Budget season comes for every AI investment eventually. When it does, "developers say they love it" will not save the line item.

CFOs think in terms of cost versus output. If your AI tooling discussion happens entirely in developer-experience terms — velocity feels better, developers are happier, onboarding is faster — you are speaking a language that does not translate to budget decisions. Here is the framework for making the case in terms finance can act on.

What finance actually wants to know

There are three questions behind every AI budget review:

What did we spend?
What did we get?
Is (2) worth (1)?

Most engineering teams can answer question 1 precisely — the Anthropic or GitHub invoice is right there. The problem is questions 2 and 3. Without concrete output data, the only honest answer is "we think it helped" — and that is not a number that survives scrutiny.

Building the output side of the equation

Developer hours saved is the most legible output metric to finance, because it maps directly to cost. If a developer earns $150,000 per year fully loaded, that is roughly $72 per hour. If AI tooling saves each developer two hours per week on average across a team of ten, that is $72 × 2 × 10 × 52 = $748,800 in recovered time per year.

You do not need to be precise to be persuasive. A defensible estimate with explicit assumptions is worth more than a vague claim. Show your math:

Assumption	Value
Average fully-loaded developer cost	$150,000/year ($72/hr)
Estimated hours saved per developer per week	2 hours
Team size	10 developers
Annual recovered time value	$748,800
Annual AI tool spend	$36,000
ROI multiple	20.8×

Even if your assumptions are off by 50%, the ratio still justifies the spend.

The commit-based approach

Hour estimates are directionally useful but hard to verify. Commit data is not. A stronger approach roots the analysis in something auditable:

Before AI tooling: Pull six months of git history and measure average lines committed per developer per week.

After AI tooling: Measure the same metric for the six months post-adoption.

If the post-AI number is materially higher — and for most teams it is — that delta is measurable output. Pair it with AI spend data and you have a cost-per-unit-output number that holds up to scrutiny.

"We committed 38% more code per developer per week and paid $3,100 more per month in AI costs to do it" is a different conversation than "productivity seems better."

Connecting commits to shipped tasks

Lines committed is a proxy for output. Closed tasks are the real thing. The strongest ROI argument links AI spend to business deliverables:

Tasks completed per month, before and after AI adoption
AI cost per closed task (total monthly AI spend ÷ tasks shipped)
Time-to-close for tasks worked with AI assistance vs. without

If you track issues in Jira, Linear, or GitHub, most of this data already exists. The gap is connecting it to AI spend data — which requires knowing which sessions contributed to which commits, and which commits closed which tasks.

What to put in the deck

When the CFO meeting comes, the slide that works has three numbers:

What we spent — total AI tooling cost, trailing 6 months
What we shipped — tasks completed or features delivered in the same period, with a note that AI-assisted commits accounted for X% of the code
The ratio — cost per feature, or recovered developer hours vs. spend

Keep the assumptions visible. Finance will probe them. If your assumptions are conservative, that is a feature — it shows you are not inflating the case.

The failure mode to avoid

The most common mistake is waiting until the renewal conversation to build this case. At that point, you are starting from zero data and arguing backwards. The teams that keep and grow AI budgets start tracking output metrics on day one of adoption — not because they expected a budget fight, but because they knew one might come.

Tazmin is built to give engineering leaders exactly this data: AI spend, committed code, and closed tasks in one place so the ROI case builds itself. Join the waitlist to get early access.