AIReturn

The CFO & the AI budget

How to Allocate the AI Budget per Team (The Cost×Rework Decision)

Allocate the AI budget by plotting each team on the Cost×Rework matrix: scale, keep, fix, or cut. The CFO's quarterly ritual.

Rodrigo Paredes BassiPublished 11 min read

TL;DR

  • Allocate the AI budget by plotting each team on the Cost×Rework matrix — AI cost per outcome on one axis, rework on the other — and take the move each quadrant dictates: scale, keep, fix, or cut.
  • The four moves: low cost / low rework → keep it running; high cost / low rework → scale it, where return compounds; low cost / high rework → fix the workflow or harness; high cost / high rework → cut or rebuild. The verdict is per team, read against each team's own baseline.
  • Make it a quarterly ritual, not a one-time audit. Worldwide AI spending is set to reach $2.59T in 2026, up 47% year over year (Gartner, May 2026) — the allocation has to be re-run as fast as the spend grows.
  • The number that forces the ritual: 66% of boards now condition further AI funding on proof of return (CloudZero, Jun 2026). The matrix is the artifact that answers them.
  • Cost the AI granularly — by skill, model, and product — so the denominator is defensible, and net rework out of the numerator so a high-adoption, high-rework team can't hide inside a blended average.

How do you allocate an AI budget across teams?

Plot each team on a 2×2: AI cost per outcome (high/low) against rework level (high/low). Each quadrant returns a budget move — scale, keep, fix, or cut — read against that team's own pre-AI baseline. Scale what's expensive but clean, keep what's cheap and clean running, remediate what's cheap but sloppy, and cut what's both costly and error-prone. Re-run it every quarter as spend grows. This is the decision layer of how a CFO governs the AI budget per team: not how to measure the return, but what to do with the measurement. Below is the step-by-step, the quadrant table with the move and its justification, and the quarterly cadence.

Definitions: the three terms the allocation rests on

Plant these once; every move below reads off them. Cost × Rework matrix is a 2×2 that plots each team on AI cost (high/low) against AI rework (high/low), returning a per-team verdict — scale, keep, fix, or cut. It is the CFO's decision artifact for allocating the AI budget team by team, not a dashboard to watch. Per-team AI budget is the practice of funding AI spend team by team based on proven output-net-of-rework, rather than one org-wide "AI budget" nobody can defend at the board. It converts AI from a blanket line item into a governed, evidence-based allocation. Output-per-dollar-net-of-rework is the business output a team ships per dollar of AI spend, after subtracting the cost of redoing flawed AI work. It is what each axis of the matrix is built from: cost is the denominator, rework-adjusted output is the numerator.

Why allocate per team — and why now

One org-wide "AI budget" can't be defended, because AI's return isn't uniform across functions. A blended figure averages the team that's compounding return with the team that's burning spend on work that gets redone — and hands the board a number that hides the decision instead of making it. Allocation has to happen where the money is actually spent: team by team. The timing is forced by two facts. First, scale: worldwide AI spending is set to reach $2.59T in 2026, up 47% year over year (Gartner, May 2026). A budget growing that fast can't be governed by an annual guess. Second, scrutiny: 66% of boards now condition further AI funding on proof of return (CloudZero, Jun 2026). The per-team allocation is the artifact that turns "trust us, it's working" into a defensible move per team.

The 5 steps to allocate the AI budget per team

Each step produces one input to the matrix. Run them in order, per team, every quarter.

Step 1 — Name the outcome each team is paid to ship

Define the unit of work each team actually delivers — not activity, the outcome. Engineering merges pull requests and closes tickets; sales closes deals; support resolves cases; product ships stories and initiatives. Pick one primary outcome per team, because the AI cost per outcome (the matrix's cost axis) is undefined until "done" is defined. This is also why the budget can't be one org-wide figure: a closed-won deal and a merged PR don't share units.

