Market sizing

A TAM number on its own is worthless. Anyone can produce one, everyone produces a different one, and the reader knows it. What carries weight in a board room or an investment memo is two independent estimates and an honest account of the gap between them.

That is the entire method here: size it twice, by methods that share no assumptions, then explain the difference rather than averaging it away.

Definitions worth being strict about

These three terms get used interchangeably and should not be:

TAM — every entity that has the problem, at full price. Not "the market for software", but the market for this product's specific job. If your TAM is a category number lifted from an analyst report, it is measuring someone else's job, not yours.

SAM — the slice of TAM you can sell to today. Right segment, right geography, right compliance posture, right integrations. The gap between TAM and SAM is a roadmap, and stating it explicitly is more persuasive than hiding it, because it shows you know what you are not yet.

SOM — the slice of SAM you can realistically win in the planning horizon given the sales capacity you actually have. This is the only one of the three that belongs anywhere near a quota model.

The most common error in this analysis is quoting TAM where SOM is the relevant number. TAM answers "is this worth building". SOM answers "what will we sell". Confusing them is how a plan gets built on a number that was never a forecast.

The workflow

1. Build bottom-up first

Bottom-up is the estimate you control, and doing it first stops the top-down number from anchoring you.

units × attach rate × annual price

Units must be countable and the counting method must be stated. "Companies with more than 200 employees and a public engineering job posting in the last 12 months, from a Crunchbase and Greenhouse export dated 2026-07-02" is a unit count. "Enterprises" is not.

Attach rate is the share of those units with enough of the problem to buy. This is almost always the softest input in the model and it deserves a wide range. Anchor it on something real: your own install-base penetration in a segment you have worked, or an observed adoption rate for an adjacent product at a comparable stage.

Annual price should be realized ACV, not list. If you discount 20% on average, list price overstates the market by 25%.

2. Build top-down independently

published market total × relevant share

The discipline that makes this useful is independence. If you derive the relevant share from the same reasoning that produced your attach rate, you have not built a second estimate, you have built the first one twice.

Be specific about what the published total covers. Most category numbers are broader than your product's job — that is not disqualifying, it is exactly what the share multiplier is for, but the multiplier needs a stated basis.

3. Reconcile

PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python3 -m gtmkit.sizing --spec market.json

The tool reports the ratio between the two methods and says what to do about it:

• Within 2x — you have a defensible range. Report both endpoints and both methods. A reconciled range beats either point estimate.
• 2x to 5x — usually a definition mismatch. Most often the top-down category includes adjacent products the bottom-up unit count excludes. Reconcile the definitions before either number leaves the building.
• Beyond 5x — one of them encodes a wrong assumption, and averaging would hide it. Find the error. Check whether the unit count is really the set of buyers, and whether the published total measures the same job you do.

The instinct when two estimates disagree is to split the difference. Resist it. The gap is the finding.

4. Derive SAM and SOM with stated constraints

SAM comes off TAM by naming what excludes the rest — geography, segment, regulatory, integration. Each exclusion should be a sentence, not a percentage pulled from nowhere.

SOM should be checked against sales capacity, not just asserted as a share. If SOM implies 240 new logos and the team can work 60, one of those numbers is wrong. The tool prompts for this check because it is the one that makes the number plannable.

Presenting it

The market slide that survives diligence has four things:

1. The range, not a point. "$300M–$375M depending on attach rate" is more credible than "$340M", because the first admits it is an estimate.
2. Both methods, briefly. One line each on how the number was built.
3. The reconciliation. Why the two differ and which you weight.
4. SOM against capacity. The number that connects to the plan.

Then the assumption ledger as an appendix. Nobody reads it, and its existence changes how the rest is received.

Common failures

Analyst-report laundering. Quoting a category number, applying an unexplained percentage, and presenting the result as analysis. If the share multiplier has no basis, the whole estimate has no basis.

TAM inflation by definition creep. Widening the definition until the number is impressive. A reader who notices this discounts everything else in the deck.

Sizing the market instead of the segment. If you sell to one segment, size that segment. A large TAM with a tiny reachable slice is a worse position than a modest TAM you can actually capture, and sophisticated readers know it.

Forgetting that price is an input. Market size scales linearly with your price assumption. If pricing is undecided, size it at two or three price points rather than picking one silently.

No time dimension. A market growing 40% annually and one that is flat are different opportunities at the same size. State the growth rate and where it came from, or say you do not know.

Reference material

• references/sizing-methods.md — additional methods (value-based sizing, analogous-market sizing, spend-displacement sizing) and when each is the right instrument.

A worked spec lives at examples/sizing/agent-observability-na.json.