TAM SAM SOM framework for PM interviews

TAM SAM SOM is a strategic sizing tool, not a Fermi estimation technique. That distinction matters in a PM interview: Fermi estimation asks you to derive a single quantity (how many piano tuners in NYC?), while TAM SAM SOM asks you to map an addressable opportunity and reason about what share is realistically capturable. Interviewers use both, but they expect different things. Using Fermi scaffolding for a TAM SAM SOM question, or vice versa, signals that you have memorized a framework without knowing when to deploy it.

The three layers nest concentrically. TAM (total addressable market) is the theoretical ceiling: the revenue available if you had 100% market share with no constraints. SAM (serviceable addressable market) is the portion your product and go-to-market can actually reach, constrained by geography, channel, customer type, regulatory fit, and pricing model. SOM (serviceable obtainable market) is the realistic near-term capture given competition and your execution capacity. The most common mistake is treating SOM as a percentage of SAM without reasoning about how you would actually win. “We’ll take 5% of SAM” is the single most reliable tell that a candidate is in investor-deck mode rather than PM mode.

Choosing your method and saying it aloud

State your method before touching math. Interviewers score the narration first.

Bottom-up wins for B2B, infrastructure, and capacity-constrained markets. You start from a countable population of buyers (companies, teams, users), apply adoption rates, and multiply by unit price. The number is grounded in specific demand signals.

Top-down fits roughly 80% of consumer-market questions. You start from a published market size or a demographic denominator, apply filters to get to SAM, and apply competitive logic to reach SOM. It is faster and relies on reference anchors that you should have memorized: US population ~342M, ~130M US households, ~1.25B global knowledge workers, ~33M US SMBs, median US household income ~$83,730.

The dual-check discipline: once you produce a number via one method, triangulate with the other. If your bottom-up gives $4B and a top-down check gives $40B, catch and explain that gap before the interviewer does. An unexplained order-of-magnitude difference signals an error in your assumptions, not rounding.

Worked example: AI coding assistant market (US, B2B, 2026)

Clarify first. “Before I size this: are we sizing the current market or a 3-year forecast? US only or global? I’ll assume US current market to keep the numbers verifiable.”

Method choice. “This is a B2B product sold to engineering teams, so I’ll go bottom-up from the supply of buyers rather than top-down from abstract developer populations.”

TAM. US software developers: roughly 4.4M. Developers at companies where the team has a budget for tooling (excludes solopreneurs and very small shops, roughly 30% of that population): ~3M developers in scope. At a $20/seat/month price point, that is $3M × $240/year = $720M TAM.

Sanity check. Top-down: the US developer tools and productivity software market is estimated at $600M–$900M for AI-specific tooling in 2026. We are in range. If these numbers were off by an order of magnitude, I would find the flaw before moving on.

SAM. Not every developer is at a company that has adopted AI tooling yet. Filtering to teams with at least one approved AI tool already in their stack (roughly 60% of addressable companies by 2026 adoption curves): ~$430M SAM.

SOM. There are two entrenched products with significant distribution advantages and switching costs. A realistic entry targets a specific niche: say, teams using a specific IDE that existing players serve poorly. In year 2, capturing 3–5% of SAM through that wedge gives a $13M–$22M SOM. By year 4, if the wedge holds, $80–120M is credible.

The implication. “A $720M TAM is large enough to build a venture-scale business, but the SAM narrows quickly once you apply enterprise procurement cycles and existing contracts. The SOM math only justifies the investment if we have a specific wedge the incumbents cannot easily replicate. Without a defensible niche, the capital required to compete on distribution alone does not return.”

That last sentence is what most candidates skip. Stopping at the number is an analyst answer. The implication step is what separates PM market sizing from a spreadsheet exercise.

The 2026 reframe: AI market sizing needs different inputs

Classic TAM SAM SOM assumed seats times price. For AI products in 2026, that model often breaks down. Usage-based and per-query pricing means SAM is frequently better expressed as “total queries addressable at viable unit economics” rather than “number of seats.” A model serving 100,000 users at $0.002 per query with average 50 queries per day generates very different revenue math than 100,000 seats at $20/month, even if the user population is identical.