Step 2 — Compute AI cost per outcome (the cost axis)

Cost the AI granularly — by skill, model, and product, including multiple products at once — then divide by the team's accepted outcomes. An aggregate token bill can't tell you which team, workflow, or model is generating spend; per-skill, per-model, per-product attribution can, which is what makes the cost axis defensible. OpenTelemetry (OTEL) instrumentation captures this at the level of the work. In AIReturn's model this is v1 scope — AI-usage (token) cost across Copilot, ChatGPT, Claude, Cursor, and in-house agents. Fully-loaded salary cost is v2 and deliberately out of scope today, which keeps HR-surveillance sensitivity off the table. The mechanics are the AI unit economics behind cost per outcome.

Step 3 — Measure the rework level (the rework axis)

Measure the redo cost — the friction and touchpoints it takes to reach an accepted outcome — through function-native signals: code churn, reverts, and review rounds in engineering; reopened tickets and escalations in support; stage regression in sales; requirements/spec volatility in product. Normalize each to an index and compare it only against that team's own history, never against another team. A team whose output rises while rework rises faster is spending more to ship less that sticks — and that only shows up on this axis. Honest limit: the lag between AI-produced work and the rework it causes isn't fully pinned in every function, and isolating AI-caused rework from a team's baseline rework is the hardest open question here. In engineering it's concrete — churn, reverts, reopens, and review rounds are already logged. For deals, cases, and initiatives the proxies are still maturing. State the boundary; don't over-claim a precise dollar of rework everywhere.

Step 4 — Plot each team and read its quadrant

Place each team on the 2×2 using cost per outcome (Step 2) and rework level (Step 3). Each quadrant returns a move — scale, keep, fix, or cut — detailed in the table below. The point of the plot is that it makes the decision visible: at a glance, which teams earn more spend, which to leave running, which need their workflow remediated before another dollar, and which should be cut. This is how to measure AI ROI team by team rendered as a single decision surface.

Step 5 — Move the money and set the next review

Read the quadrant and reallocate. Fund the scale teams (expensive but clean, where return compounds), keep the cheap-and-clean ones running, remediate the sloppy-but-cheap ones before adding spend, and cut or rebuild the rest — then hand finance a governed per-team AI budget rather than a blanket line item. Set the next review one quarter out, because rework moves with models and workflows and the spend keeps growing. To see the verdict rendered as a deliverable, look at a sample per-team AI ROI report.

The Cost × Rework matrix: the move per quadrant

This is the decision table. Read each team's position, take the move, and note what evidence justifies it. AI cost per outcome runs one axis; rework level runs the other.

QuadrantAI cost per outcomeRework levelThe moveWhy the evidence justifies it
ScaleHighLowIncrease the budget. Return compounds — the team ships accepted outcomes that rarely get redone, and more spend buys more of them.Clean output that sticks is positive output-per-dollar-net-of-rework. High cost isn't waste when rework is low — it's leverage worth funding.
KeepLowLowLeave it running. Quiet and efficient — cheap output that rarely gets redone. No change needed; revisit at the next review.Low cost and low rework is already a healthy return. There's nothing to fix and no reason to cut; the money is working.
Fix the workflow / harnessLowHighHold spend; remediate first. Cheap but sloppy — the AI produces output that gets redone. Fix the workflow, prompt, tools, or context before adding a dollar.High rework with low cost means the process is broken, not the budget. Scaling spend here just buys more rework. For agents, this is the harness to tune.
Cut or rebuildHighHighStop or redesign. Expensive and error-prone — no output to show for the spend. Cut the use case or rebuild it from the ground up.Both axes are red: high cost per outcome and high rework means negative output-per-dollar-net-of-rework. This is where budget quietly leaks; cutting it is the return.
The matrix is the single most-quoted object in the room. Each team is read against its own baseline, never as a cross-team scoreboard — output isn't comparable across functions. The model is most built-out in engineering today; other functions expand as their output and rework definitions mature.

A worked allocation: four teams, four moves

Illustrative, to show the shape finance expects — one quarter, one move per team.

