How to measure your free trial abuse rate
Define the metric, run a 400-trial sample audit with every step of the arithmetic shown, and learn why the number you get is only ever a minimum.
You really believed it when you told investors on Friday that trial starts were up 50% in five months. Now a customer on your $49 plan has cancelled, and the reason box on the form is empty. You still answer these yourself, so you send two lines asking what you could've done better. Twenty minutes later the reply comes in: "Nothing, I like it. I've been on your free trial since March, five different emails. I only paid last month because I needed an invoice for a client."
You read it twice. Those five emails are five trial starts in your investor deck, and they were all one person, who has now left. By 6:30 you've got the spreadsheet open and you're searching for how to measure free trial abuse rate, which is probably how you got here.
This is a scenario, and so are all the numbers below, so swap in your own as you go. Your deck says trial starts grew from 2,000 in May to 3,000 in October, up 50% in five months. Then you go through both months using the audit further down. May held about 100 extra trials, from people who'd already had one. October held at least 750. Take them out and you started with 1,900 real trials at most and ended with 2,250. That's 18% growth (2,250 ÷ 1,900 = 1.18), and the deck says 50%.
Now follow that 18% through everything you decided on the 50%. In August you hired a second support person at $6,500 a month, because the forecast said you'd need one by Christmas. That's $78,000 a year. Each trial costs you $3.20 to serve, so October's extra trials cost $2,400 (750 × $3.20), which is $28,800 a year. Together that's $106,800. If your company spends $60,000 a month, you've put nearly 1.8 months of runway into a forecast built on users who never existed.
The conversion rate fooled you in the other direction. October's 120 new customers on 3,000 trials is 4.0%, and that number is why you spent two quarters rewriting onboarding and cut the trial from 14 days to 7. Divide by 2,250 and it's at least 5.3%. Your real users may have been converting well enough all along. January's price rise came out of the same spreadsheet.
Thursday's investor update is already drafted. It says 3,000. You know that's wrong, you don't know by how much, and the person who told you has cancelled.
Why trial metrics can't see repeat claims
Your analytics counts accounts because accounts are what it's given. An extra trial arrives with a new ID and a new email, so it goes in as a new trial start, and its cost gets averaged into everyone else's. Every dashboard is counting exactly what it was told to count, and trial farming lives in that gap.
The obvious fix is to count unique emails, and it hands back almost the number you started with. People restarting a trial use a new address each time, because a new address is what restarts it.
How to define your free trial abuse rate
An extra trial is any trial beyond what your policy allows one person. Four rules keep it honest:
- Count each trial. Someone with five trials adds four to the top line.
- Write the policy first. If you haven't decided whether a work account and a personal account are one person, you can't mark anything. Start with a one account per person policy.
- Use a closed month. Pick a month where every trial has ended, so the paid results are final.
- Keep the people who paid. The customer who cancelled today still had four extra trials.
A sample audit you can finish in two days
Marking all 3,000 October trials at two minutes each would take 100 hours. A random sample of 400 takes about 13 (400 × 2 ÷ 60).
Export the month. Pull every trial start with its account ID, date, email and result, plus any phone number or card you required. Keep earlier months beside it, because that's where a person's first trial will be. Finding repeat trial users in your database goes deeper.
Draw a random sample. Shuffle the list and take the first 400. Pick the ones that look odd and you'll measure your own suspicions.
Look each trial up against earlier ones. A tie is something plainly in your records: the same phone number or card on an earlier trial, or the same mailbox written a second way. Gmail ignores dots in its standard addresses, and mail sent to a name with a plus sign and a word lands in that name's normal inbox, so several addresses can be one mailbox.
Sort each trial into one of three piles. A disposable domain or a shared connection only gets a trial as far as unclear, since offices and households share connections every day. A trial nobody used proves nothing either, because plenty of real people sign up and never come back.
