Free Trial to Paid Conversion: Ask the People Who Didn’t Buy
Your funnel dashboard shows exactly where trials die and tells you nothing about why. Twenty phone calls, sorted into four buckets, will.
Every team with a trial-conversion problem has the same artefact: a funnel chart with a cliff in it. Sixty-two percent make it to step three. Nineteen percent make it to step four. Everyone stares at the gap between those numbers as if staring harder will explain it.
It will not. A funnel chart is a record of where people stopped. The reason they stopped is not in the data — it was in their head, on the day, and the only way to retrieve it is to ask them.
The good news: non-converters are the easiest people in SaaS to get on the phone. They have no relationship to protect, no sunk cost to defend, and no awkwardness about having left. In our experience they respond at higher rates than churned paying customers.
The sequence: instrument the drop-off, split non-converters into four buckets, call five from each, and fix the bucket that pays. Averaging across the buckets is what produces the useless conclusion that “people want it cheaper.”
TL;DR
- Median free-trial-to-paid conversion sits around 8% across self-serve SaaS, with a roughly 10x spread between the top and bottom quintiles (Kyle Poyar / ProductLed / ChartMogul analysis).
- Trial drop-offs are the largest untapped feedback source in most SaaS companies, and the easiest segment to reach.
- Four buckets: never started, started-and-stalled, used-it-didn’t-buy, wrong fit. Each has a different fix and mixing them produces mush.
- "Never started" is a marketing and onboarding bug. "Used it, didn’t buy" is packaging or authority. Neither is a feature gap.
- Twenty calls — five per bucket — is enough to know which one you have.
What Good Actually Looks Like
~8%
Median free-to-paid conversion, self-serve SaaS
~2x
Free-trial conversion vs. freemium, roughly
10x
Spread between top and bottom quintile products
4
Distinct failure classes hiding inside one number
The number worth internalising is not the median — it is the spread. Kyle Poyar's analysis with ProductLed and ChartMogul found roughly an order of magnitude between the best and worst self-serve products. That range is far too wide to be explained by product category or price point. It is explained by whether teams know why their trials fail.
Benchmarks are also the most misused number in this discussion. “We're at 6%, the median is 8%, so we need to improve by 2 points” is not a plan. It does not say which of the four failure classes you have, and the four classes are fixed by four different teams.
A single conversion rate averages four unrelated failures into one number. Chasing the number without splitting it is how teams end up running a discount experiment to solve an onboarding bug.
Why Funnel Analytics Cannot Answer This
Analytics is excellent at one thing and structurally incapable of another. It tells you where, with precision. It cannot tell you what the person was trying to do, which is the only input that determines the fix.
Take a real-shaped example. Your data says 41% of trials abandon at workspace creation. Possible explanations, all consistent with that number:
- The form asks for a company domain and half your signups are consultants on Gmail.
- People do not understand what a workspace is in your product's vocabulary.
- They wanted to look around first and the step is mandatory.
- The step is fine and they got interrupted, then never got a reason to come back.
- They realised at that moment the product was not what the landing page implied.
Five explanations. Three completely different owners — engineering, copywriting, and lifecycle marketing. Session replay narrows it slightly; it shows you the screen, never the intent. Only a conversation separates them, and it takes about four calls.
“To design an easy-to-use interface, pay attention to what users do, not what they say. Self-reported claims are unreliable, as are user speculations about future behavior.”
Worth being precise about how this applies. Nielsen's rule warns against asking users to predict or opine — “would you pay for this?” is a fantasy question. It does not warn against asking what someone actually did on a specific Tuesday. Combine the two: use analytics to find the moment, then ask the person to narrate that moment. Neither instrument works alone.
The Four Buckets (Do This Before Calling Anyone)
Sort your last 60 days of non-converting trials into four groups before you dial a single number. This step is the whole method — the buckets each have a different cause and a different owner, and interviewing them as one population produces averaged nonsense.
| Bucket | Definition | What you are testing | Likely owner |
|---|---|---|---|
| Never started | Signed up, <2 sessions, no core action | Was the promise wrong, or the first five minutes? | Marketing + onboarding |
| Started, stalled | Hit the core action once, then stopped | Where did effort exceed payoff? | Product |
| Used it, didn’t buy | Sustained usage, no conversion | Price, budget authority, or timing? | Commercial |
| Wrong fit | Usage pattern unlike any paying customer | Is acquisition selling to the wrong people? | Marketing |
Five calls from each. Twenty total. You will usually know which bucket dominates before you finish, and the distribution itself is the finding — more so than anything any individual said.
“Wrong fit” is the cheapest win in the set and the one teams resist most. It does not need a product change. It needs a channel turned off — and turning off a channel that is producing signups feels like going backwards, right up until you measure conversion instead of volume.
What To Ask Each Bucket
Never started
“What made you sign up that day — what were you in the middle of?”
“What did you expect to see when you got in?”
“What did you actually see?”
“What did you do instead?”
Listening for the gap between the acquisition promise and the empty state. Nearly always a marketing or onboarding bug — almost never a feature gap.
Started, stalled
“Take me through what you did after you signed up.”
“Where did it stop being worth the effort?”
“What were you trying to accomplish at that point?”
“What would have made you push through?”
Ask them to narrate rather than summarise. “And then what?” is your main tool here — the moment cost exceeded payoff has a specific location, and narration finds it while summary hides it.
