Free trials are how most small SaaS products sell. They are also the easiest thing in your product to take without paying. One person with one laptop can start a new trial every 14 days with a new email, and on paper each of those trials looks like a new lead.
This guide is for small and mid-size SaaS, especially teams in Latin America that run trials with no card because their buyers pay with Pix, OXXO, PSE, Mercado Pago or a bank transfer. Without a card, the usual advice (“require a card”) doesn’t fit. Here is what to do instead, in order.
How trial abuse shows up in your data
Before you change anything, learn to see it. Trial abuse leaves a few patterns:
- Same device, new emails. The clearest one. A new account arrives from a browser or phone that already holds one or more of your accounts. The email is new; the device isn’t.
- Groups of signups. Several accounts from the same device or network within minutes. A person reselling access, or a script collecting free credits, rarely makes one account at a time.
- Disposable and throwaway emails. Addresses from temporary-inbox domains, or a single Gmail inbox used many times with dots and plus signs.
- Datacenter and VPN traffic, plus automation. Signups from cloud servers or VPNs, especially together with headless-browser signals. A VPN alone is not proof; many people use one for privacy. Treat it as context.
- Usage shaped like a loop. The trial is used right up to the last day, then a new trial starts from the same device. Or the free credits are spent in the first hour and the account is never opened again.
Device and network signals have error rates, and we publish ours on the accuracy page. In our lab, Prynt detected datacenter and VPN traffic 99.8% of the time (89.5% on a blind holdout), wrongly flagged about 1 in 1,200 IPs, and caught 89% of stealth automation tools (238 of 268) while flagging 0 of 240 real humans as bots. Those are lab numbers, not a promise about your traffic.
The playbook
Step 1: Watch for 14 days before you act
Let mirror mode run: Control starts in it on its own. For the first 14 days Prynt only observes: every signup goes through as usual, and nothing changes for your users. On day 14 you get “what we would have stopped”, case by case, so you can check each one before trusting it.
This matters more than any rule. Two weeks of watching tells you whether you have a problem worth friction at all. Some products find that repeat trials are a handful of curious people; others find that a third of their “leads” come from a few devices.
If you already have rules of your own, run them in evaluate mode during the same period, so you see what they would do before they do it.
Step 2: Use the “free trial” decision
Rather than writing rules from scratch, start from the ready-made free trial decision: send useCase: 'free_trial' with the identification. It looks at trials already started on the device, disposable emails, VPN, proxy and datacenter networks, and automation, and returns one of three outcomes:
- Allow: start the trial.
- Step up: ask for one more step before the trial starts.
- Block: don’t start a trial from this device.
The decision is a recommendation. Your signup code reads it and decides what to do, so you stay in control of every user-facing message.
By default, a trial counts as a repeat when the device started another one in the last 90 days. If your trial cycle is shorter or longer, change that window in the console.
Step 3: Climb the friction ladder one step at a time
Blocking is the last resort. The friction ladder says which step to ask for, from lightest to heaviest:
| Step | When it fits | What it costs a real user |
|---|---|---|
| None | First trial on this device, no risk signals | Nothing |
| Captcha (for example, Cloudflare Turnstile) | Automation hints | A few seconds, often none |
| Verify email | The device started one trial before, a disposable address, a VPN, proxy or datacenter network, a browser altering its fingerprint | Opening their inbox |
| Card or passkey | Two or more earlier trials on the device | A card, or one tap on their own device |
| Block | A device or account you labeled fraud, or one on your deny list | No trial from this device |
Match the step to the attacker. A captcha slows scripts but does nothing against a person who simply makes a new email every two weeks; for that person, a card is the step that costs something. Research on login challenges at Google points the same way. In a study of over 350,000 real account-hijacking attempts, Doerfler et al. (WWW 2019) found that knowledge-based challenges stopped as few as 10% of phishing-rooted attempts, while device-based challenges blocked over 94% of them and 100% of automated ones. The same study found that friction led 52% of legitimate users to fail to sign in at first, though 97% eventually got into their account in a short time. That was login, not signup, but the two lessons carry over: the type of step matters, and every step has a cost for real people, so use it only where the risk is.
Step 4: Review groups of accounts
Single accounts are hard to judge. Groups are easier. Account clusters, still rolling out in Prynt Control, will group new accounts that share a device, a network and a time window, for example “9 accounts from the same place in 20 minutes”. Until they reach your account, start from the device: when you read an event with your secret key, accountsOnDevice lists the other accounts seen on that device.
When you open a group, ask simple questions: Are the emails related? Did any of them pay? Did they use the product, or only the free credits? Coworkers signing up together look different from nine accounts that each used exactly the free quota.
Step 5: Label what you review
Mark each case you look at fraud or legit, in the console or with POST /v1/outcomes. A device or account labeled fraud gets block on its next decision. The case queue, still rolling out, will sort suspicious trials with the most suspicious at the top and learn from those labels; fifteen minutes a day at the top of the list will usually be enough for a small product.
Labeling “legit” is as useful as labeling “fraud”. It tells the system which patterns are normal for your customers, such as a family sharing a laptop or an agency testing for a client.
Step 6: Put a number on it
The monthly money statement, still rolling out in Prynt Control, will estimate what abused trials cost you, with the method shown. You can do the same by hand today, and the method matters more than the number. A conservative estimate counts only trials you labeled fraud or that came from a device that had already used a trial, and multiplies them by your own cost per trial (servers, API credits, onboarding time). It should never assume that every stopped trial would have become a paying customer.
What to measure
Watch these numbers before and after you turn on any step:
- Repeat-trial rate. The share of new trials from a device that already used a trial. This is the core number. It should drop once steps are active.
- Trial-to-paid by risk band. Split trials into low, medium and high risk, and compare conversion in each. If high-risk trials almost never pay, friction there is cheap. If they convert well, your steps are catching real buyers; loosen them.
- Step completion rate. For each step on the ladder, the share of people who finish it. A captcha that half your users abandon is a problem, whatever it stops.
- Prevalence. The share of new accounts that look abusive. Count it from your labels; the money statement will report it when it ships.
- Proactive rate. The share of abuse caught before you or a customer noticed it. The statement will report it too.
For more detail on reading these, see measuring trial abuse.
A short checklist
- Let the 14-day mirror mode run (Control starts in it on its own).
- Review “what we would have stopped” and label what you see.
- Send
useCase: 'free_trial'and, for the first weeks, cap the free trial decision at “verify email” in the console. - Raise the cap to “card or passkey”, so devices with two or more earlier trials are asked for a card.
- Leave “block” to direct evidence: devices and accounts you labeled fraud, and your deny list.
- Check the five numbers every month and adjust.
The goal is not zero abuse; it is to stop paying for the same person’s tenth trial without making the first trial worse for anyone. Prynt Control is built for that, and the pricing page shows what each plan includes.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.