Do the arithmetic. Say your 400 come out like this.
| Result | Trials | Share of sample |
|---|---|---|
| Confirmed extra trial | 100 | 25% |
| Unclear | 44 | 11% |
| No evidence found | 256 | 64% |
| Total | 400 | 100% |
A careful investor will ask about the margin of error:
With 25% of 400, that's 1.96 × √(0.25 × 0.75 ÷ 400) = 0.042, about 4 points either way. So October's confirmed rate sits around 21% to 29%, or 630 to 870 of 3,000 trials. The 25% midpoint gives you the numbers from the top: at most 2,250 real trials and a trial-to-paid rate of at least 5.3%. Activation and retention carry the same error, so clean the duplicates out of those too.
A line you can put in the update
A random sample of 400 October trials found 25% were confirmed extra trials, give or take about 4 points. That's a minimum.
Why your free trial abuse rate is always a floor
Look at the 256 in the no-evidence pile. Nobody has shown those people are honest. Your records just couldn't tie them to anything. A person with a second real mailbox who gave you no phone number or card looks exactly like a new customer, in your data and in any audit of it. Accounts you've deleted are gone too, and so are logs that have expired.
An audit can prove a repeat happened. It can't prove one didn't. So keep two figures: 25% confirmed as your minimum, and 36% confirmed plus unclear (144 ÷ 400) as a working estimate. Nothing in your own data gives you a ceiling, which makes it hard to decide how much abuse you can live with.
How basic fixes change the number, and how they break
Each of these does catch something.
| Basic fix | What it catches | What it misses | Who it wrongly blocks | Cost to keep working |
|---|---|---|---|---|
| Email verification | Typos and dead addresses | A second real inbox | People whose code goes to spam | Deliverability work |
| Gmail dot and plus rules | One inbox written several ways | A different mailbox | People who sort their mail with plus addresses | Upkeep as mail services change |
| Disposable-domain blocks | Known throwaway services | New domains, normal mailboxes | People using privacy forwarding | A list that never ends |
| CAPTCHA | Simple scripts | People signing up by hand | People on screen readers | Fees and lost signups |
| Phone or card requirement | Casual repeat trials | Anyone with another number or card | Students, people without a reliable mobile | Text fees and fewer trial starts |
| One trial per IP address | Repeats on one connection | Any other network | Offices, campuses, households | Exceptions and tickets |
| Shorter trials | Some cost per trial | More restarts make it up | Everyone who needed the time | Fewer people reach value |
| Manual review | What a person can spot | Whatever leaves no trace | Whoever looks odd to a tired reviewer | 100 hours a month at 3,000 trials |
These fixes also damage the measurement. Say you add alias rules and a disposable-domain block in November, and December's confirmed rate falls from 25% to 12%. That's 13 points, or 390 trials on 3,000. Check whether trial starts really fell by about 390, and whether the cost line fell with them. If starts held at 3,000, the same people have probably moved into the no-evidence pile, and your chart now says the problem is shrinking. Repeating the audit monthly to catch this is another 160 hours a year (13.3 × 12). By the time you've counted a trial, you've paid for it.
Counting repeat claims when they happen
Everything above works backwards from whatever your records happened to keep. AbuseZero works at the moment someone claims the trial, and it brings far more to that moment than a domain list and an alias rule. Abuse that single checks miss is what it's built to flag: multi-layer AI risk analysis examines 140,000+ signals and detects linked claims, reading those signals together, never one at a time. You also get a proprietary algorithm, with AI that checks the claim again. New inboxes are what filled your no-evidence pile, and AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing. It checks mailbox aliases and disposable domains too. Creating a fresh account isn't enough to erase previous activity. Your server makes one API call and gets a clear allow or deny, under rules you control, and you don't have to build your own abuse prevention system.
If measuring is what brought you here, start with Observe mode. It shows you what would be denied before anyone is blocked, so you can put that number next to your audit. The quickstart sets up your first benefit.
Frequently asked questions
What is a normal free trial abuse rate?
There's no benchmark worth copying, because every company's floor depends on what its records keep. Compare your own months.
How big should my audit sample be?
A sample of 400 gives a margin of about 5 points at worst. A sample of 100 gives about 10, which is usually too wide to share.
Should I restate numbers I've already shared?
Share the corrected figure with its margin, say it's a minimum and show the old number beside it.
Put a number you can defend in your next update
Start with Builder
Pay as you go. Evaluate your first claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.