Product-market fit (PMF)

Product-market fit is the state where a meaningful segment of people want your product badly enough that they pull it forward. Marc Andreessen coined the phrase in 2007: “the only thing that matters is getting to product/market fit.” The practical interview definition: PMF is present when users would be genuinely upset if the product disappeared, retention evidence confirms it, and the segment pulling hardest is large enough to support a real business. The most reliable failure mode in interviews is substituting acquisition evidence for retention evidence: citing a launch spike, a pilot NPS score, or a strong first cohort as PMF proof. Interviewers catch that and flag it.

The Sean Ellis test: the number and its real limits

Ellis benchmarked hundreds of startups on a single question: “How would you feel if you could no longer use this product?” Response options: Very disappointed / Somewhat disappointed / Not disappointed / N/A (I no longer use it). His finding: above 40% “very disappointed” correlated with sustainable growth; below 25% signaled major repositioning was needed.

That threshold is a useful heuristic, not a certification. Three limitations matter in an interview context:

• The benchmark was built on consumer SaaS. For B2B and enterprise, the survey breaks down because users who respond are not always the economic buyers. A power-user team can be “very disappointed” while procurement is consolidating vendors. Read renewal rates, net revenue retention, and shortened sales cycles instead.
• The sample must be active users who have reached core value. The test requires 30 to 40 responses from people who have actually experienced what the product does, not signups or casual dabblers. Running it on your full registered base inflates “not disappointed” responses and produces a false negative.
• A score of 42% from a segment of 200 is narrow PMF, not scaled PMF. The segment also has to be large enough to build a business on. Conflating the two is the senior-candidate trap.

Signals that actually indicate PMF

Treat these as a hierarchy. Retention evidence outweighs survey evidence. Expansion evidence outweighs retention evidence alone.

Strongest signals:

• Retention cohort curves that flatten and stabilize, with each successive cohort retaining at a higher rate than the prior. Not just “above zero” but a demonstrable upward trend across cohorts.
• Net revenue retention above 100% sustained for three or more months. Customers are deepening use, not just renewing. Expansion means the market is pulling.
• Organic referral volume that you cannot fully explain by paid spend or PR.

Secondary signals:

• Sales cycle shortening for inbound leads versus outbound. The market is coming to you.
• Support tickets shifting from “how does this work?” to “can you add X?” The latter signals workflow dependency.
• Burn multiple (net burn divided by net new ARR) below 1.0: the market is rewarding investment more than you are spending.

Weak signals that interviewers penalize:

• Acquisition spikes after a launch or press hit. Acquisition is not retention.
• High trial-to-activation numbers without week-two retention data. Novelty is not value.
• NPS above 50 from a pilot cohort. NPS measures satisfaction at a moment; it does not measure whether users have restructured their workflow around your product.

Anti-signals: why indifference is worse than anger

Almost no competing PMF definitions surface this clearly: indifference is a stronger anti-signal than anger. Users who hate your product are engaged enough to feel its absence. Users who are indifferent have no workflow dependency and leave without complaint. If your Ellis survey returns 60% “not disappointed,” the market is telling you the product is optional. That is a positioning failure if the segment is wrong, or a product failure if the segment is right and still does not care. No messaging fix resolves it.

42% of startup failures cite “no market need” as the primary cause (CB Insights, repeatedly cited through 2026). That is a PMF failure, not an execution failure.

Narrow PMF vs scaled PMF

The interview trap that eliminates senior candidates is claiming PMF too early. Narrow PMF means a small segment is genuinely obsessed. Scaled PMF means that segment is large enough to support the business model.

DoorDash validated narrow PMF in nine weeks: founders hand-delivered meals before building any tech, testing whether demand was real before investing in infrastructure. That proved the problem mattered to a segment. The scaling question was whether that segment was large enough and whether unit economics held at volume.

When interviewers ask “do you have PMF?”, they are asking about both dimensions. Miss the second and you will be asked to estimate TAM with no warning.

The 2026 reframe: viability and lovability are the remaining hard problems

Feasibility is now the floor, not the test. AI has collapsed the cost of building to near-zero for most software products. That means PMF is no longer a three-legged stool of viable, feasible, and usable. Basic usability has a strong default, and the contest is now fought on two dimensions:

Viability: is the problem worth paying to solve, at a price that supports a real business, in a market large enough to matter?

Lovability: does the product earn a durable place in how people actually work, not just a place in their trials?

A product that generates enthusiastic trials and weak retention has a lovability problem, not a feasibility problem. A product that retains a small segment well has a viability problem if the segment is too small. PMF in 2026 means clearing both bars.

For AI and agentic products, traditional PMF proxies lag. NPS and return-frequency metrics are calibrated for interface products. The relevant leading indicators for AI products:

• Task acceptance rate: do users act on AI output without revision, or do they consistently rewrite it? High acceptance signals trust; consistent rewriting signals the AI is generating effort, not saving it.
• Re-invocation rate: when a user finishes one task with your agent, do they bring it back for the next one in the same workflow? Re-invocation is the agentic equivalent of a flat retention curve.
• Rework rate: how often do users undo or override AI actions? Rising rework rate after the novelty window closes is a PMF failure signal.

These proxies matter because AI products are judged on whether users trust the output enough to act on it, not just whether they return to the interface. See lovable, not just usable and proving viability for how these connect to product decisions.

Weak vs strong interview answer

For how retention curves work as a diagnostic tool, see retention and cohort analysis. For connecting PMF evidence to your north star metric, see NPS.