Methodology · v1 · published May 2026

How we estimate
your AI ROI exposure

The AI ROI Exposure Assessment returns a one-page report with three numbers a CFO will want to defend: estimated annual AI spend, the share of that spend with a provable dollar return today, and the three moves most likely to close the gap. This page shows the math behind each one — what we assume, what we don't, and how to pressure-test it against your own data.

← All published methodologies

Our principle

Show the work. Don't assert a number you can't replicate.

The estimates in your report are directional — they're useful for framing a conversation, not for booking a number to a budget. Where the answer depends on data we don't have (your industry mix, your contract structure, what counts as "AI" in your stack), we say so and use a range. Where we use a heuristic, the heuristic is on this page in plain English, with the inputs we plugged in for your specific report. If a step doesn't survive your CFO's scrutiny, we'd rather you find out from this page than from your report.

We deliberately do not cite specific research reports for benchmark percentages we cannot independently verify. Where ranges come from public industry surveys, we say "consistent with publicly reported industry surveys" — not "according to Gartner / IDC / McKinsey." A real CFO knows what those reports actually say; a fabricated citation is worse than no citation.

Step 1

Estimating your annual AI spend

Two questions in the assessment feed this number: annual revenue and the AI tools you said you're already paying for. The math is the same for everyone, transparent on this page, and produces a range — never a point estimate.

The formula

estimated_annual_ai_spend = revenue_midpoint × pct_band(low: 0.3%, high: 1.2%)
× stack_breadth_multiplier(0.85 – 1.25)

Where the 0.3% – 1.2% band comes from

Aggregate enterprise AI spend (direct AI provider contracts plus bundled AI capabilities inside existing SaaS) sits roughly in this band for mid-market companies in 2024–25, consistent with publicly reported industry surveys and what we observe across the design partners we've talked to. The band is wide on purpose: a company spending heavily on a few enterprise contracts (Salesforce Einstein, Microsoft Copilot) lands near the top; a company experimenting with seat-licensed tools lands near the bottom. We do not narrow this band without a specific input that justifies narrowing it.

The stack-breadth multiplier

The "paid AI tools" question is a multi-select. We use the count as a rough breadth signal:

1–2 tools selected: multiplier 0.85 (you've concentrated the spend; lean low)
3–5 tools selected: multiplier 1.0 (typical mid-market footprint)
6+ tools selected: multiplier 1.25 (broader stack, more bundled AI to account for)
"Don't know" selected: we don't apply the multiplier and widen the underlying band to 0.2% – 1.5% to reflect the lower confidence

Worked example

A $500M-revenue company that selected OpenAI, Microsoft Copilot, Salesforce Einstein, and GitHub Copilot:

revenue midpoint (from the "$250M – $1B" bracket) = $625M
0.3% – 1.2% band → $1.9M – $7.5M
4 tools selected → multiplier 1.0
reported range: $1.9M – $7.5M annual AI spend exposure

What this number is not

It is not your actual AI spend — that lives in your AP system and your finance team can pull it.
It does not separate CapEx from OpEx, direct vendor contracts from bundled SaaS features, or production deployments from pilots.
It assumes "AI spend" includes both standalone AI vendors and AI features embedded inside existing SaaS subscriptions. A CFO who excludes embedded AI will land below our low end.
It does not account for hidden AI spend on company cards (Cursor, ChatGPT Team, etc.) that bypasses procurement.

Step 2

How much of that spend is provably measured today?

The "exposure" half of the report is the spend that doesn't have a defensible dollar return attached. We derive this from two questions: who owns AI ROI in your company, and how it's measured today.

The maturity ladder

We assume each rung captures a typical share of dollars under formal measurement. These percentages are stated assumptions, not measured facts:

How AI ROI is measured today	Assumed share of spend with provable ROI
Versioned formulas tied to integration data	~80%
Periodic spreadsheets owned by finance	~30%
Anecdotes in board decks	~10%
We don't measure it yet	0%

Adjustment for ownership

If AI ROI is owned by a CFO, Finance team, or Innovation team, we add 5 percentage points to the rung above — the assumption being that a named owner correlates with at least some formalised reporting, even if the methodology hasn't caught up yet. If "Each function owns their own" is selected, no adjustment is applied. If "Nobody yet" is selected, we cap the measured share at 5% regardless of the measurement answer.

