INTRODUCTION
Someone is halfway through a long application. They step away to find a document, answer a call, make a coffee. When they come back and press Next, the product shows a sign-in page. They sign in. The form is empty.
The timeout was a security decision. The empty form was a design decision, usually made by nobody. This piece is about separating the two, so that a session can expire for good reasons without taking people’s work with it.
A timeout is three decisions, not one
Teams tend to talk about “the session timeout” as a single number. It is really three choices. How long can someone be inactive before the session ends? When, and how, do you warn them? And what survives when it does end?
The first is a risk decision, and it should stay one. The OWASP Session Management Cheat Sheet describes common idle timeouts of 2 to 5 minutes for high-value applications and 15 to 30 minutes for low-risk ones, plus an absolute limit based on how long people normally use the product. Design should not argue with that number. It should make sure the other two decisions are made on purpose.
Warn before it ends, and make staying easy
WCAG 2.2 treats unexpected time limits as an accessibility problem, because people who read, type or move more slowly hit them first. Success criterion 2.2.1 Timing Adjustable, at level A, asks that people can turn a limit off, adjust it, or be warned before it expires and given at least 20 seconds to extend it with a simple action, at least ten times. There are exceptions, including real-time events and limits longer than 20 hours, but an idle sign-out on a form is rarely one of them.
HMRC’s service timeout pattern shows what this looks like in practice. Its services time people out after 15 minutes of inactivity and warn them at 13, in a dialog that says “For your security, we will sign you out in 2 minutes.” It counts down the seconds in the final minute, offers a Stay signed in button and a Sign out link, and says what will happen to answers if that matters. It is not glamorous. It is a sentence, a countdown and two clear choices.
Preserve the work across the sign-in
Warnings fail. People are away from the screen, which is the whole reason the session went idle. So the third decision does the heavy lifting: what happens to their work after they sign back in.
WCAG’s stricter level AAA criteria describe the target. 2.2.5 Re-authenticating asks that when an authenticated session expires, the person can continue without losing data after signing in again. 2.2.6 Timeouts asks that people are told how long they can be inactive before losing data, unless that data is kept for more than 20 hours. Even if you do not aim for AAA, those two sentences make a useful product requirement: sign-in returns people to the same step, with what they entered.
Mechanically, that means saving drafts against the account as people go, or holding the unsent state and restoring it after re-authentication. We wrote about the wider version of this problem in Design the return to unfinished work.
The security counterargument
A security lead might reasonably object. A warning dialog tells an attacker at an unlocked desk how to keep the session alive. Saved drafts are more personal data stored for longer. Shared computers make both worse.
Those are real concerns, and they shape the design rather than cancel it. Extending a session needs someone present to press a button; it does not remove the absolute limit. Drafts saved on the server sit behind the same authentication as everything else, and the WCAG note on 2.2.6 itself points out that privacy rules may require consent before keeping data for people who are not signed in. For high-value products with very short timeouts, the case for preservation gets stronger, not weaker, because more people will hit the limit.
A worked example
Imagine a grant application with a 20-minute idle timeout. Analytics show that many applicants spend longer than that on the budget step, reading attachments in another tab. A reasonable redesign: save each step when people leave a field, warn at 18 minutes with a one-click extension, and after re-authentication return them to the budget step with their figures intact. The security setting has not changed. The cost of hitting it has. This example is hypothetical.
What to do next
Write down three things: your idle timeout and who chose it, the longest task people actually perform in one sitting, and exactly what survives a re-login today. If the task is longer than the timeout and the answer to the third question is “nothing”, you have found a design problem with a security label on it.
Sources checked on 16 November 2026: WCAG 2.2 Understanding documents for 2.2.1, 2.2.5 and 2.2.6, HMRC’s service timeout pattern and the OWASP Session Management Cheat Sheet.