Selling to banks in MENA, and the regulator in the room
Selling to banks in MENA means clearing the central bank first: the no objection clock, where customer data has to live, and the clauses you will sign.
Read
BDGL / Insights / Entrepreneurship
Three interviews is not validation and forty is not rigour. There is empirical research on where saturation actually lands, one condition attached to it that B2B breaks, and a question habit that decides whether any of it was worth running.
A founder tells me they validated the idea. I ask how many people they spoke to. The answer is three, and two of them were friends who run companies in the right industry. That is not validation. It is three conversations, and the useful part of them was probably the bit nobody wrote down.
The opposite failure is rarer and more expensive. Forty interviews, a spreadsheet of quotes colour coded by theme, nine months gone, and nothing built. Both failures come from the same place: nobody knows what number is enough, so the number gets picked by temperament. Impatient founders stop at three. Careful ones never stop.
There is research on this. It is not B2B research, which matters and we will come to it, but it is real empirical work rather than a consultant's rule of thumb, and it gives you a defensible starting number.
The relevant paper is Monique Hennink and Bonnie Kaiser, Sample sizes for saturation in qualitative research: A systematic review of empirical tests, published in Social Science and Medicine in 2022. They went looking for every study that had empirically tested when interview research stops producing new information, and found 23: seventeen using empirical data and six using statistical modelling.
The headline finding is narrower than you would expect. In their words, studies using empirical data "reached saturation within a narrow range of interviews (9–17) or focus group discussions (4–8)". That is the convergence across varied topics and varied methods, which is the part that makes it worth quoting rather than the midpoint.
Saturation here means something specific: the point where additional interviews stop surfacing new codes or themes. It is not the point where you are confident, and it is not the point where you have enough quotes for a deck.
The finding carries a qualifier that is easy to skip and expensive to skip. The 9 to 17 range held "particularly those with relatively homogenous study populations and narrowly defined objectives". The authors are explicit that the outliers, including multi-country research and studies chasing meta-themes, needed larger samples.
Now look at a B2B buying situation honestly. You are not talking to one population. You are talking to the person whose budget it is, the person who will use the thing daily, and often a third who has to approve it on security or compliance grounds. Those three do not share a problem, a vocabulary or a success measure. A finance director describing a reconciliation problem and an operations supervisor describing the same process are two populations wearing one logo.
So the practical translation is not nine to seventeen interviews. It is nine to seventeen per group you are actually trying to understand, and the first job is deciding how many groups there are. For most early stage B2B products the honest answer is two, sometimes three. That turns a vague forty into a concrete twenty to thirty with a reason attached, and it tells you when to stop.
If your segments are still a guess, the interviews are the wrong first move. Narrow the target first; the four tests in an ideal customer profile for outbound are the cheapest way we know to do it, and they stop you interviewing a market you were never going to sell to.
Sample size is the part founders argue about. Question design is the part that actually decides whether the interviews were worth running, and it comes down to one habit: ask what someone did, not what they would do.
People are unreliable about hypothetical futures and reliable about concrete pasts. "Would you pay for this?" is a question about a stranger's imagined self, and the polite answer is yes. "What did you spend on this problem last quarter?" has a number attached, and the number either exists or it does not.
| Instead of | Ask | Because |
|---|---|---|
| Would you use a tool that did this? | Walk me through the last time you did this manually. | A past process has steps, owners and a duration. A hypothetical has none. |
| Is this a big problem for you? | What did you try already, and why did you stop? | Abandoned attempts are the best evidence a problem is real and unsolved. |
| Would you pay for it? | What is the budget line this would come out of, and whose is it? | Moves the question from willingness to mechanics, where the truth lives. |
| What features would you want? | What happens today when it goes wrong? | Feature requests are solutions from people who have not seen the options. |
| Do you like it? | Who else would have to agree before this got bought? | Surfaces the buying group early, while it is still cheap to learn. |
The right column is the reason the swap works, and it is worth keeping in front of you during the call. Under pressure, everyone drifts back to the left column, because the left column gets agreeable answers and agreeable answers feel like progress.
A sales discovery call qualifies a specific buyer against something you already sell. A customer discovery interview is research into whether a problem is worth solving at all. The questions overlap enough that founders run one meeting and believe they did both.
They did neither. The moment you start selling, the person stops telling you about their process and starts managing your expectations, because the conversation now has a price in it. The moment you start researching, the qualified buyer in front of you wonders when you are going to tell them what you do.
Keep them separate and run them in different meetings with different openings. When you are selling, the eight sales discovery questions that earn their place are the ones to work from. When you are researching, you are not allowed to pitch, and the discipline of that is most of the value.
Interviews produce talk. The decision to build needs something firmer, and there is a usable ladder of evidence.
A pre-commitment is the only one of these that costs the other person something, which is exactly why it is worth more than the rest combined. It does not have to be money. A written agreement to give you two hours a week during a pilot is a real cost to a busy operator.
Once a problem has cleared that bar, the cheapest next step is usually not a build either. A clickable prototype before development settles arguments that another ten interviews will not, and it does it in days. Our own early clients came out of this sequence rather than out of a marketing push, which is roughly the order described in how to get your first ten B2B clients.
The research is not about B2B buyers. Hennink and Kaiser reviewed qualitative health and social research. Nobody has run the equivalent systematic review on business to business product discovery, and we are borrowing a finding across contexts. We think the borrowing is defensible, because saturation is a property of how fast a population stops producing new information rather than of the subject matter, but it is a borrowing and you should know that it is.
Nine to seventeen is a stopping rule, not a target. Hitting seventeen interviews proves nothing if the first fifteen were with people who cannot buy. Sample size discipline is worth less than sampling discipline, and the order matters: segment, then recruit, then count.
Saturation on problems is not saturation on willingness to pay. You can reach the point where no new themes appear and still have no idea what anyone will pay, because price is not a theme. That question is answered by asking for money, not by asking more questions. If the number itself is the open question, how to price a B2B service without guessing is the better page.
None of this rescues a decision already made. Founders who have chosen to build will find support for it in any twenty interviews, and interviews run to confirm are worse than no interviews, because they come with evidence attached. If you cannot name in advance the answer that would stop you, you are not doing discovery.
Send us the segments you spoke to, how many conversations each one got, and the workaround people described. We will come back with where the evidence is genuinely thin, which group is under-sampled, and the one pre-commitment worth asking for before anything gets built.
Book a Free Consultation