How long should a free trial be if abuse is a problem?
Cutting a free trial in half brings repeat claims back twice as often. How to set trial length from time to first result and cost per trial, with the arithmetic.
Is our free trial too long? It cost you $13,860 last quarter, so the investor update going out this evening says yes. You've got the good part written: "1,200 trial starts last quarter, up 20% from 1,000. 8% converted to paid." Under it sits the one decision you're announcing: next month the free trial drops from 14 days to 7.
A support email arrives while you're rereading that line. "My trial ran out again, can you send me a new link?" You pause on "again" and search the inbox for the name. Six threads come back, on six trial accounts, each started the day the last one ended. By 5:35 you've found eight more people like that in 90 days of support email, and those are only the ones who wrote in. You open a new tab and type free trial length abuse, because the announcement in your draft suddenly looks different.
The numbers are invented, so use your own. Say the real count is 60 people, each running 6 back-to-back trials in a 90-day quarter (90 ÷ 14 = 6 full trials). That's 360 of your 1,200 trial starts, and 300 of them are repeats. The other 840 are one person, one trial.
| What your update says | What's underneath |
|---|---|
| 1,200 trial starts | 900 people (840 + 60) |
| 8% convert (96 ÷ 1,200) | 10.7% of people convert (96 ÷ 900) |
| Up 20% from 1,000 | Unknown, because nobody counted people last quarter either |
Your real conversion rate is better than the one you've been reporting. That feels like good news for about a minute, until you remember the ad budget and the pricing test you set from that 8%. Each of those was a decision about 1,200 people who turned out to be 900.
The money is simpler. A day of real use costs you $1.50 in model calls and compute, and these accounts use every day they get.
- 14 days × $1.50 = $21 a trial
- 300 repeat trials × $21 = $6,300 a quarter
- $6,300 × 4 = $25,200 a year
If you spend $25,000 a month, that's a full month of runway. Your 96 new customers at $49 pay you $4,704 a month (96 × $49), so it takes about 40 days of everything they pay just to cover those repeat trials ($6,300 ÷ $4,704 × 30). And those 60 people draw on the same model quota as everyone else. Say that on one Tuesday in August the quota ran out at 2:10 p.m. Your paying customers got errors for 50 minutes, you refunded three of them ($147), and you spent the next morning hunting for a bug that wasn't there.
At 5:48, twelve minutes before it's due to go out, you're still looking at the draft. You can send it as written, or you can tell your investors that the number they've watched for three quarters doesn't mean what any of you thought. Either way there's still that line about the 7-day trial, which is about to make this worse.
Why a shorter free trial brings repeat claims sooner
Someone on a trial loop wants free access for as long as they can keep it. The length of the trial only decides how often they have to come back for another one. Cut it to 7 days and the same 60 people each need 12 trials a quarter instead of 6:
- 90 ÷ 7 = 12 full trials each
- 60 × 12 = 720 trials
- 7 days × $1.50 = $10.50 a trial
- 720 × $10.50 = $7,560
At 14 days, those 60 people cost 360 × $21 = $7,560. That's the same amount to the cent, because you're paying for days of use, and trial length isn't in the sum.
Bars show one fully used trial at $1.50 a day. The line shows one repeat user across a 90-day quarter, counting full trials only. Some people will give up when the trip back comes twice as often. Anyone comfortable opening an account every two weeks can open one every week.
So who feels the cut? The 840 people actually trying your product. Say they use 5 of their 14 days today ($7.50 each, $6,300 in total) and 3 of 7 after the change ($4.50 each, $3,780). You save $2,520 a quarter, all of it from people who might have bought. Now say a quarter of your buyers run out of time before they get a result, so 96 customers become 72. That's 24 × $49 × 12 months = $14,112 a year in revenue, gone from one quarter's trials.
Your next update would read 1,560 trial starts (840 + 720), up 30%, with 4.6% converting (72 ÷ 1,560). You'd be explaining why conversion nearly halved, and you still wouldn't be able to show that it was never really 8%.
How to choose a free trial length
Start from what the trial is for: getting a likely customer to a real result, with enough time left to decide. Cost comes second.
Find when paying customers got their first real result. Pick the moment your product proved itself: a first report, a first deploy, a first finished file. Measure paid accounts only, so repeat trials can't move the number.
select percentile_cont(0.75) within group (
order by extract(epoch from first_result_at - trial_started_at) / 86400
) as days_to_first_result
from accounts
where converted_to_paid = true;That's the day by which three out of four paying customers got there.
