Most market sizing estimates don’t fall apart because the final number is wrong. They fall apart because nobody can follow how it was reached. Someone in the room asks “where does the 40% come from?”, the answer is “a report I saw”, and from that moment every other number on the slide is under suspicion too.
A defensible estimate is one that another person can retrace step by step, and then argue with your assumptions rather than your arithmetic. That is a much better conversation to be in. This guide walks through a method that gets you there: define the market precisely, build the number two independent ways, reconcile them, and keep a record of every input.
Define the market in one sentence
Before you look up a single figure, write down what you are measuring. A useful definition answers five questions:
- Who buys? Consumers, small businesses, hospitals, governments.
- What do they buy? The product or service category, described in the buyer’s terms rather than yours.
- Where? Countries or regions.
- When? A specific year, or a range for a forecast.
- Measured how? Annual revenue, units shipped, number of subscribers. Pick one.
For example (a hypothetical market, used throughout this article): “Annual spending by independent dental clinics in Germany on cloud-based practice-management software, in euros, for 2026.”
That sentence already rules out a lot. Hospital dental departments are out. On-premise software is out. Clinic chains are out. Every source you find later will have to be checked against it, and that is the point: most sizing errors come from quietly mixing figures that measure different things.
TAM, SAM and SOM, used properly
The three-layer framework is standard, but it is often used loosely.
- TAM (total addressable market) is total demand for the category if every potential buyer in your definition bought. It is a ceiling, not a forecast.
- SAM (serviceable addressable market) is the part of TAM your product could actually serve today, given its features, language, price point, regulatory approvals and sales channels.
- SOM (serviceable obtainable market) is the share of SAM you can realistically win within a stated period, given competition and your capacity to sell.
Two habits undermine the framework. The first is a TAM that is really a whole industry (“healthcare spending is enormous, so our TAM is enormous”). If the product could never serve most of that spending, the number is not informative. The second is a SOM that is simply “1% of SAM” with no reasoning. A SOM should be built from something concrete: how many customers your sales motion can close per year, the share comparable entrants achieved, or the conversion rates you have actually observed.
Top-down: start from a published total and narrow it
The top-down approach starts with a large, published figure and applies filters until it matches your definition.
- Find a credible starting total. Government statistics, regulator data, industry associations and company filings are usually better starting points than a market research press release, because their definitions are published.
- Apply one filter at a time. For the dental example you might start from total spending on practice software across all healthcare providers, then filter to dental practices, then to independent clinics, then to cloud deployment.
- Justify each filter with its own source. “Independent clinics are X% of dental practices” needs a citation, just like the starting total.
- Note every definitional mismatch. If the starting figure includes hardware, or covers German-speaking Europe rather than Germany, write that down and adjust explicitly.
Top-down is fast, and it anchors you to figures other people already accept. Its weakness is compounding: five filters that are each “roughly right” can multiply into a number that is badly off, and the published starting total often uses a category definition that doesn’t quite match yours.
Bottom-up: count buyers, frequency and price
The bottom-up approach builds the market from its parts:
market size = number of buyers × share that buy × units per buyer per year × price per unit
For the dental example, using made-up numbers purely to show the mechanics:
| Input | Illustrative value | Where a real value might come from |
|---|---|---|
| Independent dental clinics in Germany | 10,000 (hypothetical) | Professional chamber or association registers, business registries |
| Share using cloud practice software | 30% (hypothetical) | Industry surveys, vendor disclosures, interviews |
| Seats per clinic | 4 (hypothetical) | Typical clinic staffing data, vendor pricing tiers |
| Price per seat per year | €600 (hypothetical) | Public price pages, quotes |
Multiplied out, that gives 10,000 × 0.3 × 4 × €600 = €7.2 million a year. Again, these are invented inputs. The value of the exercise is that every line is a claim someone can check, and every line points to the kind of evidence that would confirm or overturn it.
Bottom-up estimates are easier to defend because each input is concrete. Their weakness is coverage: it is easy to forget a buyer segment, or to use list prices when most customers pay discounted ones.
Cross-check, then reconcile the gap
Once you have both numbers, compare them. They will rarely match, and that is useful information.
If they are far apart, don’t average them. Find the assumption that explains the difference. Common culprits:
- The top-down starting total uses a broader category definition than yours.
- The bottom-up adoption rate comes from a survey of early adopters, not the whole population.
- One side uses list prices and the other uses realized revenue.
- One side counts a segment the other leaves out.
Fix the input that is wrong, or explain in a sentence why the two methods measure slightly different things. Then add a third angle if you can:
- Supply side. Add up the disclosed revenues of the main vendors in the category, and estimate the share held by the rest.
- Value-based. Estimate what the product saves or earns each customer, and check that the implied price is plausible against it.
- Analogues. Look at an adjacent country or a comparable category where better data exists, and scale by a reasonable driver such as the number of practices or GDP.
When three approaches built on different evidence land in the same neighborhood, you can present the number with real confidence. When they don’t, you have learned which part of your understanding is weakest, which is worth knowing before anyone else points it out.
Present a range, and show what drives it
A single precise figure invites false confidence. Present a low, base and high case, and say which two or three assumptions move the result most. In the bottom-up example above, adoption rate and price per seat drive almost everything; the clinic count is comparatively well known.
A short sensitivity table helps: “If adoption is 20% rather than 30%, the market is one third smaller.” It tells your audience where to direct their skepticism, and it shows you’ve already thought about it.
Keep a source log
The single habit that makes estimates defensible is keeping a log of every number you use, at the moment you use it. A simple table works:
| Field | What to record |
|---|---|
| Figure | The value exactly as published, with its unit |
| Definition | What it counts and what it excludes, in the source’s own words |
| Period and geography | The year or quarter, and the region it covers |
| Source | Publisher, document title, page or table number, and URL |
| Accessed | The date you retrieved it |
| Adjustments | Any conversion, inflation adjustment or filter you applied |
| Confidence | Your own rating, with one line of reasoning |
Record the figure as published before you convert it. If you later discover a currency or unit mistake, you want the original to go back to. Store secondary sources (a news article quoting a report) separately from primary ones, and try to replace the former with the latter before you present.
A checklist before you present
- The market definition fits in one sentence and every input matches it.
- TAM, SAM and SOM are each defined, and SOM rests on something concrete.
- The estimate has been built at least two independent ways, and the gap is explained.
- Every input has a row in the source log, with a date and a link.
- The result is shown as a range, with the main drivers named.
- Hypothetical or assumed inputs are labeled as such, everywhere they appear.
Where research tools fit
Most of the time in a sizing exercise goes into finding inputs: registry counts, survey results, price pages, vendor disclosures, and the definitions behind each of them. That is the part where a research assistant can save hours.
Divrite’s deep research is built for that stage. It breaks a question into sub-questions, searches in several rounds and in the languages of the countries involved, and attaches a source to every figure it reports, so you can open each one and copy it into your log. It doesn’t replace the judgment calls above. Choosing the definition, deciding which filters are justified and reconciling the two methods is still your job, and it’s the part your audience will ask about.