When a CFO asks what the data team is worth, the answers usually come back in the wrong units. “We ship forty dashboards a quarter.” “We process two billion events a day.” “We have ninety-eight percent pipeline uptime.” All true, none of it answers the question. The CFO is asking about return on a million-dollar payroll line. The answers describe activity.

The metric that closes this gap is cost per decision: the fully-loaded cost of the data function divided by the number of meaningful business decisions the function supported in the period. It is awkward to compute, easy to game, and the most honest framing of data ROI most companies will ever produce.

Why output metrics fail

Output metrics measure what the data team did. They do not measure what changed because of it. Forty dashboards shipped is meaningless if thirty-five of them are unused. Two billion events processed is meaningless if the events feed a single weekly slide that nobody reads. Pipeline uptime is meaningless if the pipelines feed metrics nobody trusts.

The asymmetry is structural. Output metrics are easy to count, so teams report them. Outcome metrics — decisions changed, dollars moved, risks avoided — are hard to count, so teams don’t. The result is a budget conversation where the data team speaks one language and the CFO speaks another, and the gap gets filled by anecdote and political capital.

What “decision” means

A decision, for this purpose, is a discrete moment where someone with authority chose between options and the choice was materially informed by data the team produced. Not influenced. Informed — meaning the data was a primary input, not a confirmation of an already-made choice.

Concretely:

  • A pricing change recommended by the revenue ops analyst, approved by the CRO, and executed.
  • A vendor consolidation backed by a usage analysis from the data team.
  • A go/no-go on a market expansion driven by a cohort study.
  • An adjustment to a forecast model that materially changed a hiring plan.

What does not count:

  • Dashboards that get glanced at in a weekly meeting where the decision was already pre-made.
  • Reports that confirm what the executive already believed.
  • Self-serve queries an analyst ran for their own context-building.

The boundary is intent. If the data was the swing input, it counts. If the data was decoration on a decision already made, it doesn’t.

How to count it

The mechanic is uncomfortable but workable. Each quarter, the data team holds a decision review with the leaders of the functions it serves — revenue, product, finance, ops. The list is built collaboratively: “what decisions did your team make this quarter where data we produced was the swing input?” Each decision gets logged with a short description, the function, the approximate dollar magnitude of the choice, and a confidence rating on how central the data was.

You will be surprised by the number. Most teams overestimate. A function shipping forty dashboards a quarter typically supports somewhere between eight and twenty real decisions. The rest is reporting infrastructure, exploration, and self-justifying analysis.

Divide the data team’s fully-loaded cost — salaries, tooling, warehouse compute, the data leader’s portion of executive overhead — by that number. A team costing $2M a year that supports forty real decisions has a cost per decision of $50K. A team costing $2M a year that supports eight has a cost per decision of $250K. Both numbers are conversation-starters. Neither is automatically good or bad.

What the metric reveals

The metric does three things that output metrics cannot.

First, it forces the team to confront the gap between activity and impact. A quarter where the team shipped record output but supported five decisions is a quarter where the team built infrastructure for an audience that didn’t exist. That is a fixable problem, but only if it is visible.

Second, it changes the budget conversation. Asking for a headcount increase becomes “we believe this hire takes cost per decision from $80K to $60K because of the following capacity unlocks.” This is the language CFOs already use for every other function. Marketing speaks in CAC. Sales speaks in cost per closed deal. Finance speaks in cost per close cycle. The data team should speak the same way.

Third, it surfaces which parts of the team are decision-generative and which are infrastructure-generative. Both are necessary. The platform engineer who keeps the warehouse running rarely shows up directly in decision counts, but the analytics engineer whose models inform every decision does. A healthy data org has a recognized split between the two and budgets each accordingly.

The objections, and what to do about them

The first objection is that you cannot measure decisions cleanly. Correct. You cannot measure marketing attribution cleanly either, and marketing has been operating on imperfect attribution metrics for thirty years. The standard is not perfection; it is “directionally honest and consistently applied.”

The second objection is that this metric punishes teams that do platform work. It does, if you compute it naively. The fix is to track platform investment separately and amortize it across the decisions the platform supports over multiple quarters. Same accounting trick every infrastructure team uses.

The third objection is that the metric can be gamed by inflating decision counts. It can. The defense is that the decision review involves the leaders of the consuming functions, not the data team alone. A revenue VP will not sign off on a list of fictional decisions to pad someone else’s headcount case.

The thing that makes the metric work

Cost per decision works because it puts the data team’s value in the same unit the rest of the business uses. The CFO does not care how many dashboards exist. The CFO cares whether the company is making better choices than it was a year ago, and whether the data function is a measurable part of why.

Most data teams do not measure this because the answer might be uncomfortable. That discomfort is the point. A team confident in its impact will benefit from a metric that proves it. A team unsure of its impact needs the metric to find out.