ASO Ltd.

Blog

A delete-account page has three outcomes, not two

Google Play requires a way to delete an account without the app. We built that page for our delivery apps. The hard part was not the deleting. It was never telling someone their data is gone when we cannot prove it.

· 6 min read

You are putting an app on Google Play. The console asks for a URL where someone can delete their account without installing the app. It is one field in a form, so you treat it like one. A page, an email box, a thank you message. Done.

Then you write the code and the shape of the problem shows up. Deleting an account is a request your page sends to your own backend over somebody's mobile connection in Douala. That request has three possible endings, not two. It worked. It provably did not work. Or you have no idea.

Most pages have a screen for the first two. The third one gets rounded to whichever neighbour is closer in the code, and both roundings are a lie that costs you something.

Round it up to success and you have told a person their data is gone when it may still be there. That is the one promise in your privacy policy a regulator can check, and you broke it on the screen where you made it.

Round it down to failure and you have told a person nothing changed when their account may be in a 30 day deletion window. They try again. They email support. They assume the first attempt did nothing, so they stop watching for any confirmation that does arrive.

What we built

Earlier this month we built /delete-account/ on asoltd.info. It covers ASO Delivery, ASO Rider and ASO Restaurant. You sign in the same way you sign in to the app, with a phone number and a code or with an email and password, then you confirm, and the page calls the same deletion route the apps call, with the same 30 day window.

There is no new route and no new way to send a text message. The page is a second door onto the machinery that already exists. That decision removed most of the ways this kind of page goes wrong, because a separate deletion endpoint built for the website is a second thing to secure, rate limit and keep correct.

The first version came back from review with four findings. Not one of them was about the deleting. Here they are as rules.

1. Prove whose account it is before you show a Delete button

Signing in is not enough on its own. After the sign in succeeds, read the account back from the API and check two things: that the id you get belongs to this sign in, and that the role is one you recognise. Only then show the confirmation step.

If that read fails or comes back in a shape you did not expect, show nothing as verified. Offer a retry or a sign out. A confirmation screen that says "Signed in as" above a name you did not confirm is a page that will eventually delete the wrong account and have a reason why it was not its fault.

2. Keep "I do not know" as its own outcome

A lost answer proves nothing. So does a timeout, a 503, and a success with a body you cannot read. None of those mean the account survived.

Our page treats all of them as one state: we could not confirm what happened. The words on screen are "your account may or may not have been deleted", followed by an address to write to and a request not to try again here. When that state is reached, Delete is locked for the rest of the page, the "Start again" button stops working, and no screen after it is allowed to tell the person nothing changed. The route out is a human, not a second button.

This is the rule with the widest reach. Anyone who has reconciled Mobile Money knows the same shape: a payment whose result never came back is not a failed payment, and treating it as one is how you charge somebody twice. A deletion request is the same problem pointed at a privacy promise instead of a balance.

3. Report the sign out as it happened

Our sign out route reserves one status, a 204, for a session it actually ended. If the page gets that, the session is over and the tokens are forgotten. Anything else, including a 200, and the page says it signed you out here but could not confirm it, and offers a retry.

That wording feels clumsy and the first version did not have it. The alternative is worse. On a shared or borrowed phone, "signed out" is a safety claim. Make it only when it is true.

4. Inherit the app's limits, including the awkward ones

The page sends codes through the same route the apps use, so the server applies the same send limits it applies to the apps. On top of that, the page keeps its own 60 second hold per number, with a visible countdown that survives a second tab or an edited number.

It also inherits the wording. The app never tells you whether a number has an account, because that turns a login form into a tool for checking who your customers are. The website version says the same thing: if an account exists for this number, we have sent it a code. Slightly unhelpful, deliberately.

Say what deletion actually does

One more thing, and it is the easiest to get right if you have already written an honest privacy policy. Put the real sequence on the confirmation screen, with the real timings.

Ours says you cannot sign in from that moment, a restaurant stops receiving orders straight away, your details are held for 30 days so a mistake can be undone, and then at the next daily run, which can be up to 24 hours later, the personal fields are erased and the stored files start being deleted. Past orders stay in the accounts as financial records without your personal details.

Anyone who reads that can decide. A page that only says "this cannot be undone" is both vaguer and less true than what actually happens.

What this does not solve

The page only works for someone who can still sign in. Lose the phone number and you are outside it, writing to a person, proving the account is yours, and waiting. That path is slower and a human is in it, deliberately. An automated way to delete an account you cannot sign into is a feature for whoever is angry with you.

The page is also only honest about the moment it ran. It reports what the API told it. It cannot promise that the scheduled job which erases the fields on day 31 ran on day 31. That is a separate thing to check, in a separate place, on a schedule of its own.

That is the uncomfortable part of this work. The page is the visible half of a deletion promise and the half you can finish in a week. The rest is machinery you have to keep watching.