Total addressable market is the total annual revenue available if every company that could buy your product did buy it, from you. You calculate it by multiplying the number of potential customers by average annual revenue per customer. The difficulty sits in the first number, because there is no single correct count of how many companies exist in any market.
We went looking for that count recently. For one US industry, information technology services, we found four defensible answers from three respected sources. The smallest was 127,147. The largest was 414,507.
Nobody was wrong. That is the problem, and it is more interesting than it first appears.
There is no true market size
Every article about TAM, including this one, treats the company count as a fact you go and retrieve. Get the right source, apply the formula, receive the number.
That is not how it works, and anyone who has built one seriously already knows it. Market size is a measurement, and every measurement carries a definition, a method and a date. Change any of the three and the answer changes with it. The single figure on the slide is the last step of a long chain of choices, and the chain is where all the meaning lives.
Which leads to the argument of this article. The number is not the deliverable. The working is the deliverable. A TAM you cannot reconstruct is a decoration. A TAM with its definitions, sources and dates attached is a tool other people can argue with, and being able to argue with it is what makes it useful.
It also has a second job, and it is the one that pays. Broken into its parts, a TAM prices your growth options: what a new geography is worth against a new segment against simply selling harder into the market you already serve. A single headline figure cannot do that. A decomposed one does it in an afternoon.
So this article covers both. How to calculate a market size you can defend, and how to read that calculation as a growth plan.
What TAM measures
TAM is a ceiling. It answers one question: if we won everything, how big would this be?
Potential customers in market × Average annual revenue per customer = TAMInvestors ask for it to test whether a business can get large. Product teams use it to compare opportunities. Sales leaders mostly use it to see how much of a market they are currently ignoring.
It is a planning input and never a forecast. Your serviceable addressable market is the slice you can sell to, and your serviceable obtainable market is the slice you can win this year. TAM sits above both and sets the outer bound.
Four sources, four answers
Here is what happened when we counted one US industry four ways.
| Source | What it counts | Number |
|---|---|---|
| Census Statistics of US Businesses, 2022 | Firms | 127,147 |
| Census Statistics of US Businesses, 2022 | Establishments | 145,211 |
| Firmable, September 2026 | Companies in IT consulting and support services | 326,164 |
| BLS Quarterly Census of Employment and Wages, Q1 2025 | Establishments, private ownership | 414,507 |
A 3.3x spread between smallest and largest. Two of those numbers come from the same agency.
But before deciding one of them is wrong, look at the employment figures underneath. The Census SUSB tables put employment in this industry at 2,126,573. The BLS QCEW series puts it at 2,385,381, which is 12.2% apart across different years.
So two federal agencies broadly agree on how many people work in this industry and disagree by a factor of nearly three on how many businesses there are. Both are measuring carefully. They are measuring different things, and neither is being careless about it.
Why the counts disagree
Firms are not establishments. One company with four offices is one firm and four establishments. Census publishes both for the identical industry code: 127,147 firms against 145,211 establishments. If your product is sold once per company, you want firms. If it is sold per site, you want establishments. Choosing the wrong unit distorts your TAM before you have done any arithmetic.
Agencies count different populations. The QCEW establishment count is 2.85 times the Census one for the same industry code, because it counts unemployment insurance reporting units, which include very small and intermittent ones. Different inclusion rules, not different levels of rigor.
Taxonomies are not the same shape. NAICS is a government classification built for economic statistics. A commercial database classifies companies by what they say they do. The two overlap heavily and match exactly never. Expecting identical counts from them is a category error.
Vintages differ. Our Census figures are 2022, the BLS figures are the first quarter of 2025, and Firmable’s are September 2026. An industry moves a long way in four years, and official statistics are always behind.
None of these are flaws to be corrected. They are properties of measuring anything at scale, and the useful response is to state which choice you made and why.
Top-down and bottom-up
Two routes to a TAM. The honest answer is to run both.
