TLDR: Almost nobody chooses a trial length. They inherit one. The input that should set it is your product's natural usage frequency, meaning how often somebody would come back with nothing nudging them. A trial has to cover two complete cycles of that, because once is a demo and twice is the start of a habit.
- Natural usage frequency is the gap between sessions when no notification is pulling the user back.
- Trial length = frequency x 2 cycles + a buffer to decide, with a floor of about a week.
- Daily product, 7 to 10 days. Weekly product, 3 to 4 weeks. Monthly product, closer to 90 days, which almost nobody offers.
- A 14-day trial hands a daily product 14 cycles, a weekly product 2, and a monthly product under half of one.
- Yearly and one-off products should sell an outcome for a one-time price. You are not building a habit there, you are delivering a result.
I made the three minute version of this argument as a video. The written version below has the calculator and the awkward cases.
Full transcript on the watch page.
Why does everybody run a 14-day trial?
Because somebody else did it first and it looked fine.
Every time I ask a founder how they landed on their trial length, the answer is a version of the same thing. It was on the pricing page of the last tool they signed up for. It's a tidy fortnight. It fits neatly inside a monthly billing cycle so the finance side never has to think about it. None of those are bad reasons exactly. They're just not reasons about your product.
This is cargo cult behaviour with a Stripe integration. When you copy a competitor's trial length you are importing their usage frequency, their activation curve, and their buyer's calendar, none of which you have any reason to share. Slack and Xero could copy each other's trial length in either direction and both would be wrong.
I'm not arguing 14 days is wrong. I'm arguing it wasn't a decision.
What is natural usage frequency?
Natural usage frequency is how often somebody would open your product on their own, with no email, no push, no Slack reminder from a colleague. It's the rhythm the job has, not the rhythm your lifecycle emails can manufacture.
It's one of the more useful ideas in retention work and it turns up under a few names. Casey Winters writes about it as natural frequency, Reforge builds a lot of its retention curriculum around it, and Andrew Chen has been making the same point about frequency setting the ceiling on engagement for years. The reason it matters here is simple: it's the clock your trial is running against.
Products sort into a handful of bands.
Daily. Messaging and anything where the work happens in the tool. Slack, WhatsApp, Linear, a customer support inbox. If somebody stops opening it for three days, something is wrong.
Weekly. Planning and coordination tools. Asana, most project trackers, sprint boards. You open them when the week starts, then again when it's falling apart.
Monthly. Anything tied to a close, a cycle or a report. Xero, QuickBooks, payroll, board reporting, invoicing. The product could be perfect and you'd still only think about it once a month, because that's when the job exists.
Yearly. Tax filing, annual compliance, renewals paperwork. TurboTax is the canonical one. Nobody does their taxes for fun in March.
One-off. Moving house, a visa application, probate, a data migration. Once a year if you rent, once a decade if you own, once ever if you're lucky.
Here's the test that gets you an honest number: look at customers who never open your emails and have push turned off. The gap between their sessions is your natural frequency. Everything else is measuring your marketing.
One caveat before anyone emails me. Notifications can nudge frequency, they can't invent it. A good reminder moves somebody from every four days to every two, because the need was already there and they were forgetting. Nothing you send turns quarterly compliance into a daily habit. Duolingo's work on streaks is the best public example of pushing frequency to its ceiling, and even that only works because the underlying job (practice a language) genuinely is daily.
How does usage frequency set trial length?
The rule I use: the trial has to be long enough for somebody to complete a full value cycle twice, plus a few days to decide.
Once is a demo. Twice is the start of a habit.
The difference between those two matters more than it sounds. The first time somebody uses your product they're supervised and curious. They're following your onboarding, they've set aside time for it, and they're evaluating you rather than working. The second time is the real test, because nobody scheduled it. They chose to come back, and your product had to survive contact with an actual week: the interruptions, the colleague who wanted it done in a spreadsheet, the fire on Thursday.
Once is a demo. Twice is the start of a habit. If your trial only pays for the first one, you're selling a tour rather than a tool.
The habit research backs the shape of this even if it doesn't back a tidy number. The "21 days to form a habit" line everyone quotes comes from Maxwell Maltz, a plastic surgeon writing in 1960 about how long patients took to adjust to a new face, and it was never a study. When Phillippa Lally's team at UCL actually measured it in 2009, the median time to automaticity was 66 days, with a range from 18 to 254. Wendy Wood's research points at the mechanism: repetition in a stable context is what does the work, not motivation and not duration on its own. BJ Fogg's behaviour model and Nir Eyal's hook cycle both land in the same place.
Your trial isn't going to finish a habit. Sixty-six days is longer than almost anyone gives away. What it has to do is buy the second repetition, because that's the one that proves the context is stable. Everything after that is your onboarding emails' job, not your trial's.
