Evaluate AI search investment with agreed costs, qualified commercial outcomes and a time window that matches the sales cycle. Use answer visibility to identify information gaps, not as a substitute for revenue. Before outcomes mature, report delivery, inquiry quality and opportunity progression as leading evidence. For a budget decision, calculate a clearly labeled break-even scenario using gross profit and plausible contribution assumptions. When reporting actual return, preserve unknown and mixed-source journeys and distinguish attributed results from evidence that the program caused additional business.
Separate the measurement layers
| Layer | What to measure | What it does not prove |
|---|---|---|
| Answer visibility | Mentions, citations and recommendations in an agreed sample | Total user exposure or revenue |
| Site acquisition | Identifiable visits and landing pages | A qualified buyer or incremental demand |
| Inquiry quality | Fit against a defined target customer | Sales acceptance or a signed contract |
| Opportunity | Accepted sales process with a next step | Closed revenue |
| Commercial result | Recorded revenue or gross profit under an agreed definition | Causation from AI search alone |
Maintain separate denominators and time windows. A monthly observation report and a long sales cycle cannot be compared as if every outcome happened in the same period.
Agree the business definitions
A qualified inquiry should match the offer, target market and basic commercial criteria chosen by the sales team. Record how leads become opportunities and what counts as won business. Keep disqualified and unknown outcomes visible.
Cost should include the items required by your analysis: service fees, approved third-party spending and any internal implementation costs you choose to include. A positive revenue total does not necessarily mean a positive return after delivery costs.
Use formulas with honest inputs
Illustrative calculation, not a TANTU AI client result: suppose a business records $30,000 of gross profit from opportunities attributed under its chosen method and spends $20,000 on the evaluated program. The arithmetic return is ($30,000 − $20,000) ÷ $20,000 = 50%.
This calculation is only as credible as the profit, cost and attribution inputs. It does not prove that the program caused the entire $30,000. If the only available number is sampled brand mentions, the ROI calculation cannot be completed.
Keep the unknown part visible
A buyer may research with AI, search for the brand later and submit through a direct visit. Another may arrive through a source link that can be identified. Use available referral data, landing pages and optional self-reported discovery information, while keeping privacy and data quality in mind.
Do not force every sale into an AI category. Record unknown and mixed-source journeys. Compare alternative explanations such as pricing changes, sales activity, seasonality, promotions and changes to the product itself.
Run a monthly business review
- Check that the same question set and business definitions are being used.
- Review the pages and sources that changed during the period.
- Examine inquiry quality and the reasons for sales rejection.
- Track progress of accepted opportunities over a realistic sales cycle.
- Choose the next work based on an observed problem, not a single score.
The purpose is to improve decisions. A report that admits attribution limits but identifies a useful action is more valuable than a precise-looking revenue claim without evidence.
Use break-even arithmetic as a planning test
Start with the total cost included in the evaluation and the expected gross profit per won customer under your own finance definition. Divide cost by that gross profit to estimate the number of additional wins needed to cover it. State which costs and margins are included; otherwise two apparently similar calculations may answer different questions. Round required customer counts upward when whole customers are the unit.
Hypothetical planning example: a $20,000 investment and $5,000 gross profit per additional customer require four additional wins to break even. If only half of a customer’s gross profit is assumed to be incremental to the program, the scenario requires eight wins. Neither assumption predicts results. The exercise tests whether the required commercial outcome is plausible before you treat visibility movement as an investment case.
Follow inquiry cohorts through the sales cycle
Group inquiries by their first recorded period and track qualification, accepted opportunity and outcome without resetting their history each month. Keep open opportunities separate from lost and won business. A long sales cycle means this month’s closed revenue may relate to much earlier discovery, while recent inquiries have not had time to mature. Explain this lag rather than forcing all activity into one monthly ratio.
When evidence is incomplete, present a range of attribution assumptions alongside the observed commercial totals. Compare those assumptions with known campaign, sales and product changes. If qualified inquiries rise but accepted opportunities do not, investigate fit and handoff before increasing acquisition spend. A financial review should identify the next decision and the unresolved attribution question, not manufacture a precise return from an immature pipeline.
Common questions
Is an AI mention an impression?
Not in this reporting framework. A sampled answer records one observation; it does not reveal how many real people saw similar content.
Can we report ROI before deals close?
You can report costs and pipeline progress, clearly labeled. Projected pipeline value is not realized return.