The exposure number

ai_roi_exposure = estimated_annual_ai_spend × (1 − measured_share)

So for the $500M company above, with "Periodic spreadsheets owned by finance" and "The CFO" as the owner:

measured_share = 30% + 5pp = 35%
exposure = $1.9M – $7.5M × (1 − 0.35) = $1.2M – $4.9M
reported exposure: $1.2M – $4.9M of AI spend without a defensible dollar return today

Step 3

The three highest-leverage moves

The bottom third of your report names three specific changes most likely to close the exposure gap. These come from Roiva's library of initiative templates, ranked for your specific company by:

Function match. We surface templates aligned to the functions you said are running or planning AI work in question 11.
Estimated impact for your company size. Each template carries an impact band (low / medium / high). For larger companies we weight high-impact templates more heavily; for smaller companies we weight effort/time-to-deploy more heavily.
Deduplication against your existing stack. If you said you already use a tool that would deploy a given template (e.g., Zendesk AI for support routing), we down-rank that template — you've already paid for the capability and the ROI work is measurement, not deployment.

The three returned are the top of that ranked list. You will see different three depending on which functions you selected, which AI tools you said you already pay for, and your company size band.

Important

What this report is not

We say this on the report itself, but it's worth saying again here. The exposure report is:

Not an audit

We don't see your actual contracts, GL accounts, or finance reports. The estimate is built from five answers on a five-minute form.

Not predictive

It does not forecast next year's AI spend or the ROI a specific initiative will return. The number is directional, anchored in your current portfolio.

Not industry-specific

The same band applies regardless of whether you're in financial services, healthcare, or manufacturing. Industry-specific calibration is planned but not in v1 of the methodology.

Not a substitute for your real numbers

If your finance team can pull actual AI spend from QBO / NetSuite / Coupa, that number beats ours every time. The report exists to frame the conversation, not to replace the data.

Pressure-test it

How to validate the number with your CFO

If you're taking this report into a finance conversation, here are the three questions a CFO will ask first — and how to answer each one.

It's a directional band consistent with publicly reported enterprise AI adoption surveys (Stanford AI Index, S&P Global Market Intelligence, McKinsey State of AI), cross-referenced against the AP spend patterns we've seen in mid-market design-partner conversations. We deliberately do not cite a single source because no single source covers the full picture of direct + bundled AI spend, and any specific report's number drifts within a quarter. If your CFO wants the source for a specific number, the answer is: this is a heuristic, not a citation — validate it against your own AP.

Then your real number is the number — ours was the wrong frame for your company. We'd want to know about it: it tells us our band is too wide, too narrow, or biased high/low for your industry, and we use that to retune the methodology. Drop a note to luke@roiva.ai with the actual percentage and we'll fold it in.

The 10% reflects the typical observation that even in anecdote-heavy environments, the most-cited 1–2 initiatives usually have at least one number attached — a deflection rate, a time-saved estimate, a cost-per-ticket savings — even if that number isn't audited. If your environment is pure narrative with no quantified anecdotes, treat that rung as closer to 0% than 10%.

Versioning

How we update this

The methodology version (v1) is printed at the top of this page and on the footer of every exposure report PDF. When we change the bands, the multipliers, or the maturity ladder, we bump the version and date so any historical PDF can be matched back to the methodology that produced it. We do not silently retune; the change log lives here when there is one to log.

v1 — published May 2026

Initial published methodology covering spend exposure, measurement maturity, and template ranking.
Spend band: 0.3% – 1.2% of revenue; "Don't know" widens to 0.2% – 1.5%.
Stack-breadth multiplier: 0.85 / 1.0 / 1.25 for 1–2 / 3–5 / 6+ tools.
Measurement maturity: 80% / 30% / 10% / 0% with +5pp ownership adjustment, capped at 5% for "Nobody yet" owner.

Now run yours

Five minutes, no account, no card. The math on this page is the same math that produces your report.

Get your AI ROI exposure report →

How we estimate your AI ROI exposure