TAM Analysis for B2B: How to Size Your Market With AI Agents
A B2B TAM analysis estimates the total revenue available to you by counting the companies that fit your ideal customer profile and multiplying by what each would realistically pay per year. Done bottom-up, it produces two things: a number for your board deck and a named list of accounts your team can work. AI agents now make the second part practical for small teams, because they can research and qualify thousands of accounts with a source behind each one.
What are TAM, SAM and SOM?
TAM is the total revenue if you won every possible customer; SAM is the part you can actually reach; SOM is the part you can realistically win in a set period. Standard definitions [1]:
- TAM (total addressable market): total annual demand for your product if 100% of the market bought it.
- SAM (serviceable available market): the portion of TAM your product, geography and sales channels can serve today.
- SOM (serviceable obtainable market): the share of SAM you can realistically capture against competitors, usually over the next one to three years.
For B2B, define all three in accounts first and dollars second. "4,000 companies at an average of $20,000 a year" can be checked, challenged and turned into territories. "$80 million" on its own cannot.
Top-down vs bottom-up TAM: which should you use?
Use bottom-up for any decision you will act on; use top-down only as a sanity check. Top-down starts from an industry-wide spending figure in an analyst report and takes a percentage. Bottom-up starts from a count of real accounts and multiplies by price.
| Top-down | Bottom-up | |
|---|---|---|
| Starting point | Industry spend figure from an analyst or market report | Count of accounts that match your ICP |
| Formula | Market size × share assumed relevant | Accounts × realistic annual contract value (ACV) |
| Speed | Hours | Days, or hours with agents |
| Main weakness | The share assumption is a guess, and report categories rarely match your product | Depends on account data quality and an honest ACV |
| Produces an account list | No | Yes |
| Best for | Early sanity check, very new categories | Planning, territories, hiring, fundraising |
The problem with top-down is that the most important number, the share of the category that applies to you, is usually an assumption. Two people can start from the same report and end an order of magnitude apart.
Bottom-up is harder, but every input is a fact you can check or a price you can defend. It also hands sales a list, which is the real payoff.
How do you run a bottom-up TAM analysis? A worked example
Here is a complete bottom-up calculation. Every input below is hypothetical, chosen to show the method; replace each with your own counts and prices.
The hypothetical product: a SaaS spend-management tool sold to US software companies with 200 to 2,000 employees. It integrates with Okta, so accounts must use Okta to be serviceable.
Step 1: Count accounts by size band and set ACV per band.
| Size band (employees) | Accounts (hypothetical) | ACV (hypothetical) | TAM |
|---|---|---|---|
| 200 to 499 | 3,500 | $12,000 | $42.0M |
| 500 to 999 | 1,700 | $24,000 | $40.8M |
| 1,000 to 2,000 | 800 | $48,000 | $38.4M |
| Total | 6,000 | $121.2M |
ACV per band should come from your own closed deals or, before you have any, from a pricing hypothesis tied to how you charge (per seat, per spend, per module). Do not use list price if you discount.
Step 2: Apply serviceability filters to get SAM. Our hypothetical filter is "uses Okta," and suppose research shows half the accounts in each band do.
| Size band | Serviceable accounts | ACV | SAM |
|---|---|---|---|
| 200 to 499 | 1,750 | $12,000 | $21.0M |
| 500 to 999 | 850 | $24,000 | $20.4M |
| 1,000 to 2,000 | 400 | $48,000 | $19.2M |
| Total | 3,000 | $60.6M |
Step 3: Estimate SOM from sales capacity, not from a percentage. Suppose you will have 4 account executives, each closing 30 new logos a year (hypothetical), for 3 years: 4 × 30 × 3 = 360 customers. At the SAM's blended ACV of about $20,200 ($60.6M ÷ 3,000), that is about $7.3M in ARR, or 12% of SAM.
Step 4: Check SOM against reality. If 12% of the serviceable market in three years is far above what any competitor has achieved, your win rate or capacity assumption is too high. If it is tiny, you may be under-hiring.
The capacity method is the important habit. "We will take 1% of a huge market" is the most common SOM mistake, because it has no mechanism behind it. Capacity times win rate does.
How do you build an account-level TAM list with AI agents?
Give agents your ICP as written criteria, let them qualify each candidate account with a cited source, then audit a sample before you trust the count. The steps:
- Write the ICP down as testable rules. "B2B software, US HQ, 200 to 2,000 employees, uses Okta, sells to enterprises" is testable. "Innovative mid-market companies" is not. Our guide on giving an AI agent your ICP without a data contract shows the file format.
- Pull the candidate universe wide. Use registries and filings for firmographics (SEC EDGAR for US public companies [2], Companies House for UK companies [3]), plus LinkedIn and Sales Navigator searches, Crunchbase, and public ATS job feeds. Our 11 free B2B data sources lists more.
- Sanity-check the size of the universe. US Census County Business Patterns gives free establishment counts by NAICS industry and employment size [4]. It has no company names, but if your candidate list is far larger than the Census count for that industry and size, you have duplicates or misclassified companies.
- Have agents qualify each account against every rule and record the evidence: the job post that names Okta, the LinkedIn page that shows headcount, the company site that shows the customer segment. Separate firmographic and technographic fields so you can see which filter removed which accounts.
