What is serviceable addressable market (SAM)?
Serviceable addressable market is the portion of your total addressable market that your product, your geography and your permissions allow you to sell to. It is calculated by applying your real constraints to your TAM, then multiplying the surviving company count by average annual revenue per customer.
TAM says how big the opportunity could be. SAM says how much of it you can reach from where you are standing, with what you have built. It is almost always far smaller, and the gap between the two is where go-to-market plans quietly come apart.
SAM is the only one of the three you control
Most teams treat SAM as a waypoint. They calculate a TAM for the investors, a SOM for the board, and pass through SAM on the way as a bit of arithmetic that makes the first number look responsible.
That gets it backwards. TAM is a property of the world, and you cannot change it by wanting to. SOM is largely a property of your capacity this year. SAM is the one number in the set that moves when you make a decision.
Every constraint you apply is a choice: which segments to serve, which markets to enter, which compliance regimes to take on, which problems to solve next. Apply them and you get a smaller number. The smaller number is not the output. The choices are the output, and the number is the receipt.
Which is why a SAM with no record of what was excluded, and why, is close to worthless. Anyone can produce a smaller figure than their TAM. The value sits entirely in being able to say what you deliberately gave up.
SAM against TAM
| TAM | SAM | |
|---|---|---|
| Question it answers | How big could this get? | How much can we sell to now? |
| Constraints applied | None | Product fit, geography, regulation, reachability |
| Used for | Fundraising, opportunity comparison | Territory planning, headcount, segment choice |
| Changes when | You build new products or enter new markets | Any of the above, plus your own decisions |
| Who controls it | Nobody | You |
If you have documented your ideal customer profile, you have done most of the thinking SAM requires. SAM is your ICP, counted.
The four constraints that shrink TAM into SAM
Product fit
The obvious one, and the one applied too loosely. Your product does not serve everyone in your industry category. It serves companies with a particular problem, at a particular size, with a particular existing setup.
That last one is worth pulling out, because for a great many B2B products it is the whole of product fit. You integrate with one system. You replace another. A third has to be present before you are any use at all. Fit is decided by what a company already runs, and stated that way it stops being a description and becomes a count.
The difference is not subtle. “Companies that would benefit from us” cannot be sized, so it gets estimated, and the estimate is generous because nothing pushes back on it. “Companies running the product we replace” can be sized exactly, and the number that comes back is usually smaller and always arguable. Prefer the second, and write down which systems define your fit before you count anything.
Geography
Where you can sell, support and invoice, which is often narrower than where your website loads. Time zones, languages, local entities, data residency and the location of your support team all bite here.
Regulation and permission
Part of your market is legally closed to you, and no amount of selling opens it. A sales lead at a wholesale fibre network described a ceiling written into his company’s own constitution: permitted to sell wholesale, forbidden from selling retail. Half his apparent market was never his to win.
Compliance cuts the other way too. The Privacy Act 1988 in Australia, the Spam Act 2003, the GDPR in Europe, the TCPA in the US and Canada’s National Do Not Call List all shape who you can contact and how. Our guide to data privacy and compliance in B2B sales covers the detail. This is general information and not legal advice, and your obligations depend on your jurisdiction and your data.
The attributes you cannot filter on
This is the constraint teams never plan for, and in 2026 it has quietly become the binding one.
You define a segment by a characteristic, then discover that no source holds that characteristic as data. A commercial lead at an automotive marketplace wanted dealerships whose stock was mostly priced under $20,000. Commercially sharp, and unbuildable. The attribute could not be supplied:
“We can, but we can’t give you that as a data set, if that makes sense.”
The information sat on public websites. Turning it into a filterable field raised privacy and permission problems that made it unusable.
His segment definition was sound. It was also unqueryable, and those are different failures with the same consequence. A segment your systems cannot express does not exist operationally, however well you can describe it in a meeting.
This is a genuinely new constraint. When market sizing was done by hand, on a spreadsheet, any characteristic a human could recognize was usable, and the limit was how many companies one person could research. Now that segmentation runs through filters, saved searches and increasingly through an agent calling an API, the limit has moved. Your segment has to be machine-expressible, or your team cannot act on it at scale.
Two options when you hit this. Find a proxy attribute that is held as data, or change the segment. Proceeding on a definition you cannot query is not a third option, though plenty of plans attempt it.
He took the second. The team moved to an adjacent vehicle category where the defining attribute did exist as data, and a ten minute working session produced a national list of about 270 dealers, which became the basis of a trial. The original segment was never built. The adjacent one was, and it was the one that moved.
One hopeful note, because this constraint moves. The set of filterable attributes is not fixed, and things that were guesswork a few years ago are now readable. Technology usage is the clearest case: it used to be inferred from job ads and whatever a company chose to publish on its own site, which made “companies running X” an opinion. Detection now runs against infrastructure signals with a confidence score and an observation date attached, which makes the same thing a filter. Tech Intelligence covers how that detection works and what it can see.
So a segment that was undefinable when you last looked may be queryable now. Recheck the constraint before designing around it, because the answer has a shelf life.
How to calculate SAM
One. Start from your TAM. You need the outer bound before cutting it down. If you have not built one, start with how to calculate total addressable market.
Two. Apply each constraint separately, recording the count after each. The sequence matters more than the total, because it shows which constraint costs you the most market. That is the finding worth taking to a product conversation.
Three. Check what the segment contains. Do not trust the label. Open the list.
