Every strategy deck needs a market size, and the TAM / SAM / SOM framework is how you make it defensible. Here is the method in practical steps: what each level means, how to build the numbers, and how to avoid the trap of a precise-looking guess. The core message is simple: your market size is only as good as the assumptions under it, so this guide shows you how to make those assumptions visible, testable, and honest.

What TAM, SAM and SOM actually mean

TAM: total addressable market: the full revenue opportunity if everyone who could buy, did. It is the ceiling, usually built from top-down data about the whole industry.
SAM: serviceable available market: the slice you can realistically serve, cut by geography, segment, or constraints such as language, regulation, or distribution.
SOM: serviceable obtainable market: the share you can actually capture in the near term, typically three to five years, and driven by your go-to-market reach and budget.

Most sizing arguments are really two people measuring different levels and not noticing. One person quotes a TAM of $2 billion and the other points out their revenue is $5 million, and both walk away confused. The fix is to label every number with its level in the heading and in the slide title, so the disagreement surfaces early instead of at the review.

A useful sanity check for the three levels: TAM should be much bigger than SAM, and SAM should be several times bigger than SOM. If your SOM is close to your SAM, you are either planning to capture nearly the entire addressable slice in a few years, which is almost never true, or your definition of SAM is too narrow.

Step 1: Define the unit first

Is the market "spend on X per year" or "companies that do Y"? Pick one unit and hold it across all three levels. Switching units mid-way is where numbers stop being comparable.

You have three workable choices:

Revenue units: dollars of spend per year on the product or service category.
Volume units: number of transactions, units sold, or seats, converted to revenue later via price.
Population units: number of customers, companies, or locations that could buy.

The best choice depends on your data. If you have reliable per-customer spend figures, use revenue units. If you have census-style counts of companies, use population units and multiply by an average annual spend. Do not mix them: a TAM in dollars and a SAM in companies cannot be compared.

Step 2: Segment before you build

Segment by two or three variables: geography, industry, and customer size is usually enough. Every added dimension multiplies the cells you need to fill, so keep it tight.

Worked example: a payroll software vendor in the UK might segment by country, then by employee band (1 to 9, 10 to 49, 50 to 249), then by industry only where it matters, such as construction versus professional services. That yields a manageable number of cells, each with a distinct buyer and a distinct price point. If you find yourself with more than 15 cells, collapse the dimensions that do not change the price or the buyer.

Step 3: Build a driver model

Price per unit multiplied by volume per customer multiplied by customer count. Every input needs a source or a written assumption. The discipline is what makes the number defensible: you can hide a guess inside a formula, but not when every input is labelled.

For a population-based estimate the formula is: number of eligible customers, times the share that would buy in a year, times average spend per customer per year. For a spend-based estimate it is: total category spend, times the share in your geography, times the share in your segment.

Worked example for the payroll vendor:

1.2 million UK businesses have 1 to 9 employees (published government statistics).
45% use some form of payroll software today, and 15% outsource payroll entirely, leaving 40% as a serviceable target: 480,000 businesses.
Average software and support spend per small business is GBP 40 per month, or GBP 480 per year.
That gives a SAM of 480,000 times 480, roughly GBP 230 million per year.

Every one of those inputs is now on the page with its source, and a reviewer can attack any single one without collapsing the whole model. That is the point of the exercise.

Step 4: Show the range, not a point

A point estimate is a lie. A band from pessimistic to optimistic is a consultable answer. Clients trust the thinking behind a range, because it tells them what would have to be true for each end of it.

Build the range by varying the two or three inputs that matter most, usually the adoption share and the average spend. Keep everything else fixed. In the example above, if adoption sits between 35% and 50% and average spend between GBP 420 and GBP 540 per year, the SAM swings from roughly GBP 176 million to roughly GBP 324 million, with a midpoint near GBP 250 million.

Present it as "in the range of GBP 180 million to GBP 320 million, centered around GBP 250 million". This is more honest and more useful than a single number, because the client can stress test the assumptions that drive each end. For investors and boards, a range with identified drivers is a much stronger signal of analyst quality than a false precision.

Step 5: Write the assumption next to every number

"I assumed 40% of SMEs outsource this" is worth more to a client than a third-decimal precision that came from nowhere. It is also what survives the review: when someone asks where a number came from, the answer is already on the page.

Two rules keep the assumption list honest. First, mark each assumption as anchored or judgment. Anchored means it has a source: a government statistic, an industry report, your own sales data. Judgment means it is an estimate you are choosing. There is nothing wrong with judgment assumptions, but they are the ones a reviewer will probe, so flag them. Second, add a sensitivity note: for the two or three judgment assumptions, write what would happen if they moved up or down by 25%. If a 25% swing in one assumption changes your answer by 60%, that assumption is the one to go research next.

Where the numbers come from

A practical sourcing order, cheapest first:

Your own data: invoice history, CRM records, pipeline win rates. Most defensible, and free.
Public statistics: government censuses, tax agency business counts, industry association surveys.
Industry reports: paid reports for the anchor numbers, used carefully for the segment you actually serve.
Interviews: five to ten conversations with buyers and resellers to test price points and adoption shares.

In a two week engagement, plan the split roughly as: three days on your own data, four days on public sources, two days on interviews, and the rest on building and stress testing the model. That split keeps the model anchored in evidence instead of speculation.

Common mistakes to avoid

Unlabelled levels: quoting a TAM as "the market size" so the client overestimates your reach. Always label TAM, SAM, and SOM.
Mixed units: dollars at one level, customer counts at another. Hold one unit across all three.
A single point estimate: implying precision that no model has. Give a range with drivers.
Hidden guesses: inputs without sources or assumptions. If you cannot label it, you cannot defend it.
Top-down only: a TAM built entirely from an industry report with no bottom-up check against your own data.
Ignoring the buyer count: a market of 5 million companies with 2 buyers per category is smaller than it looks. Check who actually signs.
Confusing served and obtainable: treating your SAM as your plan. SOM is the part you actually pursue.

Speed it up

The research is the slow part; the structure should not be. Generate a free market sizing estimate and spend your time on the assumptions that actually move the answer. The tool builds the TAM / SAM / SOM skeleton in minutes, so you can focus your effort on the few inputs that matter.

How to size a market: TAM, SAM and SOM