So the maths is boring:
Trial length = usage frequency x 2 + days to decide. Round up to a sensible calendar number, and put a floor of about a week on it, because a two-day trial is over before a human has had a normal work week.
| Frequency | Cycle | Trial that covers two cycles | What 14 days buys | Verdict |
|---|---|---|---|---|
| Daily | 1 day | 7 to 10 days | 14 cycles | Generous. You could shorten it and convert faster. |
| Weekly | 7 days | 21 to 30 days | 2 cycles | Survivable, but one bad week kills the trial. |
| Monthly | 30 days | About 90 days | 0.5 cycles | Broken. They never do the job once. |
| Quarterly | 90 days | About 190 days, so not really | 0.16 cycles | Use a paid pilot with checkpoints instead. |
| Yearly or one-off | 365 days or none | Not applicable | 0.04 cycles | Wrong model. Sell the outcome, one-time. |
Put your own numbers in:
Rounded up to the nearest number a billing system and an email sequence can both live with.
Get weekly churn reduction tips.
No spam, unsubscribe anytime.
You're in.
Check your inbox shortly.
Where this number actually comes from: it's a coverage calculation on the calendar, not a model of buying psychology. If your value event repeats every F days, then a trial of length T contains T divided by F complete cycles, and the trial only pays for the habit-forming second use when T is at least 2F. Add the days somebody needs to decide, because the decision happens after the second cycle rather than during it, and you get T = 2F + buffer. The floor of seven days exists because human weeks have a shape: a trial that starts on a Thursday and ends on a Saturday never touched a working week at all. The snapping to 7, 14, 21, 30, 60 and 90 is pure practicality, since your billing system, your reminder emails and your sales team all think in those units. Two things the maths deliberately ignores. It assumes the first cycle starts promptly, which is only true if your onboarding actually gets people to the value event, and it assumes the frequency is stable rather than seasonal. If either is false, fix that before you touch the trial length.
What does a 14-day trial cost a monthly product?
Let me make it concrete, because the abstract version doesn't land.
You sell a month-end close tool. Somebody signs up on the 5th, because that's when their last close finished and the pain was fresh. They poke around, import a chart of accounts, think "yeah, this looks right". Their trial expires on the 19th. The job they bought it for, the actual close, happens on the 1st through the 3rd of next month. They never did it. They never once used your product for the thing it exists to do.
They cancel. Your analytics record a failed activation. Somebody on your team opens a doc titled "improving onboarding".
That's the expensive part. It wasn't an activation problem, it was a calendar problem, and you're about to spend a quarter rebuilding a flow that was fine. This is the same category of mistake I wrote about in the retention experiment attribution problem: the number moved (or didn't) for a reason that isn't in the dashboard, so you learn something confidently false.
There's a second cost that's easier to miss. A trial that ends mid-cycle teaches your prospect that your product is something they set up rather than something they use. That framing is hard to undo later, and it shows up again at renewal as the vaguely embarrassed "we never really got going with it" conversation.
What if your product is only used once a year, or once ever?
Then stop selling a habit and sell the outcome.
This is the part of the video people argued with, so let me put it plainly. If somebody genuinely needs your product once a year, a monthly subscription is a bet that they'll forget to cancel. That's not a business model, it's a tax on inattention, and it produces a churn number that no amount of retention work will move, because the customer leaving is behaving correctly.
You're not building a habit, you're selling an outcome. Stop trying to turn a one-off payment into monthly recurring revenue.
The dashboard will lie to you about this in a specific way. You'll see high annual churn, low engagement between events, and a save flow that converts terribly. Every one of those looks like a retention problem and none of them is. It's a pricing model problem, and I'd rather you fixed the model than shipped another save flow that has nothing true to say.
What to do instead, roughly in order of how often it's the right answer:
- One-time price for the outcome. The filing, the migration, the report. Charge properly for it, once. Stripe handles one-off payments just as happily as subscriptions, and your conversion rate usually goes up because you've removed the "will this bill me forever" hesitation.
- Per-event pricing. Per filing, per move, per audit. Works when the frequency is irregular but non-zero, and it scales with the customer's actual usage rather than the calendar.
- A cheap custody tier. Only if there's real value in you holding their data between events: history, compliance records, the ability to pick up where they left off. This one is legitimate, but you have to be able to name the value out loud without wincing.
- Annual plan with a real deliverable in between. If you're going to charge yearly, ship something yearly. Quarterly reports, monitoring alerts, regulation change notices. Annual plans work when there's something to renew for.
The honest exception: some low-frequency products do have continuous value that the customer just doesn't see. Backups, security monitoring, uptime checks, compliance watching. The value accrues every day and the conscious visit happens twice a year. Subscriptions are correct there. But you then have a communication problem instead of a pricing one, and the fix is making the invisible work visible, which is the whole premise of why silent value gets churned.