Used it, didn’t buy
“It looked like you got real use out of it — what for?”
“What happened when the trial ended?”
“Who else would have had to say yes for this to become a line item?”
“If the price were half, would you have bought it? What about a tenth?”
The third question is the one that surprises people. A large share of this bucket is not a price objection at all — it is a user with no budget authority and no internal champion. That is a completely different fix.
Wrong fit
“What were you hoping this would do?”
“Where did you first hear about us?”
“What kind of work do you actually do day to day?”
You are looking for the channel producing people your product was never built for. The second question is the load-bearing one.
Two questions to ask everyone regardless of bucket: “What are you using instead?” and “What would have had to be true for you to be paying us today?” The second is the same counterfactual that anchors the six-question churn script, and it is just as load-bearing here.
Trial Drop-off Interview Guide
The bucket definitions, the opening line, question sets for all four buckets, the logging sheet, and how to read the distribution.
- Bucket definitions with the instrumentation each one needs
- The opening line for non-converters (easier than churned customers)
- Four question sets, one per bucket, with what to listen for
- The two questions to ask everyone
- An eight-field logging sheet and how to read the bucket distribution
Reading the Results Without Averaging
After twenty calls, count the buckets, not the reasons. This is the step where most teams undo their own work — they tally the stated reasons across all twenty calls, find “price” appearing six times, and go run a pricing experiment.
| Dominant bucket | What it means | What to do | What NOT to do |
|---|---|---|---|
| Never started | Problem is upstream of the product | Fix the landing-page promise and first-run | Build anything |
| Started, stalled | One specific step is too expensive | Instrument it, then remove the cost | Add a feature elsewhere |
| Used, didn’t buy | Packaging or authority, rarely price | Look at who signs vs. who uses | Discount |
| Wrong fit | Acquisition targeting problem | Turn off a channel | Build for them |
The right-hand column is there because each of those wrong moves is the intuitive one. Discounting a “used it, didn't buy” population feels like the obvious response to price objections, and it converts a handful of people while training your market to wait for a discount — and it does nothing about the actual constraint, which was usually that your user could not get a purchase order signed.
The Fix For Each Bucket
Never started
Cut the distance between the promise and the first useful moment. Concretely: make the empty state do something, defer every mandatory setup step that is not required to deliver value, and check that your landing page's headline describes what the first screen actually does. This bucket is fixed with copy and defaults far more often than with code.
Started, stalled
Find the specific step where cost exceeded payoff — the calls will name it — then reduce the cost or increase the perceived payoff at exactly that step. Not generally. At that step. This is the highest-leverage fix in the set and also the one that requires the most precision.
Used it, didn't buy
Separate the three sub-causes before acting: genuine price sensitivity, absent budget authority, and wrong timing. Only the first is fixed by pricing. The second is fixed by giving your user the material to sell internally — an exportable summary of what they did in the trial does more than a discount. The third is fixed by a follow-up sequence with a date on it.
Wrong fit
Turn the channel off, or qualify harder at signup. Painful, immediate, and it improves every other number in your funnel because you stop diluting the denominator. It also makes the other three buckets easier to read next quarter.
Twenty calls, one dominant bucket, one fix, then measure. Then twenty more. Teams that do this quarterly compound; teams that do it once as a “trial conversion project” get one improvement and go back to guessing. The mechanics of running it as a standing loop are in the churn feedback loop framework.
Frequently asked questions
- What is a good free trial to paid conversion rate?
Around 8% is the commonly cited median for self-serve SaaS, with free-trial products converting roughly twice as well as freemium ones. But the spread between the top and bottom quintile is close to tenfold, which makes the median a weak target. A more useful question is which of the four failure buckets dominates yours — that determines how much headroom you actually have.
- Should we require a credit card for the trial?
It moves the problem rather than solving it. Requiring a card raises conversion rate and lowers trial volume, because it filters at signup instead of at the end. Whether that is good depends entirely on which bucket dominates: if you are full of “wrong fit”, a card requirement helps; if your problem is “never started”, adding friction at signup makes it worse.
- How do we reach people who never really used the product?
Better than you expect. Non-converters have no relationship to protect and no awkwardness about leaving, so a short, honest, no-pitch email or call gets a higher response rate than the equivalent outreach to churned paying customers. The framing that works: “you tried this and didn't stick with it, and you're the person whose opinion I can't get any other way.”
- How long should a free trial be?
Long enough to reach the first genuinely useful moment, and no longer. Most teams pick 14 or 30 days by convention rather than by measuring when their converting users hit value. If your converters activate on day two, a 30-day trial is just 28 days of forgetting you exist. The bucket data tells you this directly.
Sources & further reading
- 1What is a good free-to-paid conversion rate? — Lenny’s NewsletterKyle Poyar’s analysis with ProductLed and ChartMogul across self-serve SaaS products.
- 2First Rule of Usability? Don’t Listen to Users — Nielsen Norman GroupWhy predictions and opinions are unreliable, and behavioural recall is not.
- 3SaaS Retention Report — ChartMogul
- 4The Product-Led Growth Method — Wes Bush — ProductLed
Keep reading
Twenty calls answers it. Nobody has time to make twenty calls.
saasfeedback.ai detects trial expiries from Stripe, calls the drop-offs, and returns the bucket distribution with transcripts and themes. Book 15 minutes and see it on real data.
Book a demo