Skip to content
Ashish Kumar AgnihotriOperating Partner

The KPI library

COO KPIs: The Scorecard Operating Leaders Actually Steer By

The useful question is not which KPIs a COO could track — the answer is hundreds — but which few a leadership team will actually trust and act on weekly. This library organises COO KPIs by theme — throughput, quality, decisions and cash — and by business model, with the pairing logic that keeps a scorecard honest.

Most operating dashboards fail by addition. Every review adds a metric, nobody ever removes one, and within a year the leadership team owns forty numbers it glances at and none it steers by. The discipline that works runs the other way: a handful of measures — typically five to nine — each with one named owner, one agreed definition, and one home on a single source of truth. A newsroom I scaled to roughly 400 stories a day ran on a small set of throughput and quality numbers leadership could read in real time; the restraint was not a simplification of the operating model but the core of it.

The themes below cover the ground a COO actually governs. Throughput: how much work moves, and where it waits. Quality: how much of it is right first time, and what the late catches cost. Decisions: how long the operation waits on judgement. Cash mechanics: how fast work becomes invoice becomes money — the operating side of working capital, distinct from finance’s view of it. Any operating scorecard worth the name draws from all four, then weights them by business model, because the number that steers an agency is background noise in a GCC and vice versa. The second half of this library maps that weighting.

The library reflects scorecards I have actually run, not a metrics glossary. Nineteen years in operations — Lean Six Sigma Green Belt, most recently Senior Director, Business Excellence at Publicis Groupe across 500+ clients and 2,000+ teams — including quality governance that lifted a score from 95% to 99% across 2,000+ campaigns and 450 clients, and billing governance that cut a cycle from roughly two months to fifteen days across 75 entities. Those numbers appear here not as trophies but as calibration: they are what it looks like when a scorecard is small enough to be trusted and connected enough to be acted on.

In depth

What you need to know.

The philosophy: fewer metrics, trusted more

A metric earns its place on a COO scorecard by passing three tests. Someone owns it — one named person whose number it is, not a department. Everyone believes it — the figure comes from a source the business already trusts, with one written definition, because the moment two versions of a number exist, meetings become debates about data instead of decisions about work. And someone would act differently if it moved — a metric nobody would respond to is decoration. Five to nine measures pass these tests in most operations; forty never do. The counterintuitive consequence: removing metrics usually improves control, because attention is the scarce resource a scorecard allocates. A leadership team that knows its seven numbers cold outperforms one that owns a dashboard it has to study.

Throughput KPIs: how work moves, and where it waits

The core set: cycle time — elapsed time from work accepted to work delivered, the single most revealing operating number because it captures every queue and handoff in one figure; throughput rate — units completed per day or week against capacity, whatever the unit is in your business (stories, campaigns, claims, shipments); on-time delivery — the share of commitments met as promised; and work-in-progress ageing — how long items have sat unfinished, which exposes the invisible queues averages hide. The Republic World turnaround was steered on almost exactly this set: mapping where stories waited, then watching cycle time and daily throughput as the redesign took hold — a fourfold lift to roughly 400 a day, with the numbers proving the gain came from flow, not pressure. Measure the waiting; the working takes care of itself.

Quality KPIs: right first time, and what late catches cost

The core set: first pass yield — the share of work accepted without rework or correction, the honest quality number because it prices the rework loop that error-rate figures flatter; defect or error rate at the point of use, counted where the customer meets the work rather than where the team finishes it; cost of poor quality — rework hours, credits, penalties, and in media the makegoods owed when delivery misses its guarantees; and audit or compliance scores where an external standard applies. Pair volume with quality always: a throughput gain that quietly spends quality is a deferred cost, not a gain. Staged checks weighted by exposure are what move these numbers at scale — the discipline behind a score lifted from 95% to 99% across 2,000+ campaigns, and behind more than USD 20 million in client billings protected from becoming makegoods.