Four. Multiply by average annual revenue per customer. Use your own billing data, not list price.
Five. Write down what you excluded and why. In six months somebody will ask. Usually it is you.
A worked example
Continuing the professional services automation platform from the TAM article.
| Step | Definition | Companies |
|---|---|---|
| TAM | US companies in information technology and services | 786,476 |
| Product fit | Narrowed to IT consulting and support services | 326,164 |
| Geography | Headquartered in California | 48,763 |
All three counts come from Firmable searches run on 19 September 2026, and any account can rerun them.
At $12,000 average contract value, that California SAM is worth roughly $585 million a year. Set beside a $3.9 billion TAM it makes the point better than any argument: the market you can sell to this year is about 15% of the market that exists.
Cross-check it the same way you cross-checked the TAM. For California, the BLS QCEW series counts 35,645 private establishments in NAICS 5415 in the first quarter of 2025. Our 48,763 is 1.37 times that.
Which is worth a second look, because the national comparison ran the other way. Nationally, 326,164 against 414,507 establishments put us at 0.79 times. The ratio flipped, and it flipped for a reason worth knowing: one count is by headquarters, the other is by establishment. A California company with offices in four other states is one headquarters here and several establishments spread across the BLS state tables. A New York company with a California office is the reverse. California is a headquarters hub for this industry, so counting by headquarters inflates it relative to counting by site.
The lesson generalizes past this example. A ratio between two sources that holds at national level will not hold in a single state. This one moved by a factor of 1.7 across two geographies measured from the same pair of sources, so a state estimate carried over from a national ratio would have been badly wrong, and wrong in a direction nobody would have thought to check.
Notice which step did the damage. Product fit removed 59% of the market. Geography removed 85% of what survived. If someone asks where growth comes from next year, the table has already answered: a second state costs far less than a second product.
Your count is a floor, not a census
Every number above counts the companies a database could identify and classify. The real figure is higher, because classification is imperfect and some companies are missing from any source.
Our own team says this to prospects. On a call about sizing a niche segment, a Firmable rep told a customer his filtered count was too low before the customer could ask:
“we don’t actually have something for developers in particular… there’s still going to be false positives, and there’s probably more than, almost definitely more than 63 across [the country].“
That is the right way to hold a number. A count is a verified floor you can act on today, not a census. And a floor beats a census you cannot obtain, because you can call every company on a floor.
There is a stronger version of a floor, though, and it is worth asking your sources for. A detection that carries a confidence level and the date it was last observed is a different object from a bare yes. It lets you weight a list rather than treat every row as equally true, and it tells you how old the evidence is before you spend a quarter acting on it. Treat any count that arrives with no date attached as the one to distrust, including your own from last year.
This is also why SAM should be revisited instead of carved into a strategy document. Classification improves, companies get founded, and your filters get sharper as you learn which ones actually predict a sale.
The trap in every first search
Getting a segment clean is iterative work, and the exclusions do more than the inclusions. A business development lead narrowing a broad category to something workable described the difference:
“definitely going through 500 companies and doing a little bit more tuning, it’s definitely easier than the 7,000 that were before.”
So seven thousand to 500 was never one clever filter. It was a sequence of exclusions, each learned by looking at what came back wrong.
A count on its own is never enough. An industry category is a classification someone else built for their own purposes, and your segment is yours. The two overlap and they never match exactly. Open the list, read 50 rows, and find out what the category contains before building a plan on the number.
Frequently asked questions on servicable addressable market (SAM)
Serviceable addressable market is the portion of your total addressable market that your product, geography and permissions allow you to sell to. It is your TAM with your real constraints applied, expressed as a company count and usually also as an annual revenue figure.
Start from your TAM, apply each constraint as a separate filter while recording the count at each step, check the contents of the resulting list, then multiply the surviving company count by your average annual revenue per customer.
TAM is every company that could buy your product anywhere. SAM is the subset you can serve given your product’s fit, the geographies you operate in, the regulations that apply to you, and the segments you can identify. SAM is always smaller.
Because TAM answers an investor’s question and SOM answers a board’s question, so SAM has no obvious audience. That is a mistake. SAM is the only one of the three that changes when you make a decision, which makes it the number most worth examining.
Four recur: product fit, geography, regulation and permission, and whether the attribute defining your segment exists as data you can filter on. The last is the one most plans overlook and increasingly the one that binds.
Yes, and for many B2B products it is the most accurate way to define one. If your product integrates with, replaces, or depends on a particular system, then “companies running that system” is a segment you can count exactly, where “companies that would benefit from us” can only be estimated.
It cannot be worked at scale, however well you can describe it. Either find a proxy attribute that is held as data, or change the segment definition to something your systems can express.
Both, and they should be reported separately. The company count is measured and verifiable. The revenue figure depends on an assumption about average contract value, which is far less certain.
An ICP describes the characteristics of a company worth selling to. SAM is what you get when you count how many companies match it. The ICP is the definition, SAM is the measurement.
Because any count reflects the companies a source can identify and classify correctly. Treat it as a verified floor instead of a census. It remains the most useful number available, because every company in it can be contacted today.
Whenever a constraint changes. Entering a geography, shipping a feature that serves a new segment, or gaining a compliance approval each move the boundary, and so does learning which of your filters actually predict a sale.
Count your own SAM, and keep the list. The three searches behind the worked example take a few minutes each. Start a 14 day free trial, apply your constraints one at a time, and record what each one costs you.
Next in this series: how to calculate your serviceable obtainable market.