TeamAI cost per outcomeRework vs. own baselineQuadrantThe budget move
Engineering$310/PR (high)−12% (low)ScaleIncrease next-quarter budget; expensive but clean, return is compounding
Support$71/case (low)+2% (low)KeepLeave it running; cheap and clean, no change needed
Sales$104/deal-touch (low)+24% (high)Fix workflowHold spend; remediate the process before scaling
Marketing$540/asset (high)+29% (high)Cut / rebuildCut the use case; rebuild the workflow from scratch
Numbers are illustrative; the structure is the point. Each team is read against its own baseline. A blended average across these four would report "AI is roughly working" and bury the two teams leaking budget — which is exactly what the matrix exists to prevent.

Make it a quarterly ritual, not a one-time audit

Allocation isn't a project; it's a cadence. Three reasons the review has to repeat every quarter, not annually:

  • Spend outruns the annual guess. At 47% year-over-year growth (Gartner, May 2026), a budget set in January is materially wrong by mid-year. The allocation has to move at the speed of the spend.
  • Rework drifts as models and workflows change. A team in the Scale quadrant this quarter can slide into Fix after a model swap or a workflow change. Only a re-run catches the drift.
  • The board keeps asking. With 66% of boards gating funding on proof of return (CloudZero, Jun 2026), the matrix isn't a one-time exhibit — it's the recurring answer to a recurring question. Run the five steps each quarter, re-plot every team, and reallocate. The ritual is what turns a static "AI budget" into a governed allocation that grows — and gets defended — with the spend.

Common mistakes when allocating an AI budget

  • One org-wide AI budget. A single blanket figure averages the compounding team with the leaking one. Allocate per team, always.
  • Confusing high cost with waste. A high-cost, low-rework team is a candidate to scale, not cut — the spend is buying clean, shipped work. Cost alone doesn't condemn a team; cost and rework do.
  • Scaling a high-rework team because it's cheap. Low cost per outcome is not permission to add spend if rework is high. Fix the workflow first; scaling just buys more rework.
  • Skipping the baseline. A cost or rework number with no pre-AI baseline is an anecdote. Every quadrant read is a comparison to the team's own history.
  • Allocating once a year. At 47% annual growth, an annual allocation is stale before the next board meeting. Re-run quarterly.

FAQ

How do you allocate an AI budget across teams? Plot each team on the Cost × Rework matrix — AI cost per outcome against rework level — and take the move the quadrant dictates: scale (high cost, low rework), keep (low cost, low rework), fix the workflow (low cost, high rework), or cut (high cost, high rework). Read each team against its own baseline, and re-run quarterly. 66% of boards now gate AI funding on exactly this kind of proof (CloudZero, 2026). What is the Cost × Rework matrix? It's a 2×2 that plots each team on AI cost (high/low) against AI rework (high/low), returning a per-team verdict — scale, keep, fix, or cut. Cost is the denominator; rework-adjusted output is the numerator. It's the CFO's decision artifact for allocating the AI budget team by team, turning "we can't tie spend to outcomes" into a move per team a board can act on in one glance. What do you do with a team that has high AI cost but low rework? Scale it, don't cut it. Low rework proves the output is sound and shipping, so a high bill is leverage, not waste — the team is buying clean, accepted outcomes. This is the "Scale" quadrant: fund it, because more spend buys more of a proven return. Cutting a team that ships good work just to save cost destroys a return you could have compounded. How often should we re-allocate the AI budget? Quarterly. Worldwide AI spending is set to grow 47% in 2026 (Gartner, May 2026), so a budget set annually is materially wrong within months. Rework also drifts as models and workflows change — a Scale-quadrant team can slide into Fix after a model swap. A quarterly re-run of the matrix catches the drift and keeps the allocation defensible as the spend grows. How is this different from just tracking AI cost? Tracking cost tells you the bill, not the return. Allocation needs the second axis — rework — because a cheap team can be net-negative if its output keeps getting redone, and an expensive team can be worth every dollar. The matrix pairs AI cost per outcome with rework level so the move is based on output net of rework, not on cost alone. Cost is the denominator, not the decision.

Sources

See the return on your own numbers.

Get a clear, per-team return-on-AI picture in weeks — output net of rework, reconciled to your invoice.