Decision and cash KPIs: latency is the silent tax

Two families most scorecards omit, and the ones founders feel most. Decision latency: the elapsed time from an operational question being raised to being decided — measurable from the ageing of items on the weekly review’s decision log. When every judgement routes through one founder, this number is the size of the queue the company stands in, and watching it fall is watching the organisation learn to move without permission. Cash mechanics: billing cycle time — work completed to invoice raised to payment approved — plus unbilled work and approval ageing. This is operations’ share of working capital, and it is usually larger than finance assumes: a billing cycle cut from roughly two months to fifteen days across 75 entities released capital no financing decision could have. Neither number appears in standard KPI lists; both belong on the first scorecard.

Pairing leading and lagging indicators

A lagging indicator tells you what happened — revenue delivered, quality score achieved, on-time percentage for the quarter. A leading indicator tells you what is about to happen — pipeline coverage, work-in-progress ageing, mid-flight pacing against a delivery guarantee. A scorecard built only on lagging numbers manages by autopsy; only on leading ones, by forecast nobody is accountable for. The discipline is pairing: every lagging measure that matters gets a leading partner watched on the same cadence. On-time delivery pairs with WIP ageing, which deteriorates weeks before delivery dates slip. A quality score pairs with in-flight check results. Makegoods illustrate the logic perfectly: the settlement is the lagging number, and the pacing data — visible days before any shortfall becomes a liability — is the leading one. Catch the leading indicator and the lagging one stops needing to be explained.

KPIs by business model: agencies and professional services

An agency sells promises made against a calendar, so its scorecard weights delivery integrity and margin protection. The set that earns its place: on-time delivery against client commitments; first pass yield on client-facing work, because errors that reach a client cost trust as well as rework; delivery against guarantee for media — pacing mid-flight, shortfalls caught while correctable, makegoods as the lagging tally; utilisation and realisation, watched together since high utilisation with leaking realisation means busy people and vanishing margin; and revision cycles per deliverable, the quiet number that explains where capacity disappears. Scope creep hides in that last one. The agency failure mode is flying blind between monthly client reports; the fix is a weekly internal read on delivery health, before clients become the measurement system.

KPIs by business model: BPO, GCC and multi-entity operations

Process businesses at scale steer by consistency, and their KPIs must expose variance, not just averages. For BPO and GCC operations: service-level adherence and turnaround time by process; first pass yield per team, because an aggregate that blends a 99% team with a 91% team reports a comfortable 95% while a client relationship burns; cost per transaction trend; and attrition-adjusted capacity, since headcount on paper is not trained capacity on the floor. For multi-entity operations — groups, franchises, market networks — the unit of measurement is the entity: the same few numbers, defined identically everywhere, compared side by side, with the spread itself as a KPI. Closing the gap between best and worst entity is usually worth more than raising the average. The 75-entity billing transformation was steered exactly that way: one definition, every entity visible, the laggards unmistakable.

KPIs for founder-led companies — and how to build the scorecard

A founder-led company’s first scorecard has one extra job: making the operation visible without the founder in the room. Start with five numbers, not fifteen — one throughput, one quality, one cash, decision latency, and the single number that best expresses the promise customers buy. Build it in four moves: define each measure in writing, one sentence, one owner; baseline before targeting, because targets without baselines are wishes; put the scorecard on a single source of truth with one update rhythm; and review it weekly in an operating cadence where every red number leaves with an action and a date. Then hold the line on addition — every new metric must displace an old one. The scorecard is not reporting; it is the instrument panel the operating cadence flies by, and instruments only work when there are few enough to read.

Questions

Common questions.

Draw from four themes, then weight by business model. Throughput: cycle time, throughput rate, on-time delivery, work-in-progress ageing. Quality: first pass yield, error rate at the point of use, cost of poor quality. Decisions: decision latency — how long the operation waits on judgement. Cash mechanics: billing cycle time, unbilled work, approval ageing. A working scorecard takes five to nine of these, each with one owner and one written definition, on one source of truth. The selection matters less than the discipline: numbers the team trusts and reviews weekly beat any theoretically complete set.

