TL;DR
A market number is reliable when you can name the source that produced it, it is recent, and it measures exactly what you need, over a stated period, in your zone. Three pages repeating the same 2018 survey are one source from 2018, not three confirmations. Rate each number honestly and treat anything you cannot trace as an assumption.
A market number is reliable when you can name the source that produced it, it is recent, and it measures exactly what you need, over a stated period, in the zone you serve. If any of those is missing, the number is weaker than it looks, and you should say so next to it. Three pages quoting the same figure do not make it more reliable: a statistic from 2018 cited by three pages is still a statistic from 2018.
Why one bad number breaks everything after it
Market sizing is a chain of multiplications. A typical version for one customer segment needs three numbers:
- How many of them there are.
- What they already spend on the problem you solve, per year.
- The share that is ready to pay for a solution.
Multiply the three and you get a revenue potential. Then you build on it: how many customers you aim for, whether the project is worth your evenings, what you tell a partner or an investor.
If one of the three numbers is wrong, every conclusion downstream inherits the error, and nothing in the final total warns you. A wrong number looks exactly like a right one once it has been multiplied.
Find the source that produced the number
Most numbers you find online are copies. A blog quotes a report, a second blog quotes the first blog, a slide deck quotes the second blog. By the time you meet the figure, it has lost its date, its scope and its definition.
So the first job is to walk back up the chain until you reach the organisation that actually measured something: a statistics office, a trade body, a survey with a published method, a company reporting its own data. Then ask:
- Who measured it, and how? A survey of 200 people and a national census are not the same kind of number.
- Can you open it? If the trail ends at "according to studies" or at a dead page, you have not found a source.
- Is it really there? Check that the figure appears in the original, with the same value. Copies drift.
This matters even more if you research with an AI assistant. In an international study of more than 3,000 answers given by AI assistants to news questions, 45% contained at least one significant issue, and 31% had problems with their sources (EBU and BBC, 2025). An assistant is a good way to find leads. It is not where a number's reliability comes from.
Check four things on every number
Once you hold the original, check four things before you use it.
1. The date. When was the data collected, not when was the article published? A 2024 post quoting a 2018 survey is a 2018 number. Markets change, prices change, habits change. Past about three years, treat a figure as indicative at best.
2. What it measures. "Spending on software" is not "spending on scheduling software". "Small businesses" may mean fewer than 10 employees in one source and fewer than 250 in another. Read the definition, and compare it with the segment you are actually counting.
3. The period. An amount without a period cannot be used. "Freelancers spend €25 on tools" could mean per month, per year or per project. You cannot turn it into a yearly figure without guessing, and the guess can be off by a factor of 12.
4. The zone. A worldwide figure does not describe one country, and a US figure does not describe France. Use numbers for the zone you will really serve in the next 12 to 24 months, or scale them down and say so (more on that in how to choose the zone you size).
Rate each number, out loud
You will rarely have three perfect numbers. That is fine. What is not fine is mixing a census figure and a guess in the same multiplication without saying which is which.
A simple scale is enough:
- Highly reliable. A named, recent source that measures exactly your segment.
- Reliable. A named source with a small gap: slightly old, slightly wider definition.
- Unreliable. Old, indirect, or a definition that only roughly matches.
- Assumption. No named source. Your own estimate, or a number you could not trace.
Two rules keep the scale honest. First, a number with no named source is an assumption, whatever its precision: "37.4%" with no source is not better than "about a third". Second, when in doubt, rate down. People, and models, tend to overestimate how sure they are.
A worked example: one segment, three numbers
Take a tool that helps freelance translators manage quotes and invoices, sold in one country. All the figures below are invented for the example.
You collect three numbers for the segment "freelance translators registered as self-employed in the country":
- Count: 12,000. From the national statistics office, data from last year. Highly reliable.
- Spend: "€25 on admin tools". From a blog post that cites no survey and gives no period. Unusable as it stands.
- Share ready to pay: 30%. From a 2018 survey, quoted on three different websites. Unreliable.
Here is why the spend figure cannot simply be plugged in. If €25 is per month:
12,000 × 30% × (€25 × 12 months) = 12,000 × 0.3 × €300 = €1,080,000 a year
If €25 is per year:
12,000 × 30% × €25 = 12,000 × 0.3 × €25 = €90,000 a year
Same three numbers, a result 12 times larger or smaller depending on a word the source left out. A potential of about a million a year and one of €90,000 lead to completely different decisions. So the number goes out until you find a version with a period, or you replace it with your own estimate, clearly labelled.
Suppose you find a trade body survey from two years ago stating that freelance translators spend €180 a year on admin software. That is reliable: named, recent enough, close to your definition. The calculation becomes:
12,000 × 30% × €180 = 12,000 × 0.3 × €180 = €648,000 a year
Note what has not changed: the 30% still comes from 2018. Written honestly, the line reads "€648,000 a year, with a share ready to pay taken from a 2018 survey". That sentence tells the reader exactly where to push.
What this does not cover
These checks tell you whether a number deserves your trust. They do not tell you whether your market is big enough, or what share of it you can realistically win. That is a separate question, and a reliable total can still hide an unrealistic plan.
They also do not replace talking to customers. A survey says what people answered, not what they will pay you. A handful of conversations with people in your segment will often tell you more about the share ready to pay than any published figure.
And they do not make an assumption disappear. Sometimes the right answer is "no reliable figure exists for this". Write your estimate, label it, and move on.
How Banast does it
In the market chapter of a Banast report, each segment gets the same three numbers: how many they are, what they already spend on the problem, and the share ready to pay. Each one shows a reliability badge (Highly reliable, Reliable, Unreliable, Assumption) and a drawer with its source, what it measures and its limit. You can set a number aside, enter your own or run the search again.
The model suggests a confidence level, and the code can only lower it, never raise it: a source more than three years old cannot get the top rating, and a number with no named source becomes an assumption. An amount with no period is excluded, because turning it into a yearly figure would mean guessing.
FAQ
How old is too old for a market statistic?
There is no universal cut-off, but past about three years you should treat a figure as indicative rather than solid, and more so in a market that moves fast. If you keep an old figure, show its year next to it so nobody reads it as current.
Can I use a number that ChatGPT or another AI assistant gave me?
Use it as a lead, not as a source. Ask where it comes from, open that page yourself and check that the number is actually there, with the date and scope you were told. If you cannot find it, the number is an assumption.
What if I cannot find any reliable figure for my niche?
Write your own estimate, show how you built it and label it as an assumption. A clearly labelled guess is useful because everyone knows how much weight it can carry. The danger is a guess dressed up as a statistic.
Adrien Lebas
Founder of Banast, building it alone from Avignon. Writes about what it takes to make a living from what you build.