AI Search

SEO ROI: How to Build a Case Your Finance Team Will Not Send Back

Stefka Team October 26, 2026 6 min read

SEO ROI is incremental revenue from organic search, minus the cost of the work, over a period — and every contested word in that sentence does more damage than the arithmetic. "Incremental" and "attributable" decide the answer before any number is entered. Agreeing what they mean, in advance and in writing, is the whole job.

There is a straightforward piece of evidence that the market does not see it that way. When we pulled this family on 26 October 2026, seo roi calculator drew 170 searches a month at competition index 2 — while seo payback period, saas seo roi, seo roi benchmark and how to prove seo value returned no measured search demand at all. People want a tool that produces a number. Almost nobody is looking for a way to think about it.

Why the calculators do not work

An SEO ROI calculator multiplies four things: keyword search volume, an assumed click-through rate for the position you hope to reach, an assumed conversion rate, and an assumed deal value.

Each of those is uncertain, and uncertainty compounds. If each is within a factor of two of the truth — generous, in most cases — the output is within a factor of sixteen. That is not a forecast; it is a range wide enough to contain both "obviously do this" and "obviously do not", presented as a single figure with a currency symbol in front of it.

There is a second problem specific to 2026, and it is getting worse. The click-through-rate curves these tools use are mostly drawn from studies predating AI Overviews holding absolute position 1. On both of the queries we pulled for this batch, the AI answer sat above every organic result and ended by asking the reader a question back. Whatever position 1 is worth now, it is not what the curve in the calculator says.

Calculators are useful for one thing: sizing. If the answer comes back at €5,000 of annual value, stop. If it comes back at €5 million, the arithmetic is wrong. Anything in between tells you nothing you did not already know.

Build it from the five things that are actually knowable

1. The cost, with the full number. Not just the fee. The fee, plus the internal time, plus the period before you can judge it. What it costs covers the first; how long it takes covers the third, and it is usually the larger of the two.
2. The queries you can realistically reach. Not category volume. The terms where a page you could actually write could plausibly land on the first two pages inside the period. This is where most forecasts break: they are built on the head term, and the head term is not available to you.
   
   A worked example from this site's own programme, because we would rather show our own than borrow someone's. The commercial term ai seo agency carries 1,600 US searches a month at competition index 7 — the most attractive number in our own research. We do not rank for it. What we rank for, after six weeks of weekly publishing, are terms like best aeo agency for early-stage saas companies (40 a month, position 38) and partner agencies for white-label collaboration europe (70 a month, position 5). Twenty-five ranking keywords and five monthly US organic visitors in total. A forecast built on 1,600 would have been wrong by three orders of magnitude. A forecast built on reachable long-tail terms would have been roughly right, and would have looked far less exciting in the slide.
3. Your own conversion rate, not an industry one. You have this number. Use it, and if organic converts differently from paid at your company, say so and use the organic one.
4. The one number this will be judged on. Agreed before the money moves, countable within a quarter, and countable by someone who was not involved in producing it. Pipeline created, qualified conversations, activations. Not impressions, not rankings, not a score out of 100.
5. The condition under which you would stop. Write it down. "If we have not seen X by month six, we stop and do Y." A business case without one reads as advocacy, and finance teams have learned to read it that way.

Present it as a range, and say what it is sensitive to

Give three cases, not one. A conservative case built on terms you already have some position on. A central case. An optimistic case you do not expect.

Then add the sentence that makes the whole document credible: name the single assumption the range is most sensitive to, and say what would make you revise it. Usually it is the conversion rate or the achievable position. Volunteering it costs nothing, and it is the clearest available signal that the forecast was built rather than reverse-engineered from a conclusion.

What about AI search, where there are no benchmarks at all?

The honest answer is that nobody can forecast it, and pretending otherwise is the main way money gets wasted in this category right now.

We checked. aeo roi, ai visibility results, ai search visibility kpi and how to measure ai visibility all return no measured search demand. There is no body of published experience to draw a benchmark from, because the question is not being asked in public.

What you can do instead is measure the mechanism rather than forecast the outcome. Citation share on a fixed prompt set, measured weekly, is a real number with a real baseline, and it moves or it does not. Tracking your brand in AI search sets out how. Build the case on the search side where numbers exist, treat AI visibility as measured upside rather than forecast upside, and say which is which in the document.

We hold ourselves to that: our own AI mention count has been one for twelve consecutive weekly readings, and the single mention is about a given name rather than this studio. We report it as zero and publish it anyway.

The shape of a one-page case

Six lines, which is usually enough.

1. What we would spend, over what period, including internal time.
2. What we expect to be able to see, and when — split into what moves in weeks and what moves in quarters.
3. The one number we will be judged on.
4. Three cases, with the assumptions stated as assumptions.
5. The assumption it is most sensitive to.
6. What would make us stop.

If it runs longer than a page, the extra pages are usually defending the forecast, which is the part nobody believes anyway.

For the wider picture this sits inside, what to budget overall is the page above this one; what an audit costs is the one-off line that often precedes the retainer.

Where Stefka fits

We publish our price — engagements start at €5,000 a month — and our own results, including the unflattering ones: twenty-five ranking US keywords, five monthly US organic visitors, zero genuine AI mentions. We have run this site's own search programme weekly since September 2026 and published every number it produced, including the weeks nothing moved.

That is not modesty. It is the only honest basis on which anyone in this category can tell you what to expect, and it is why the worked example above is ours rather than a client's.

If you are building this case internally and want someone to argue with the assumptions before finance does, talk it through.

Frequently Asked Questions