Five to nine, reviewed weekly — few enough that the leadership team knows them cold, enough to cover throughput, quality, decisions and cash. Every candidate metric passes three tests: one named owner, one trusted source and definition, and someone who would act if it moved. Dashboards fail by addition — each review adds a number, none ever leaves, and within a year there are forty figures and no steering. Hold a displacement rule: a new metric must retire an old one. Attention is what a scorecard allocates, and attention does not scale past single digits.

First pass yield is the share of work accepted without rework, correction or repeat — right first time, measured honestly. It matters because it prices the loop most quality metrics hide: an operation can report a flattering final error rate while burning enormous capacity redoing work internally before anyone outside sees it. FPY exposes that cost, and improving it compounds — less rework releases capacity, which shortens cycle time, which improves delivery. It is also the number staged, in-flow quality checks move most directly, which is how a quality score goes from 95% to 99% across 2,000+ campaigns without adding inspection headcount everywhere.

Decision latency is the elapsed time from an operational question being raised to being decided — the silent tax on founder-led companies, where everything queues behind one calendar. Measure it without ceremony: keep a decision log in the weekly operating review — item, date raised, owner, date decided — and track the ageing of open items. The trend matters more than the absolute number. Falling latency means ownership and escalation rules are working and the organisation is learning to move without permission; rising latency almost always locates a bottleneck person, not a bottleneck process.

The ones that protect promises and margin: on-time delivery against client commitments; first pass yield on client-facing work; delivery against guarantee for media businesses — mid-flight pacing as the leading indicator, makegoods as the lagging tally; utilisation paired with realisation, because one without the other misleads; and revision cycles per deliverable, where scope creep hides. The pairing logic is the point: pacing watched weekly is what keeps shortfalls from becoming settlements — the discipline that protected more than USD 20 million in client billings at a global advertising network.

At process scale, variance is the enemy and averages are the disguise. Track service-level adherence, turnaround time and first pass yield per team and per entity — never only in aggregate, because a blended 95% can hide a team at 91% burning a client relationship. Add cost per transaction trend and attrition-adjusted capacity for BPO/GCC work. For multi-entity groups, define the same few numbers identically everywhere, compare entities side by side, and treat the spread between best and worst as a KPI in its own right — closing it is usually worth more than lifting the average.

Lagging KPIs report what happened — quarterly on-time percentage, the final quality score, makegoods paid. Leading KPIs signal what is about to happen — work-in-progress ageing, mid-flight pacing, in-process check results. A scorecard of only lagging numbers manages by autopsy; only leading ones, by unaccountable forecast. The working rule: every lagging measure that matters gets a leading partner on the same review cadence. WIP ageing deteriorates weeks before delivery dates slip; pacing shows a shortfall days before it becomes a makegood. Catch the leading number and the lagging one rarely needs explaining.

Judge any scorecard against five criteria. Small: five to nine numbers a leadership team knows cold. Owned: one named person per metric, one written definition, one trusted source. Balanced: throughput, quality, decisions and cash all represented, weighted for the business model. Paired: each lagging measure partnered with a leading one on the same cadence. And governed: reviewed weekly in an operating rhythm where red numbers leave with an action and a date. The best scorecard is not the most complete — it is the one the team actually steers by, unchanged in structure six months after it was built.

The next step

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

Book a consultationThe diagnostic

Keep exploring

The operations governance scorecard frameworkThe full framework this KPI library plugs into.The COO 90-day planWhere the scorecard gets baselined and installed.How to hire a fractional COOMeasurement is step five of the hiring process.The fixed-fee diagnosticBaseline your numbers before you target them.Fractional COO costJudge any retainer against the numbers it should move.Fractional COO vs a full-time COOEither way, the scorecard is the same.