How should a SaaS startup run win-loss interviews when it keeps losing to an established competitor?

Illustrative situation: a SaaS startup loses four deals in a quarter to the same established product. The sales lead blames price. The product lead points to a missing integration. The founder notes that every buyer already used the incumbent. Each reason is plausible, and each fits the job of the person giving it. Nobody asked the buyers. The scene is invented. It shows what this guide helps with: a loss reason written by the team that lost.

Humbleteam, a digital product design agency, publishes this guide. It covers whom to interview, who should run the call, consent and recording, a question guide, sample size, coding answers and separating product causes from sales causes, with an invented worked example. Sources are described as of November 10, 2026. The recording section is general information, not legal advice.

Why are your own loss reasons not enough?

The CRM field is filled in by the person who lost the deal. Buyers are polite to people who just sold to them: Clozd, a win-loss vendor, says buyers are more candid with a neutral third party, and Drive Research, a market research firm, says even talented reps struggle to get candid answers from buyers they just courted. Both have a commercial interest, so treat this as a claim, not a measurement. Memory also fails. Jakob Nielsen wrote for Nielsen Norman Group (NN/g) that people forget how they used a website and tend to invent stories that make what they remember sound logical.

Our guide to running a UX teardown of an established competitor studies the rival’s product. This guide studies the buyer’s decision.

Whom should you interview, and how soon?

Interview three groups: buyers who chose you, buyers who chose the incumbent, and teams that decided nothing. Drive Research recommends a mix of wins, losses and no-decisions. Clozd and Pragmatic Institute ask for equal numbers of wins and losses, since losses alone push teams toward overly negative decisions (Clozd) and wins alone skew the picture (Pragmatic).

Pragmatic says to interview within 3 months of the decision and that sooner is better, because implementation can overshadow memories of the evaluation. Our suggestion for short sales cycles: within 2 to 4 weeks. Ask for the people who shaped the decision, not only the signer (Drive Research). Our suggestion: separate calls with the person who ran the trial and the person who held the budget.

Who should run the call?

Not the account owner. Pragmatic says sales teams should not be solely responsible and suggests product team members. Drive Research and Clozd prefer a neutral outside firm. We rank no firm and recommend none. Our suggestion for a team without budget: a product manager or designer who never worked the deal. The sales lead reads the notes afterward, not before.

Open by saying who you are and that you are not selling. Offer no discount and make no promise about a contract.

What about consent and recording?

Ask in the invitation and again at the start of the call, so the yes is on the recording. For example: “Is it all right if I record this call for note-taking? Only our product team will hear it, and we delete recordings after 90 days.” The 90 days is our suggestion.

Federal law, 18 U.S.C. 2511(2)(d), allows recording when one party consents, unless the purpose is a crime or a tort (statute text). Some states ask for more. Florida’s statute 934.03, 2024 edition, makes recording lawful when all parties have given prior consent. The Digital Media Law Project guide lists California, Florida and Pennsylvania among all-party states and advises getting every party’s consent when callers are in different states. It says it is no longer updated, so check with a lawyer.

Our rule: get consent from every participant, every time, and take notes if anyone declines. We did not research rules outside the US, including privacy law.

Which questions belong in the guide?

NN/g defines a leading question as one that includes or implies the desired answer. Clozd’s sales example: “What did the sales team do poorly?” plants the answer, while “What was your interaction with the sales team like?” does not. Clozd also prefers open questions to yes-or-no or 1-to-10 ones. Pragmatic prefers standard scales for comparing interviews. We keep the call open and add structure afterward, in coding.

Ask what happened, not what the buyer would do, because Nielsen wrote that predictions and recall are weak data. When you follow up, repeat the buyer’s own words, as NN/g advises. If the buyer kept an evaluation sheet or email thread, ask to see it. That last habit is our suggestion.

How many interviews are enough?

NN/g’s Maria Rosala says there is no golden number and that five is often too few for exploratory work. Start with 5 or 6, analyze as you go, and add people while new themes keep appearing. The win-loss sources we read give no target number.

Our suggestion: start with 6, two per group, and extend to 12 or 15 if themes keep moving. At that size, count; do not compute rates. Say “3 of 6 non-wins,” not “50% of buyers.” Buyers who agree to talk are also not typical.

How do you code answers and separate product from sales causes?

NN/g describes thematic analysis as tagging observations and quotes with codes, then finding themes that recur across participants. Use one row per interview: outcome group, the reason in the buyer’s words, the evidence (an event they described, not an opinion) and one primary code. Wins use the same codes, read as the reason the buyer chose you.

Treat “price” as a prompt for one more question, not a finding. To split product from sales, ask two things (our suggestion). Could a trial user hit this with no salesperson involved? Then it is product and UX. Would a different rep showing the same product change the buyer’s account? Then it is sales. Two yeses mean two owners.

Rank codes by how many interviews name them as primary. Our threshold: 3 interviews across at least 2 outcome groups before a pattern earns roadmap time. Anything less is logged and checked against another source, such as the CRM.

Illustrative worked example: Quillmark and Dunmere

Everything here is invented. Quillmark is a fictional expense-approval startup. Dunmere is a fictional established finance suite. In one quarter Quillmark lost four deals to Dunmere. A product manager who never worked those deals interviewed nine buyers within 4 weeks of their decisions, with consent recorded on each call: three who bought Quillmark, three who chose Dunmere and three who decided nothing. A fourth Dunmere buyer declined.

Switching risk is primary in interviews 4, 5 and 7: 3 interviews across two groups, which meets the rule. Price came up as a side remark in 4 and 5. Asked what they compared it with, both named the hours needed to rebuild approval rules, so it was recoded. Single sign-on appears once, so it is logged and checked against lost deals in the CRM. The import problem appeared in the self-serve trial, which makes it a product and UX cause, not a sales one.

The team’s hypothesis: if the import step previews which approval rules will carry over before anything moves, the switching worry shrinks. They ship that one change and ask the same switching questions in the next six interviews. No result is claimed. Our guide to switching costs and migration UX covers how to design such a preview.

Where does this method fail?

• Small samples show patterns, not rates.
• A neutral interviewer lowers politeness, but answers can still be rationalized later. Clozd’s own material says blind interviews remove social pressure and add recall errors, since buyers may confuse vendors. We found no independent test of how candid win-loss answers are.
• Writing on win-loss comes mostly from vendors and training firms. The interviewing and sample-size guidance comes from NN/g’s general user research, not win-loss studies. We found no independent evidence that a program raises win rates, and we give none.
• The 2-to-4-week window, the six-interview start and the 3-across-2 rule are our suggestions, not standards.
• The legal notes rest on the federal statute, one state statute and one secondary guide that is no longer updated. We did not read the California or Pennsylvania laws themselves.

What can you produce now?

1. A list of deals closed in the last 3 months, sorted into won, lost to the incumbent and no decision, with decision dates and the people who shaped each one.
2. An invitation that asks for the call and for recording consent.
3. A guide of seven questions, each checked against the “Not this” column.
4. A coding sheet with the six codes and one row per interview.
5. A written rule for what counts as a pattern.
6. One change, one hypothesis and a date to re-check.

Decide which change goes first, and write down what you would expect to hear in the next six interviews if it worked.