Top-down
Start with a published market size, then take your share of it. Fast, cheap, and fine for a board slide. The weakness is that you inherit someone else’s definition of the market, written for their purpose instead of yours. If the report counts every company with an IT function and you sell only to IT service providers, the number is not yours.
Bottom-up
Count the companies matching your criteria, then multiply by what they would pay. Slower, more defensible, and the version investors prefer, because you can show the list.
Its weakness is that your count is only as good as your source, and practitioners tend to be more honest about this than the articles written about them. A business development lead at an industrial equipment supplier described his process:
“I did a little bit of Google, but also I used the ChatGPT for another market research and then Sales Navigator. Then I tried to basically cross-check everything and I’ve created an Excel file with all the companies that I was able to find, but they’re not all of them, of course.“
Three sources, cross-checked by hand, and he still cannot tell you what he missed. That last clause is the most honest sentence in this article.
Run both methods. Landing close together gives you a number you can defend. Landing three times apart teaches you something about your market definition before you stake a hiring plan on it.
A worked example
Say you sell a professional services automation platform to IT service providers in the United States.
Step one, define the market. All US companies in information technology and services. A Firmable search on 19 September 2026 returns 786,476 companies headquartered in the US.
Too broad. That includes software vendors, hardware manufacturers and cloud hosts, most of whom will never buy a PSA tool.
Step two, define who could plausibly buy. Narrow to IT consulting and support services. The same search returns 326,164.
Step three, cross-check top-down. The nearest government classification is NAICS 5415, computer systems design and related services. Census counts 127,147 firms. BLS counts 414,507 establishments. Our 326,164 sits between them.
That is what you want from a cross-check. Not agreement, which would be suspicious, but a number inside the range of official answers with a clear account of why it differs from each.
Step four, apply revenue per customer. At $12,000 average contract value:
326,164 companies × $12,000 = $3.9 billion TAMNow separate the two halves of that sum, because they are not equally reliable. The company count is measured, dated and reproducible. The $12,000 is an assumption about pricing. Presenting them as one figure hides the weaker input behind the stronger one, and a reader who cannot tell them apart will trust the total less, not more.
Keep both numbers from step one and step two. The gap between 786,476 and 326,164 is not waste, it is your first growth option priced, and we come back to it shortly.
What changes when you cross a border
The same filter, run in three markets on the same day:
| Market | Companies in IT and services |
|---|---|
| United States | 786,476 |
| Australia | 88,761 |
| Singapore | 17,764 |
The US is 44 times Singapore on an identical definition.
This matters more than it looks. A segment that supports a dedicated sales team in one market may not support a single rep in another. Teams that build one global TAM and divide it by headcount get this wrong constantly. Size each market separately, then decide where people go.
Turning TAM into a growth plan
A TAM does not contain a strategy. Decomposed, it prices your options, which is more useful.
Most growth decisions get made the other way round. Somebody decides the company is going upmarket, or going international, or building the next product. The sizing work is commissioned afterwards to support a decision already taken, and it arrives with the conclusion baked in. Running the decomposition first inverts that: price every lever, then choose one.
There are four levers. The same worked example prices three of them.
| Lever | What it means | Headroom in the example |
|---|---|---|
| Win more of what you already target | Same product, same market, more sales capacity | The gap between your SOM and your SAM |
| Extend geography | Same product, new market | 277,401 more US companies outside California, plus 88,761 in Australia and 17,764 in Singapore |
| Extend segment | Same product, adjacent category | The wider IT and services category holds 786,476 against your 326,164. Computer software on its own returns 447,887, though the two sets overlap and should not be added |
| Extend product | New product, same buyers | Not a counting question |
Notice which one refuses to be measured. Geography and segment are arithmetic, and you can price them in an afternoon. Building a new product is a bet about demand that does not exist yet, so no market count will settle it.
