Top-Down TAM SEO Forecast
Forecast SEO from market size down, not from keyword-tool volumes up.
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 92%
Bottoms-up SEO forecasting — pull search volume from a keyword tool, guess a rank, guess a CTR, guess a conversion rate, then arbitrarily gross it up — produces a number that is both too small to fund and built on data that can be wrong by an order of magnitude. Schwartz replaces it with a TAM-style top-down forecast that starts from the addressable population and cuts down to buyers, then applies a market-penetration assumption you can revisit and defend.
Origin
Developed by Eli Schwartz while at Faire (the wholesale marketplace) to size the SEO upside of launching in a new country, after bottoms-up keyword forecasts kept producing numbers too small to win funding.
Core principles
- 01Keyword tools are estimates from proprietary algorithms, not truth — Google itself cannot publish real volumes.
- 02A forecast built on a wrong first number is wrong all the way down.
- 03Every assumption in a top-down model is nameable and adjustable; a keyword-tool number is not.
- 04You will never reach the truth — you are trying to make a better decision, not a perfect one.
- 05Google Search Console is real data; third-party tools are indicative only.
How to run it
- 1
Start with the addressable population
Take the total population of the market you are entering. For a shoe business launching in Japan, start at (say) 100 million people.
Pro tip Name the source of each number so a reviewer can challenge it later.
- 2
Cut down to plausible buyers
Apply successive demographic and behavioural filters — gender if the product is gendered, age brackets, then the share that buys this category online rather than in a store. Schwartz's worked example: 100M → 50M men → 25M in the right age band → 10% who buy shoes online → 2.5M.
Pro tip Keep each cut as a single explicit percentage so you can sensitivity-test it.
Watch out Do not hide multiple assumptions inside one blended factor — you lose the ability to correct the specific one that was wrong.
- 3
Apply a market-penetration target and monetize it
Pick the share of that buyer pool you intend to capture (10%, 50%, 100%), multiply by purchase frequency per year and AOV. That is your SEO forecast.
Pro tip State the penetration target as a decision, not a prediction — it becomes the goal the SEO product is built to hit.
- 4
Revisit and adjust the named assumptions
At any point you can go back and correct a specific input — AOV was off, penetration was optimistic, online-purchase rate was wrong for this market — and the forecast updates coherently.
Pro tip Validate against Google Search Console for the queries you already own, rather than against a third-party tool.
Watch out Third-party tools have been observed over- and under-estimating real volumes by 10x. Use them for normalization (how people phrase things) and relative comparison, never as a source of truth.
In the wild
Working with WordPress, Schwartz found that 'WordPress' — the largest query in the entire web development space — had a volume in every keyword research tool that was completely wrong versus what Google Search Console actually reported.
→ Proof that a bottoms-up forecast anchored on the single biggest keyword in a category can be wrong from its first line.
Early in COVID, a public company's board member emailed the CMO saying they were being crushed by competitors, based on a third-party traffic tool. Schwartz checked Google Search Console: the company's organic traffic had actually quadrupled during COVID.
→ The board member's conclusion was inverted. Search Console — imperfect but real — overruled the estimating tool.
Common mistakes
Grossing up the bottoms-up number by an arbitrary multiple
Teams estimate 'shoes' volume, know it misses 'white shoes' and 'running shoes', then multiply by 10 because it feels right. The result is a guess dressed as a model — and usually too small to get funded.
Treating semrush/Ahrefs/Similarweb numbers as truth
They all run different proprietary estimation algorithms, which is why they disagree with each other. They are useful for how people phrase queries and for relative comparison, not for absolute planning numbers.
Is it for you?
Best for
A PM or growth lead who must justify an SEO investment or a new-market launch to a board or exec team.
Not ideal for
Small tactical decisions about which of two existing pages to optimize — the top-down model is too coarse for that.
From the transcript
“the way most people do SEO forecasting is they do a Bottoms Up forecast which is they look at keywords”
“we'll take 10% of our 25 million people in our Market two and a half million people buy shoes on the internet”
“they're helpful tools are helpful but I don't think they are a source of actual truth”
From the episode
Rethinking SEO in the age of AI
Eli Schwartz (SEO advisor, author)