Add time for the decision. Someone on a team has to show a manager or get a card approved. Treat this as a starting rule:
If the first result takes 4 days and approval takes 3, that's 11 days, so a 14-day trial fits. Doubling leaves room for a busy week.
Price it per customer. Cost per trial is active days × cost per active day. Divide the total by customers won. In the scenario, the 840 honest trials cost $6,300 ÷ 96 = $65.63 per customer, about 1.3 months of a $49 plan. Add the 60 repeat users' $7,560 and it's $13,860 ÷ 96 = $144.38, nearly three months. If the first figure is fine and the second isn't, a different length won't help.
Cap the expensive part before you cut days. If a used day costs too much, limit what a free user can spend. The person who needs 11 days still gets 11 days.
Change one thing and count people. Run the new length on new trials for a full cycle, then compare customers per person, since fake signups distort the rate per trial.
The basic fixes for repeat trials, and how they break
You still have the same 60 people. Each of these basics catches something.
Email verification and unique emails. Keep both. They catch typos, dead addresses and the same address used twice. They miss any second real inbox, because a verified email only proves someone can open a mailbox. Gmail also delivers an address with its dots moved to the same inbox, and mail with a plus sign and a word added lands there too. A rule to catch that needs upkeep of its own. Real people get stuck when the code lands in spam.
Disposable-domain blocks. They stop the throwaway services on your list. A new domain isn't on it, and ordinary mailboxes never were. People using privacy forwarding get turned away, and someone has to update that list for as long as you run trials.
CAPTCHA. It stops simple scripts. Someone opening an account by hand isn't what it was built for. People using screen readers pay for it, and you pay in lost signups.
A phone number or a card up front. This is the strongest of the basics, and it does put off casual second accounts. It misses anyone with another number or card. It also turns away students and anyone who wants to try before paying, and you're billed for every text.
One trial per IP address, and VPN blocks. These catch repeat signups on one connection and known VPN ranges. Everything on another network gets through. Offices and households share addresses, and people on a company VPN include your buyers. Each new block brings exceptions and support tickets.
Usage caps and rate limits. Caps are the one basic that lowers your cost per trial, so set them. A new account gets a fresh cap, though, and a rate limit never notices 60 people staying under it. A real buyer having a heavy first day hits the same wall.
Manual review. It found those eight people this afternoon. At 2 minutes a trial it's 40 hours a quarter (1,200 × 2 = 2,400 minutes). It finds whoever happens to write in, and it can flag whoever looks odd to a tired reviewer.
Stack all of that and a stranger clears five hurdles before seeing your product, which pushes out the time to first result your trial length depends on. Every list and exception then needs someone to keep it working, and the person on a trial loop simply comes back with a new form.
Let trial length go back to being a product decision
Every fix in that list looks at one thing, and one thing is easy to replace. AbuseZero was built for the moment a trial is requested and you need to know whether this claim looks like a repeat. Its multi-layer AI risk analysis examines 140,000+ signals, reads them together and never one at a time, detects linked claims and flags abuse that single checks like those miss. AbuseZero uses a proprietary algorithm, with AI that checks the claim again. AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing, and it checks mailbox aliases and disposable domains as well. It sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. Creating a fresh account isn't enough to erase previous activity, so you can pick 7 days or 30 for what they do for your customers.
Your server makes one API call before a trial starts and gets a clear allow or deny, under rules you control. Eligibility can be once per user, or once every 90 days. Observe mode shows you what would be denied before anyone is blocked. You don't have to build your own abuse prevention system, and the quickstart protects your first benefit.
Frequently asked questions
Is a 7-day or a 14-day free trial better?
Use whichever covers the first real result plus the time to get approval. If three out of four paying customers get a result by day 2 and nobody else has to sign off, 7 days is plenty.
Does a shorter trial reduce abuse?
It lowers the cost of each trial and raises how often the same person needs a new one. Days of use, and so your cost, barely change for anyone who keeps coming back.
Should I report trial conversion per trial or per person?
Per person wherever you can, with repeat trials on their own line. Measuring your trial abuse rate gives you a place to start.
Keep your trial for people deciding whether to buy
Start with Builder
Pay as you go. Evaluate your first trial claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.