That matters because the unmeasurable lever is usually the one that gets chosen. It is the most exciting to talk about, the hardest to argue with, and the only one where nobody can produce a number that contradicts you. Knowing which kind of decision you are making is most of the discipline here.
The cheapest lever is frequently the dull one. A second state needs a rep and a plan. A second product needs a roadmap, a launch and a year.
When a lever is exhausted
Compare your SOM to your SAM and the answer usually falls out.
Winning a small fraction of your SAM? Your constraint is capacity, and entering a new market will not fix it. You will simply under-serve two markets instead of one.
Already holding most of your SAM? Capacity will not help, because there is nothing left to sell to. Expansion is the only lever with anything behind it. The sales lead at a wholesale fibre network knew exactly which position he was in before we showed him anything, which is why the conversation moved to new regions within minutes.
That comparison is the practical payoff of doing all three calculations rather than only the flattering one. TAM alone tells you a market is large. TAM against SAM against SOM tells you which of your problems is real.
The method was built for a world where you could not count
Here is the part worth sitting with.
TAM as a discipline comes from investor conversations, where the question was whether a market was big enough to be worth backing. In that setting an estimate was the only thing available. You could not enumerate a market, so you modeled one, and the entire apparatus of top-down multiplication exists because counting was impossible.
Counting is no longer impossible. Every figure in this article came from a search that takes about a minute, and the output is not an estimate. It is a list of named companies you can open, filter and call.
The method has not caught up with that. Teams still produce a slide with a number on it, when the same effort now produces a list of 326,164 companies that doubles as the top of the pipeline. The sizing exercise and the prospecting exercise have become the same exercise, and most go-to-market plans still run them as two unrelated projects months apart.
The more interesting question is direction, not size
Counting is only the first thing that got cheap. The same infrastructure that makes a market countable makes it observable over time.
TAM has always been a static quantity, and a static quantity answers a fairly boring question. How big is this. The question revenue teams actually care about is usually about movement: which part of this market is growing, what are those companies adopting, what are they retiring, and which of those changes creates a reason to call. A market of 326,164 companies where a third are visibly adopting something adjacent to your product is a different market from one where nobody is, and the two have identical TAMs.
That used to be unanswerable outside an analyst subscription. It is now a property you can read off the market itself. Firmable’s Tech Intelligence is built for this, letting you map your total addressable market and track aggregated trends such as AI maturity across it.
Our head of sales for ANZ, Luke Williams, hit this on a call with an online course provider trying to size its opportunity in Japan.
Japan is a hard market to size from the outside. English-language sources thin out quickly, the registries work differently, and a top-down estimate is close to all you can get. The team on the call had run into exactly that wall.
So instead of estimating the market, he looked at what Japanese companies were actually running, and used AI adoption as the segment definition.
The logic holds up well. AI tools appearing in a company’s stack represent a purchase that has already been made, and that purchase creates work. Tools arrive faster than the skills to use them, and somebody then has to bring a team up to a standard nobody had last year. That gap is exactly what the courses are for, which makes visible adoption a better buying signal than stated interest. One is a decision already taken. The other is a survey response.
So the question stopped being “how big is the Japanese market for online courses”, which nobody was going to answer on a call, and became “which Japanese companies have adopted AI and now need their people to be good at it”. The second question comes with a list attached, and the list is the outreach plan.
Notice what moved there. Not the size of the market. The definition of the segment, using an attribute that was not available as a filter a couple of years ago.
Which is a change in what the artifact is. A TAM has been a number on a slide, produced once a year, correct on the day it was made. The same work now produces something with a direction attached, which is worth more than the number was.
The test
Which suggests a different test for whether your TAM work was any good. Not “is the number defensible”, though it should be. Ask instead: at the end of it, do we have a list? If the answer is no, you spent a week producing a slide.
Five ways TAM goes wrong
Counting the wrong unit. Firms, establishments, or accounts in your CRM. Decide which one your product is sold to, then count that.
