RevOps Metrics: The Numbers the Function Owns

Two dashboards open side by side in a Monday meeting report different pipeline totals for the same quarter, and both are correct. One counts opportunities from the moment a meeting is booked, the other from the moment a rep marks a stage. The argument that follows runs for forty minutes and settles nothing, because the disagreement is about a definition and the meeting is about a number.

That is the situation a revenue operations function exists to remove, and it is also the reason RevOps metrics have to be separate from the sales scoreboard. The sales numbers on the dashboard measure selling. They say nothing about whether the machinery producing them is sound, and a confident figure computed from inconsistent inputs is more expensive than an honest gap.

This page sets out the numbers the function itself owns, what each one is computed from, and who is supposed to act on it. Every figure used below to illustrate a calculation is invented for the illustration and describes no real company.

Why the function needs separate numbers

What the RevOps function actually owns is definitions, the systems and the wiring between them, data quality, and routing. None of those four shows up on a sales dashboard, and all four decide whether the sales dashboard is true.

The distinction that matters is between an outcome metric and an integrity metric. Closed revenue against plan is an outcome. The share of closed-won opportunities carrying a populated segment field is an integrity metric, and it is the one that says whether the outcome can be cut by segment at all. A team reporting only outcomes cannot tell a real change from a reporting artefact.

The second reason is ownership. Who owns which decision is the boundary that causes the most argument at most companies, and a metric with two owners has none. Every number below has a single owner by construction, because it measures something one function can change.

The RevOps metrics worth carrying

Six numbers cover most of what the function can be held to. The point of the list is its shortness: a dashboard with fifty measures is a dashboard nobody reads, and the operator rules on reporting put the working ceiling at five per dashboard, which is the discipline this set is meant to sit just inside.

Field completeness on decision fields. The share of records carrying a populated value in the fields any report cuts by: segment, source, owner, close date, and whichever custom fields the business actually plans on. Computed as populated records divided by records in scope, per field, on the population a report uses rather than on the whole database. This is the number that decides whether every other cut is honest.

Definition compliance. The share of open opportunities whose current stage is supported by the exit criteria written for that stage. It is a sampled measure rather than a computed one, because no system holds the evidence, and it has to be sampled by somebody willing to move a deal backwards. A stage set that nobody audits drifts within two quarters into a set of labels reps apply by feel.

Routing latency. Elapsed time from a qualifying event to first human contact, measured at a percentile rather than as a mean. The mean hides the tail, and the tail is where the leads are lost. Read the ninetieth percentile beside the median, because the gap between them is the story.

Reconciliation gaps between connected systems. For each integration that matters, the count of records present in one system and absent or contradicted in the other, plus the age of the oldest unresolved gap. Every seam between two tools produces this, and an integration nobody instruments produces it silently.

Forecast accuracy. Forecast at a fixed point in the period against the eventual actual, expressed as a signed variance rather than an absolute one, and tracked across periods so that a consistent bias becomes visible. A team that is consistently light by the same margin has a correctable bias. A team whose variance swings in both directions has a pipeline data problem, which is a different repair.

Cost of the stack per revenue-carrying head. Total annual spend on the revenue tooling divided by the number of people whose work it supports. This is the number that gets a renewal conversation started early enough to matter, and it is the one most often computed only after a renewal has auto-renewed.

The properties that make one of these usable

A measure earns its place on two conditions, and both are easy to check.

It has a denominator you can state out loud. A bare completeness figure means nothing until the population is named. A rate computed over all records in the database and the same rate computed over the records in this quarter's closed-won report are different claims, and the second one is what a segment cut depends on. The operator rule is blunt about this: show the denominator or do not show the rate.

Somebody can act on it this week. A number that moves only after a quarter of structural work is a project status rather than a metric. Field completeness moves when a validation rule ships. Routing latency moves when an assignment rule changes. Forecast accuracy moves slowly and belongs on a quarterly read for that reason.

Applying both filters is what keeps the list at six. What usually gets added to a RevOps dashboard fails the second test: it is genuinely interesting, nobody can change it inside a period, and it stays on the screen until it becomes furniture.

