Every channel a customer uses
A real browser, phone number, inbox and card. They text you, call your AI agent and click the link in your email.
Invite-only beta · Limited spots
AI customers that find your bugs before real ones do.
Get confirmed bugs: the double charge, the booking that never moved, the sign-up code that never came. All you send is a URL. No scripts, no SDK.
Made by engineers who ship AI agents to real customers, and got burned by these bugs.
When customers find the bugs first
-
Air Canada had to pay a refund its website chatbot promised.
Moffatt v. Air Canada, 2024 (opens in a new tab) -
Cursor's support bot made up a one-device policy. Complaints and cancellation threats followed.
Ars Technica, 2025 (opens in a new tab) -
17% of US shoppers who abandoned a checkout cited site errors or crashes.
Baymard Institute, 2025 (opens in a new tab)
Bugs your tests can't see. They hide in texts, inboxes, phone calls and calendars, and show up days later.
Days, not minutes
They come back for the reminder, the renewal and the follow-up, like real customers.
Checks what really happened
Not just what the screen says: the booking in the calendar, one charge, the order on the seller's side.
Several customers at once
Two people on the last spot, a buyer and a seller, a team that invites each other.
From a link to confirmed bugs. Nothing to install, nothing to maintain, no access to your code.
-
01
Send a URL
And a few lines on who it's for. Docs help, but they're optional.
-
02
We map every journey
The happy paths and the dumb stuff people do: double-clicks, the Back button, a link opened a day late.
-
03
Fake customers run it
Again on every release. We keep the map current, so you never write or fix a test.
Only confirmed bugs. No flaky noise to sort through.
-
Reproduced twice
If a second fake customer can't make it happen again, you never see it.
-
A video of every bug
Watch it happen, then hand it to an engineer or your coding agent with the repro steps.
-
In plain words
What broke, for which customers, and what it costs you.
Not another AI testing tool. Most stop at the screen.
| Most AI testing tools | FakeCustomers | |
|---|---|---|
| Setup | You describe or record each test | Send a URL, we map every journey |
| Where they look | The screen of your web or mobile app | Your app, texts, inbox, phone line and card |
| How long | One session, a few minutes | Days, like a real customer |
| Passes when | The screen looks right | It really happened: the booking, the charge, the order |
| You get | Failed tests to sort through | Only bugs a second run reproduced |
Pricing after the beta. The closed beta is free. Nothing is charged until your team agrees to a plan.
-
Minimum
$299 a month
Join the beta with Minimum- 1 product
- Up to 10 journeys, 150 runs a month
- Web and email, with real inboxes and test cards
- Your customer's side
- A full run every week, re-tests on request
- Bugs confirmed by a second run, with a video
-
Optimum
Recommended$599 a month
Join the beta with OptimumEverything in Minimum, plus:
- Up to 30 journeys, 500 runs a month
- SMS and calls to your agent, 100 call minutes
- Your customer's side and your business's
- A run after every deploy
- Journeys that span up to 7 days
- Bugs sent to Slack, Linear, Jira or GitHub
-
Maximum
$999 a month
Join the beta with MaximumEverything in Optimum, plus:
- Up to 2 products or environments
- Up to 75 journeys, 1,000 runs a month
- 500 call minutes
- Every role, many customers at once
- Daily runs and production checks
- Journeys that span up to 30 days
- A person reviews every bug
Planned prices in US dollars, billed monthly after the beta, cancel anytime. Every plan: unlimited seats, no code access. A run is one fake customer taking one journey.
Questions
What's in the closed beta?
Fake customers go through the journeys that matter most in your product, on every channel you use, and you get every bug a second run confirmed, with the steps and the evidence. It's free and invite-only, with no commitment, and beta teams help shape what we build.
When can we start?
Spots open a few teams at a time, and we email you when yours does. Then you send a URL and a few lines about the product, and fake customers start on the journeys that matter most. Journeys that take days, like a reminder or a renewal, report as they finish.
Do you need our code, an SDK or test scripts?
No. Fake customers use your product from the outside, the way your customers do. A URL and a few lines about the product are enough; if part of it sits behind an invite or a firewall, we ask for just that.
Is our data safe?
Fake customers see only what a customer would, plus any test logins you choose to give us. We never ask for your code or your database, results go only to your team, and we delete everything from your runs when you ask.
Staging or production?
Either. Staging is the safest start. On production we agree up front which journeys fake customers may take, and they keep away from your real customers: on a marketplace or booking platform they set up their own test business and book with that.
Will fake customers mess up our data?
They're easy to tell apart: each signs up with an address on our own domain and a phone number we own, so you can filter them out of your analytics and delete them after a run.
How do payments work?
With test cards in your payment provider's test mode, declined cards and 3-D Secure included. On production fake customers pay only if you allow it, for example with a 100% coupon or small charges you refund.
Can they really text and call?
Yes. Each fake customer has a real phone number: it gets your texts and codes, and it calls your phone line or AI voice agent and talks like a person. Every call comes with its recording and transcript.
What counts as a confirmed bug?
Something a real customer would hit that a second fake customer reproduced on its own. Each comes with the steps, the evidence (screens, messages, the call) and who it affects. What we can't reproduce, we hold back.
What if you report something that isn't a bug?
A second run has to reproduce every bug before you see it, so flaky failures stay out of your list. If something we report is how you meant it to work, tell us once and we won't flag it again.
What can you test?
Anything customers use through a browser, a phone number, email or texts: web apps, checkouts, booking flows, marketplaces, SMS and voice agents, on any stack. Mobile web, yes; native iOS and Android apps, not yet.
How is this different from other AI testing tools?
Most check the screen of your web or mobile app in a session of a few minutes, from tests you describe. Fake customers also use your texts, inbox, phone line and card, come back days later, check what really happened and send only bugs a second run reproduced.
What happens after the beta?
You keep every report and bug, no strings attached. If you want fake customers on every release, your team picks one of the monthly plans. Nothing is charged until you agree to one.
Who's behind FakeCustomers?
A small team of engineers who build and run AI voice, SMS and booking agents in production, and know these bugs first-hand.
Real customers shouldn't be your QA team.
Free during the invite-only beta, with limited spots. We'll email you when yours opens. Beta teams help shape what we build.