For AI products, three questions replace the standard SAM calculation: (1) What is the total volume of queries or tasks this market generates per unit time? (2) At what price per query or outcome does the product remain economically viable for the buyer? (3) What portion of that volume is addressable given latency, compliance, and integration constraints? Candidates who can reframe SAM from seats to queries signal AI PM fluency at Anthropic, OpenAI, Perplexity, or Cursor interviews.

The viability bar is also higher now. Because almost anything can be built quickly, the economic argument for building has to be airtight. Interviewers at AI-native companies increasingly probe unit economics: can the product generate profit at scale, or does the cost of inference eat the margin? Know your model cost structure before you claim a SOM.

What interviewers score

Structure comes first: a logical tree before any numbers. Judgment is second: are your assumptions anchored to reference points or invented? Communication is third: did you narrate every step so the interviewer could follow and challenge? The final number needs to land within one order of magnitude of reality. Precision beyond that is irrelevant.

Claiming more than 5% SOM in year 1–3 without naming a specific entry wedge and a reason incumbents cannot replicate it is a red flag. It signals optimism rather than product thinking.

Strong and weak answers

strong

"Before I size this, let me clarify: are we sizing the current market or the potential market in 3–5 years, and should I focus on US only? [Gets answer: US, current.] I'll use a bottom-up approach here because food delivery is a capacity-constrained business: the supply of drivers and restaurant partners is the real ceiling, not abstract consumer demand. I'll segment by order frequency. US population is about 342M; delivery-age adults with disposable income to use delivery regularly is maybe 120M. Heavy users (3+ orders/week), roughly 8% of that group, $40 AOV: that's ~9.6M people × 3 × 52 × $40 = about $60B. Moderate (1–2/week), 20% of 120M, $35 AOV: 24M × 1.5 × 52 × $35 = ~$65B. Light (a few times a month), 35%, $30 AOV: 42M × 24 × $30 = ~$30B. Raw TAM roughly $100B. Sanity check: industry estimates quote the US food delivery market at $90–100B for 2024, so we are in the right zone. SAM: if we are targeting urban and suburban markets with existing restaurant density, that is roughly 60% of TAM, so ~$60B. SOM: there are two entrenched leaders with significant liquidity advantages. Realistically in the first two years, entering one or two underserved metros targeting a specific niche (say, premium health-food restaurants with no existing delivery integration) gives a credible wedge. I would model a realistic SOM of $300–500M in year 2, growing to $2–3B by year 5 if the wedge holds. The implication: this is a winner-take-most market, so the investment case only works if we can identify a defensible niche the incumbents are structurally unable to serve well. Otherwise the SOM math does not justify the capital required to compete."

weak

"The TAM for food delivery in the US is about $50 billion. Our SAM would be the major metro areas where we operate, so maybe $15 billion. And our SOM: if we can get just 5% of that, that's $750 million in revenue." Why it fails: the candidate started with a number they recalled rather than building one; they applied an arbitrary percentage to SAM with no competitive or execution reasoning; "5% of SAM" is the textbook lazy SOM move that every experienced interviewer recognizes immediately; and the answer stops at the number. There is no implication, no recommendation, and nothing that shows how this person thinks about actually winning the market.

Handling pushback mid-calculation

If the interviewer challenges an assumption, do not defend it or abandon it: update it transparently. “You’re right that 8% for heavy users might be high: if it’s closer to 4%, that drops the TAM to roughly $70B. Does that change my SAM and SOM materially? Let me re-check.” Demonstrating that you can flex assumptions without losing the structure is itself what the interviewer is scoring. The question “what if your assumption is wrong?” is not a trap. It is an invitation to show that your framework holds even when inputs change.

For more on the viability logic that underpins when TAM SAM SOM is worth doing at all, see feasibility is free and proving viability. For the unit economics layer specific to AI products, see LLM unit economics one-pager.