An invented worked example, to show the shape

The following figures are invented for illustration and describe no real company.

A team reports 480 closed-won opportunities in a quarter, an invented figure as are all of the numbers here. Segment is populated on 361 of them, which is an invented completeness rate of about 75% on the reported population. The quarterly review then presents average deal size by segment across those 480 deals.

The presented cut leaves out a quarter of the book, and the omission is not random: records created before a form change carry no segment, and those skew toward one acquisition channel. The segment comparison in that deck is therefore a comparison between two populations rather than two segments, and nothing on the slide says so.

The repair is not analysis. It is a backfill against the source that carries the value, plus a validation rule that stops the field being empty on creation, plus a completeness figure printed beside the cut from then on. That sequence, in that order, is what the function does day to day.

Two data-quality calls the numbers depend on

Reconciliation and completeness both rest on judgement calls about individual records, and two come up often enough to need a rule.

Is a vendor's closed-business flag trustworthy? Treat the flag as a claim with a source and a date. Before it removes an account from targeting or reporting, check where the vendor got it, when that source was last refreshed, and whether a second, independent source agrees, such as the business's own website or listing. A flag one source asserts and nothing corroborates goes to a review queue rather than the suppression list, and the share of a vendor's flags overturned on review is that vendor's reconciliation number.

Is it a true duplicate, or two similarly named local businesses? Treat two records as distinct until an identifier says otherwise. Match on things that cannot coincide by accident, such as a registered company number, a website domain, a phone number or a street address, and require two of them before merging. A shared name at a different address is two businesses. Log every merge, because an unlogged merge is a reconciliation gap nobody can reverse.

What none of these numbers says

They say nothing about whether the business is selling well. That is the point of them, and it is also the failure mode: a function measured only on integrity metrics optimises for a clean database that nobody uses to decide anything.

The counterweight is that every integrity metric should be traceable to a decision it protects. Field completeness protects segment analysis. Routing latency protects conversion on inbound. Reconciliation protects the forecast. If a measure cannot name the decision it protects, it is hygiene for its own sake and it should come off the dashboard.

The sales-side set sits beside these rather than under them. Which numbers change a decision covers the outcome half and the pairing rule that makes each of them readable: one leading figure, one lagging, one diagnostic, and a companion for each.

Tooling comes after all of it. RevOps tools sorted by the job each one does is the buying question, and buying a forecasting platform before the definitions are settled produces a confident forecast from inconsistent inputs, which is a more expensive kind of wrong than a spreadsheet because it carries more authority.

Which metrics should a RevOps team report on weekly?

Three RevOps metrics belong in a weekly report: field completeness on the fields reports cut by, reconciliation gaps per integration, and routing latency at the median and the ninetieth percentile. All three move inside a week and degrade quietly. Definition compliance is a quarterly audit, forecast accuracy a per-period read, and stack cost an annual number.

That split follows the operator rule of weekly operating metrics and quarterly strategic ones, never the reverse. Definition compliance stays quarterly because sampling it properly costs a person a day and doing it badly is worse than not doing it. Forecast accuracy is a per-period read by construction. Stack cost is annual, timed to land a quarter before the largest renewal.

The instruction that keeps the whole set honest is that each number goes to a named person with the authority to change the thing it measures. A completeness figure reported to a room where nobody owns the form is a complaint.

Where the reading says the constraint is not machinery at all but the number of qualified conversations entering it, no dashboard repairs that. We are paid on attended meetings that meet criteria agreed in writing before launch, which prices supply against the same unit a pipeline target is written in. See what a first campaign produces.

The short version

A revenue operations function needs integrity metrics rather than a second copy of the sales scoreboard. Six carry most of the weight: field completeness on the fields reports cut by, definition compliance sampled by audit, routing latency at a percentile, reconciliation gaps per integration, signed forecast variance across periods, and stack cost per revenue-carrying head.

Each one needs a stated denominator, a single owner, and a decision it protects. A measure that fails any of the three is furniture, and the fastest way to make this list useful is to keep it short enough that somebody reads all of it.