Borrowing a market definition that is not your market. If the report’s market includes segments you do not serve, your TAM is inflated and every derived number with it.
Presenting an estimate as a measurement. A TAM rests on at least two assumptions. Publish the range. A midpoint on its own oversells how much you know.
Ignoring what you are not permitted to sell to. Some of your market is closed by regulation instead of competition. The sales lead at a wholesale fibre network described his ceiling:
“We’re not allowed to be retailers, so it’s against our constitution with our government, so when we were tasked to build the fibre network, that was one of the rules that we’re never allowed to be a retailer, we’re only ever allowed to be a wholesaler.”
No sizing exercise will discover that constraint. You have to know it and subtract it.
Calculating it once. A TAM from 2024 is a historical document. Markets move, your product moves, and the count moves with both.
Where the numbers come from
Government statistics are free, authoritative and slow. Census SUSB and BLS QCEW in the US, the ABS in Australia, SingStat in Singapore. Use them for the top-down pass and to sanity-check anything commercial.
Analyst reports give you revenue-sized markets instead of company counts, which saves a step when their definition matches yours.
A B2B sales platform like Firmable gives you the bottom-up count and hands you the list behind it. That is the practical difference. A statistic tells you a market holds 326,164 companies. A database tells you which ones. Firmable covers Australia, New Zealand, SEA, the United States and Canada, and every count here came from a search anyone with an account can rerun.
Your own CRM is the one teams forget. You already know what customers pay, how long they take to close and which segments renew. That is where average revenue per customer should come from, not a guess.
Frequently asked questions on TAM (Total addressable market)
Total addressable market is the total annual revenue available if every company that could buy your product bought it from you. It is calculated by multiplying the number of potential customers by average annual revenue per customer, and it represents a ceiling rather than a forecast.
Multiply the number of potential customers in your market by average annual revenue per customer. Get the customer count either top-down from published statistics or bottom-up by counting companies that match your criteria. Running both and comparing them is more reliable than either alone.
Because sources count different units and different populations. Firms are not establishments, government and commercial taxonomies classify companies differently, agencies apply different inclusion rules, and published data carries different vintages. For one US industry, four sources returned counts from 127,147 to 414,507, a spread of 3.3 times.
Yes, but as a reasoning tool rather than a headline figure. A TAM presented as a single number invites false confidence. A TAM presented with its definitions, sources and dates can be checked and argued with, which is what makes it worth having.
Top-down starts with a published market size and takes your share of it. Bottom-up counts the companies matching your criteria and multiplies by what they would pay. Top-down is faster, bottom-up is more defensible because you can show the underlying list.
TAM is everyone who could buy. SAM is the portion you can serve given your product, geography and constraints. SOM is the portion you can realistically win in a given period, after subtracting existing customers and accounts you cannot reach.
Yes, and it is usually more useful than the size itself. Company counts, segment composition and technology adoption across a market can all be measured repeatedly, which turns TAM from an annual exercise into a reading you can take whenever a plan depends on it.
Decompose it rather than reading the headline figure. Comparing the size of your current segment against the wider category, other geographies, and adjacent segments prices each growth lever before you commit to one. Comparing SOM to SAM then tells you whether your constraint is sales capacity or market exhaustion.
There is no universal threshold, and a large TAM proves less than a well-evidenced one. Investors generally care more about whether you can show your working, name your sources and explain your assumptions than about the size of the headline figure.
Both, reported separately. The company count is measured and verifiable. The revenue figure depends on an assumption about pricing. Separating them lets a reader judge the reliability of each.
At least annually, and whenever you change pricing, enter a market, or ship a product serving a new segment. Each of those changes one of the two inputs in the formula.
Finish with a list, not a slide. Every count in this article came from a search that takes about a minute, and the output is a set of named companies you can work. Start a 14 day free trial and build your bottom-up TAM against real accounts.
Next in this series: how to calculate your serviceable addressable market.
