How to forecast what an SEO change is worth in pounds
Revenue forecasting · 6 min read · Updated
To forecast what an SEO change is worth, multiply four numbers and subtract one: the keyword's monthly searches, the extra share of clicks you expect from the position you are moving to, your site's conversion rate for that kind of page, and the value of a conversion — then take off the cost of making the change. The output is a monthly revenue figure with a stated set of assumptions attached. It is a projection, not a promise, and the only way it stays useful is if you record it and compare it with what actually happened.
The five numbers, and where each one comes from
Most forecasting arguments are really arguments about one of these five inputs. Getting them from the right place, and being clear about which are measured and which are assumed, settles almost all of them.
| Input | Where it should come from | How wrong it usually is |
|---|---|---|
| Monthly searches | Google Ads, if you have an account with spend. Otherwise your own Search Console impressions for that query. | Very. Third-party volume figures are modelled, rounded into buckets, and often out by a factor of two either way. |
| Click-through uplift | A published position-to-CTR curve, adjusted down for queries with an AI Overview, ads or a local pack. | Moderately. The curve is an average across every query type, and yours is not average. |
| Conversion rate | Your own analytics, for that page or that page type — never a site-wide average. | Usually optimistic, because the site-wide figure includes your brand traffic. |
| Value of a conversion | Your order value, or lead value multiplied by your lead-to-sale rate. From your own books. | Least of the five, if you use the real figure rather than the one on the website. |
| Cost of the change | Hours, at your real rate, plus any tooling or content cost. | Routinely forgotten entirely, which is how unprofitable work gets prioritised. |
If a forecast does not name all five, it is not a forecast — it is a number that sounds like one.
A worked example
A plumbing company ranks eleventh for "boiler replacement Leeds". Search Console reports 880 impressions a month for the query and 9 clicks, which is about a 1% click-through rate — consistent with a position just off the first page.
Moving to position five would, on a conservative curve, take that to roughly 5%. That is 44 clicks a month against the current 9, so 35 additional visits. The company's quote-request rate on service pages is 2.4%, giving 0.84 extra enquiries a month. They win one job in four, and an average boiler replacement is £2,400 — so a won job is worth £2,400 and an enquiry is worth £600.
0.84 enquiries × £600 = roughly £504 a month, or £6,048 a year. The change needed is a rewritten service page with proper local signals and three internal links, which is most of a day: call it £400 of someone's time. The forecast is therefore about £500 a month against a £400 one-off cost, paying back inside a month.
Notice what the arithmetic did not do: it did not assume position one, it did not use a site-wide conversion rate, and it did not quietly leave out the cost. Each of those three shortcuts roughly doubles the answer, and all three together produce the £4,000-a-month version of the same change that nobody ever achieves.
What to do with the things you cannot price
Plenty of worthwhile work resists this arithmetic. Fixing a crawl error on a page with no rankings yet, adding schema, improving a page's readability for answer engines — none of these has a keyword and a curve behind it.
The temptation is to invent a figure so the item can be ranked alongside everything else. Resist it, for a reason that is practical rather than moral: once one made-up number is in the list, the list stops being sortable by money, and you lose the only thing the exercise bought you.
Price it at zero and rank it separately, on effort and expected upside. A zero is not a claim that the work is worthless — it is a statement that its value is not measurable in this model, which is true and which everyone reading the list can verify.
The honesty rules that keep a forecast useful
- Label every figure a projection, every time it appears. A number that travels into a slide deck without its label comes back as a commitment.
- State the assumptions next to the number, not in an appendix. Position target, CTR curve, conversion rate, conversion value.
- Record the forecast at the moment you make the change, not afterwards. A forecast written after the result is a description.
- Compare it with the outcome and publish the comparison, including the ones you got wrong. This is the only thing that makes the next forecast trustworthy.
- Let the comparison change the model. If your forecasts for metadata changes run 40% high across thirty measured changes, the model should learn that for your site specifically.
That last point is where forecasting stops being a spreadsheet exercise. Confidence should be earned per site and per change type from measured outcomes, not inherited from an industry average that was never about your business.
Doing this continuously rather than once
The method above takes ten minutes per opportunity. The problem is that a site of any size has hundreds of them, they change weekly, and the ranking that justified last month's priority list has moved.
01Collect
Pull rankings, impressions, conversion rates and page data continuously rather than when someone remembers to run a report.
02Price
Apply the model above to every finding, using the site's own revenue inputs rather than defaults.
03Order
Sort by expected profit, adjusted for effort and for how confident the model has earned the right to be.
04Act
Do the top of the list. Snapshot before every change so any of it can be put back.
05Measure
Compare the outcome against the forecast recorded at apply time, and feed the difference back into the confidence for that change type.
SEOGrowPilot runs exactly this loop. It is worth saying plainly that the method is not proprietary — it is arithmetic, and it is written out above so you can do it by hand. What software buys you is doing it for every page every day, and never quietly rounding a forecast up.