What if you can't afford a 90-day free trial?
Nobody can, which is why almost nobody offers one. The move is to compress the cycle rather than extend the window.
Seed the account with their real data. If the job is month-end close, import last month's numbers on day two and run the close on data they already know the answer to. You just delivered a full cycle without waiting for the calendar. This is the single highest-leverage thing on the list and it's how the activation milestones playbook gets used most often.
Run the first cycle with them. A scheduled call where you do the job together. Expensive per customer, and worth it when the alternative is a 14-day trial that structurally cannot work.
Swap the free trial for a refund window. A 60-day money-back guarantee covers two cycles of a monthly product, gets the card on file on day one, and costs less than 60 days of free access because most people never claim. I went through the numbers on this in whether you should offer a money-back guarantee.
Use a reverse trial. Full product for a short window, then a permanently free tier rather than a locked door. Good when frequency is unpredictable across your user base, and it stops punishing the slow customers. The trade-off is the same one covered in whether you should have a free tier at all.
For high-value deals, run a paid pilot. Named checkpoints, a human attached, a decision date. An unattended 90-day trial for an enterprise buyer is a slow way to get ghosted, and Bessemer's Atlas is full of teams learning that the expensive way.
How do you measure your product's natural usage frequency?
1. Pick the value event, not the login
A login is somebody checking a box. The value event is the thing that, if it never happened, the customer would have no reason to pay: the invoice sent, the report generated, the deploy shipped, the message posted. If you can't name it in one sentence, that's the real work, and the aha moment guide is the shortest route to naming it.
2. Only look at customers who paid and stayed
Six months minimum. Trials and churned accounts will drag the number in both directions and tell you nothing about what a healthy rhythm looks like. You want the frequency of people your product is working for.
3. Strip out the nudged sessions
Look at sessions that didn't follow an email open or a push within the previous 24 hours. Or take the cohort with notifications disabled entirely, which is a small group in most products but an honest one. What's left is the pull, not the push.
4. Take the median gap between consecutive value events
Per customer, then the median across customers. The mean will be wrecked by one person who came back after eleven months. Watch the shape of the distribution too, because if you see two clusters you probably have two products glued together, and they may need two different trials.
5. Multiply by two, add the buffer, round up
Then compare it to what you're currently offering. If you're wildly over, you're paying for dead air at the end of every trial and finding out slower than you need to. If you're under by a whole cycle, you have an explanation for a conversion rate you've been blaming on the product.
Where does this break down?
Four places, and they're worth knowing before you go and change your pricing page.
Team products have two frequencies. One user opens it daily, the account as a whole does something meaningful weekly. Set the trial by whichever cycle the buying decision hangs on, which is almost always the account-level one.
Seasonal products lie in the off-season. A retail analytics tool measured in July has a very different frequency to the same tool in November. Measure across a full year or you'll build a trial for the quiet months.
Background-value products have no conscious cycle. Monitoring, backups, fraud detection. The value is continuous and the visits are rare, so a frequency-based trial measures the wrong thing entirely. Your trial has to manufacture a conscious moment instead: a report, an incident replay, a "here's what we caught this week" email.
Frequency changes with tenure. Plenty of products are daily during setup and monthly afterwards. That's fine, and it means the trial should be sized on the setup-phase frequency while your retention work should be sized on the steady-state one. Getting those two the wrong way round is how you end up with great trial conversion and terrible month-two retention.
What to do this week
Pull the median gap between value events for your retained customers. It's one query and it'll take an afternoon. Multiply by two, add a buffer, compare it to the number on your pricing page. If the two disagree by more than a cycle, you've found something worth more than most of the things on your roadmap.
And if the honest answer is that your customers need you once a year, take the harder decision. Price the outcome, drop the subscription, and stop measuring yourself against a churn benchmark that was never yours.
Work out your frequency first, and your trial length picks itself.
The companion piece to this one is how long should your free trial be, starting from time to value. Frequency tells you how wide the window has to be. Time to value tells you how quickly the first cycle can start. You need both, and they fail in different ways: a trial too short for your frequency loses people who would have bought, while a trial too long for your time to value just burns cash and attention. What happens at the end of the trial matters at least as much as the length, which is what reducing trial expiry abandonment and the trial expiry email sequence are for.
If you want to know whether trial length is even in your top three problems, the free 60-second churn health check asks a handful of questions and tells you where your biggest leak actually is. And if you want to see what those trial conversions are worth once they're paying, the MRR churn impact simulator runs it out over 24 months. Worth reading alongside Lenny Rachitsky's retention benchmarks, Amplitude's work on retention curves, David Skok on why churn is critical, a16z's 16 startup metrics and OpenView's product-led growth benchmarks, all of which treat frequency as an input rather than an afterthought.