What a Fake-Door Test Can Tell You Before You Build
Eighty clicks aren't eighty buyers. Follow a proposed bookkeeping feature from exposure through informed opt-in to the evidence needed for a limited pilot.
A fake-door test puts an entry point for an unavailable feature or offer in front of potential users and records whether they explore it. Also called a painted-door test, it can help decide whether a problem deserves more investigation before the team builds the solution.
The difficult part begins after the click. A person might expect something different from what you plan to build, discover the price is unsuitable or decline to hear any more. The useful result is the path from a realistic encounter to an informed next step, with those differences intact.
The feature behind the door
Consider a bookkeeping app for independent freelancers. Its team is deciding whether to build a monthly handoff pack for a bookkeeper: a consolidated export of categorized invoices and receipts. The pack doesn't exist yet. This is an illustrative product and test, with worked figures rather than observed results.
The team could add an item called “Monthly handoff pack” in the relevant export menu. After a click, the page explains that the pack isn't available and that the team is exploring it. It describes the proposed contents and offers an optional invitation to discuss a future pilot.
That encounter tests interest in the entry point and the described possibility. It doesn't test whether the export works, whether a bookkeeper can use it or whether anyone will pay once it exists.
Amplitude's fake-door guide describes the entry-point, interaction and transparent follow-up sequence. For your own decision, keep the immediate click separate from what happens after the person understands the feature's availability and scope.
A truthful follow-up changes what the signal means
The follow-up should answer the user's immediate question: why didn't the expected feature open? For the bookkeeping example, a suitable message might be:
This handoff pack isn't available yet. We're exploring an export that brings categorized invoices and receipts into one monthly file. You can return to your current export, or ask us to contact you about a possible pilot. Signing up doesn't place an order or commit you to a purchase.
Only promise a delivery date if the business has actually committed to it. “Coming soon” and “in development” also describe states the team should be able to support.
The original task still matters. A person closing their monthly books shouldn't lose access to an existing export while the team tests a new one. Keep the test out of a critical path if failing to deliver the implied function would leave work unfinished.
An entry point labeled “Explore a proposed handoff pack” is more explicit before the click, but it measures a different encounter from an apparently available export. Record the exact wording. Don't compare the resulting click rates as if the only difference were feature demand.
The denominator belongs to the opportunity to respond
Suppose 800 eligible freelancer accounts encounter the item during a two-week test. Each account belongs to one freelancer, and the tally counts each account once at each stage. Exposure means the item was actually shown in the intended menu, not merely that an unrelated page loaded.
Stage | Example count and rate |
|---|---|
Eligible accounts exposed | 800; the base for the full path |
Accounts clicking the item | 80/800 = 10% |
Accounts opting in after the explanation | 24/800 = 3%; also 24/80 = 30% of clickers |
Accounts attending a follow-up conversation | 8/800 = 1%; also 8/24 = 33.3% of opt-ins |
Product use or payment | Unavailable: no working pack was offered |
Reporting “30% conversion” without its base would obscure that only 3% of exposed accounts opted in. Reporting 80 clicks as 80 committed buyers would skip the informed choice entirely.
Repeated clicks may tell you about frustration or navigation, but they aren't additional interested accounts. If several people share an account, decide whether the business question concerns people or buying accounts and retain both identifiers when appropriate. Don't silently move between them.
The test period and eligibility also shape the result. A monthly handoff feature shown mainly in the middle of the month may receive a different response from one encountered while closing the books. That is a reason to inspect the context, not to extrapolate the 10% click rate to every account all year.
What the conversations need to resolve
The eight follow-ups are a chance to find out what the proposed pack is expected to do. Ask about the person's last real handoff: the files involved, the recipient's requirements, what had to be corrected and what the current workaround costs in time or effort.
Someone who wants a tax-ready statement and someone who wants a folder of receipts may have clicked the same label while imagining different products. The next investment should address that difference before engineering treats the click total as agreement on a specification.
An informed opt-in is still a selection from the exposed group. It doesn't tell you how common a problem is among people who ignored the item. A small comparison with nonclickers can help investigate whether they didn't notice the item, didn't understand it or didn't need it. The appropriate evidence depends on that question.
Sapien can compare descriptions and investigate modeled audience interpretations before a live test. Actual clicks, use and payment need their corresponding observed records.
Our concept-testing guide can help sharpen the offer being described. The live door then investigates behavior in a specific encounter, rather than substituting for that earlier clarity work.
The next investment should match the evidence
There is no universal good fake-door conversion rate. A team considering a small export may need a different signal from one planning a costly platform integration. Agree beforehand what the next decision is and what evidence could change it.
For the bookkeeping team, 24 informed opt-ins and eight conversations might justify a limited prototype or a small manual pilot. They don't, on their own, justify a full automated build. The team still needs to see whether the pack can be delivered, whether it fits the recipient's workflow and whether its value survives an actual price.
If the follow-ups reveal that people expected a different output, revise the concept rather than repeatedly changing the button to collect more clicks. If the offer is understood but the people with the relevant task show little interest, the team may have learned enough to stop.
Our preorder research guide distinguishes later forms of commitment, while our AI-research-versus-A/B guide explains when an observed experiment is needed. Preserve those stages as the project progresses.
A record the build decision can survive
Keep the tested label, placement, proposed scope, availability explanation, eligibility, period and account-level event counts with the recommendation. Include the reasons from follow-ups and the outcomes the test couldn't observe.
The door has done useful work when it reduces uncertainty about the next investment. Sometimes that means building a small version. Sometimes it means a clearer promise or a different problem. A click opens that investigation; the rest of the evidence decides where it goes.
