cadence and its audiences

Three tiers, three intervals, three different documents. Teams see their own operating detail daily or at shift level, in whatever granularity helps them work — this tier can be large, because it is used by the people who own the process. Leadership sees five to nine metrics weekly, on one screen, in a short exception-driven review where each breach gets a cause, an action and a date. The board sees a smaller set monthly or quarterly, with trend and narrative, oriented to outcomes rather than to operations. The frequent mistake is showing the board the leadership screen, which invites directors into weekly operational management, or showing leadership the team detail, which buries the exceptions. Match the granularity to the decision rights of the audience.

Failure modes worth designing against

Four recur. Vanity metrics: numbers that only ever go up, such as cumulative totals, which cannot generate a decision. Gaming: any metric owned by the person who also controls its measurement will eventually be managed rather than improved, which is why sampling and independent audit belong beside self-reported quality. Metric inflation: the slow return to forty, prevented by a one-in-one-out rule on the leadership scorecard. And orphaning: a metric whose owner has left or changed roles, still on the screen, answered by nobody. A quarterly review of the framework itself catches all four in about ninety minutes. Skip it for a year and you will be rebuilding rather than maintaining.

Questions

Common questions about KPI reporting framework.

Five to nine on the leadership scorecard. That is what a leadership group can genuinely hold, debate and act on in a weekly review, and it is the size I hold to across engagements. It is not a cap on organisational measurement — individual functions may reasonably track many more operating numbers — it is a cap on what reaches the leadership screen. Every metric beyond nine dilutes the attention available for the others. If a tenth metric is genuinely essential, the honest move is to decide which of the existing nine it replaces.

Lagging KPIs measure outcomes after they have happened: delivered quality, on-time rate, cost per unit, escaped defects. Leading KPIs measure the conditions that produce those outcomes while there is still time to act: queue ageing, first-pass yield early in the process, capacity against booked demand, readiness before a launch. Use both, paired. My rule is one leading metric for each lagging outcome leadership cares about. Lagging alone gives you accurate history and no steering. Leading alone drifts away from the results anyone is accountable for.

Five. Decision: does a bad reading change behaviour next week? Ownership: can one named person move it? Definition: would two people computing it independently agree? Timeliness: does it arrive while the situation can still be changed? Resistance to gaming: if someone optimised it cynically, would the business be better off? Fail two and the metric is out; fail the decision test and it is out immediately. The gaming test eliminates the most candidates, and it is the one companies most often skip.

Do it in one session with the whole leadership team, using written tests rather than opinion, so the exercise is about rules and not about whose area is being demoted. Separate leadership metrics from function-level operating metrics first — that alone usually removes half the list without anyone losing their numbers. Then apply the tests and keep five to nine. Retire the rest to a documented reserve list with the reason recorded, so the decision can be revisited. Finally, adopt a one-in-one-out rule, or you will be back at forty within a year.

Start from an honest baseline over enough history to show normal variation, not one good month. Then separate two numbers: the acceptable threshold, below which the business is materially harmed, and the improvement target you are working towards by a stated date. Show the threshold on the screen as a binary condition. Attach the improvement target to a named owner, a date and a stated mechanism — the specific change expected to produce the movement. Targets without mechanisms become aspirations, and a target missed every week quietly stops being information at all.

The business. Specifically, the one named leader who can change the number. The data team owns the pipeline, the refresh and the accuracy of the calculation; the operating leader owns the performance and answers for it in the review. Confusing the two turns every review into a debate about whether the report is right instead of what the business should do. Shared ownership behaves identically to no ownership. If two leaders genuinely both influence a metric, either split it or assign it to whoever is accountable for the outcome.

Operate on three tiers. Teams look at their own operating detail daily. Leadership reviews the five-to-nine scorecard weekly, exception-driven, with a cause, an action and a date recorded for each breach. The board sees a smaller, outcome-oriented set monthly or quarterly with trend and narrative. Separately, review the framework itself once a quarter: check for orphaned metrics with no current owner, definitions that have quietly changed, thresholds nobody has revisited since they were set, and metric creep. That review takes about ninety minutes and prevents an eventual rebuild.

Ambiguous definitions above all — two teams computing on-time delivery differently, then arguing about the report instead of the business. After that: metrics nobody owns, metrics with no threshold so drift goes unnoticed, cumulative vanity numbers that can only rise, and self-reported quality with no independent sampling behind it. Each has a cheap fix applied early and an expensive one applied late. Write the definitions before you build the reporting, name an owner per metric, agree thresholds before seeing the data, and audit a sample of anything that is self-reported.

The next step

A short conversation settles most of this — and a fixed-fee diagnostic settles the rest.

Book a consultationCall +91 98115 73155

Keep exploring

COO KPIs worth trackingCandidate metrics by operating area.COO scorecard templateA structure for the five-to-nine set.The Operations Governance ScorecardThe framework this reporting sits inside.Operations dashboard consultantTurning the metric set into one usable screen.How I measure engagementsThe same tests applied to my own work.BI, reporting and insightsThe service this framework belongs to.
Call