- Deduplicate by root domain and resolve parents and subsidiaries, so one company is not counted three times.
- Audit a random sample of 50 qualified and 50 rejected accounts by hand. Measure how often the agent was right in each group. If precision is low, fix the rules, not the sample. Our post on evals for GTM research agents covers how to score this.
- Tag each account with segment, size band and ACV band, then compute TAM, SAM and SOM from the list itself.
Running this with agents changes the economics. A person qualifying accounts at a few minutes each cannot cover thousands; agents running in parallel can, and they leave a source trail a person can check.
Drevon is built for steps 2 and 4: it turns a plain-English ICP into a live Sales Navigator list with the reason each match qualifies, and researches company sites, job boards and Crunchbase with a source link on every claim, on the AI subscription you already have. The Drevon skills library also includes a build-tam skill for sourcing accounts into a TAM list.
How do you pick a beachhead segment from your TAM?
Pick the segment where pain is sharpest and deals close fastest, even if it is not the biggest. A TAM list split into segments lets you compare them on evidence instead of opinion.
Drevon's beachhead-segment skill scores candidate segments on burning pain, willingness to pay, winnable share and referral potential. Its output has "a cited source behind every score and unsourced scores marked as guesses" [5]. Whatever tool you use, keep those four criteria and keep the sources: a segment chosen on unsourced enthusiasm is a guess with a spreadsheet around it.
In the worked example, the 500 to 999 band might win even though it is not the largest by account count. Its ACV is big enough to support a sales team, and its companies are small enough that one champion can buy. The point is to decide with segment-level data, then point your lead scoring model at that segment first.
How do you present a TAM analysis so it holds up?
Show the funnel from accounts to dollars, the source behind each filter, and how the answer moves when your shakiest input moves. A TAM slide that hides its inputs invites the one question you cannot answer.
A layout that survives scrutiny:
- Start with the account funnel: 6,000 accounts in the ICP, 3,000 serviceable, 360 obtainable in three years (from the hypothetical example above).
- Next to each step, name the filter and its evidence: "uses Okta, from job posts and DNS records, sample-audited at 50 accounts."
- Show ACV per segment and where it came from: closed deals, pilots or a pricing hypothesis.
- Add a sensitivity table for the two inputs you trust least.
Using the same hypothetical numbers, here is what that sensitivity table looks like:
| Assumption changed | SAM |
|---|---|
| Base case (50% use Okta, ACV as above) | $60.6M |
| Only 40% use Okta | about $48.5M |
| 60% use Okta | about $72.7M |
| ACV 20% lower in every band | about $48.5M |
The table tells a reader that the market is roughly $50M to $70M serviceable under reasonable assumptions, which is more honest and more useful than a single precise figure. It also tells you which input to research next: here, the Okta share moves SAM as much as a 20% price change, so it is worth an audit.
What are the most common TAM analysis mistakes?
The most common mistake is sizing a market you cannot reach. Others we see often:
- Top-down only. A percentage of an analyst figure, with no account count behind it.
- List-price ACV. If you discount 30% on average, your TAM is 30% too high.
- Counting companies, not buying units. If each business unit buys separately, a large enterprise is several accounts, not one.
- Double counting subsidiaries and regional entities.
- Skipping the serviceability filter. Required integrations, languages, regulations and sales channels all shrink SAM.
- SOM as a flat percentage. Tie it to sales capacity and win rate.
- Static TAM. Companies grow into and out of your size bands, change stacks, get acquired. Refresh the list at least twice a year.
- No sources. A TAM number nobody can trace back to accounts will not survive a board question.
If your TAM analysis feeds a broader plan, our AI GTM strategy framework shows where market sizing sits in the sequence.
Frequently asked questions
What is a TAM analysis?
A TAM analysis estimates the total annual revenue available for your product by sizing the full market of potential customers. In B2B, it is best built as a count of fitting accounts multiplied by realistic ACV.
How do you calculate TAM bottom-up?
Count the accounts that match your ICP, group them by size or segment, multiply each group by its realistic annual contract value, and sum the groups.
What is a good SOM as a percentage of SAM?
There is no standard percentage. Derive SOM from your sales capacity and win rate over a set period, then check whether the resulting share is plausible against competitors.
How often should you update your TAM?
At least twice a year, and whenever you change ICP, pricing or product. Account-level TAM lists drift as companies grow, shrink and change tools.
Can AI calculate my TAM?
AI agents can build and qualify the account list, which is the hard part. You still set the ICP rules, ACV assumptions and capacity inputs, and you should audit a sample of the agents' work.
Want a TAM list where every account comes with a reason and a source? Download Drevon or book a demo to run it on your ICP.
Sources
- Wikipedia, "Total addressable market" (TAM, SAM and SOM definitions), accessed 2026-10-10.
- U.S. SEC, "EDGAR Application Programming Interfaces", accessed 2026-10-10.
- GOV.UK, "Search the Companies House register", accessed 2026-10-10.
- U.S. Census Bureau, "County Business Patterns", accessed 2026-10-10.
- Drevon Skills, "beachhead-segment", accessed 2026-10-10.
- Drevon Skills, "build-tam", accessed 2026-10-10.