Beermat Software All articles
Startup Strategy

The Lock-In That Never Ended: Retiring Features Without a Riot

Beermat Software
The Lock-In That Never Ended: Retiring Features Without a Riot

Photo: David Howard , CC BY-SA 2.0, via Wikimedia Commons

Every software product has a regular who will not leave. Not a customer — a feature. It sits in the corner of the interface, nursing the same pint since some previous release, occasionally catching the eye of a new starter who asks, 'What does that do?' Nobody knows. Nobody has used it since 2019. And yet every time someone suggests removing it, a product manager says, 'We've got a few accounts who might rely on it.' That sentence is the software equivalent of a lock-in: technically the doors are shut, but nobody has the heart to turn the lights on and ask everyone to go home.

The cost of keeping the lights on

A dead feature is not free. It needs tests, documentation, support scripts, security reviews and design attention. It makes onboarding slower because new users wonder whether they have missed a module. It adds another branch to every release. It is the dusty fruit machine in the corner of the pub: nobody plays it, but it still takes up space, still needs power, and someone still has to wipe it down.

The real cost is not the code. It is the attention. Every dead feature is a small tax on every person who has to remember it exists. Multiply that by a decade and you have a product that feels tired before it has even opened.

Why teams keep the lock-in going

Most teams do not keep dead features because they have carefully decided to. They keep them because nobody has decided not to. There is no owner. There is no date. There is a vague fear of upsetting a hypothetical user, a sales deck that still mentions it, and a general British instinct to avoid a difficult conversation. So the feature stays. The lock-in continues. The lights stay on and everyone pretends they are having a nice time.

Sunk cost does the rest. Someone built it in 2016, so removing it feels like admitting the build was a mistake. It might have been. That is allowed. Software is not a cathedral; it is a pub. You can change the menu.

Start with a last-orders warning

Deprecation should begin with a clear, boring announcement. Not an apology. A date. 'This feature will be retired on 31 March. Here is what replaces it. Here is how to export your data. Here is who to contact.' Say it in the product, in the docs, in the changelog and in an email that does not look like a phishing attempt.

The pub landlord does not creep round at 11pm and turn the lights off mid-sentence. They call last orders. Everyone gets a warning, a final drink and a chance to settle up. Your users deserve the same courtesy. The warning is the service.

The slow clear-out

After the warning, move in stages. Make the feature read-only. Hide it for new accounts. Disable it for everyone else. Keep the export route alive for longer than feels necessary. Monitor support tickets. Give the support team a script that is one paragraph long and does not contain the phrase 'as per my last email'.

Treat deprecation as a project. Give it an owner, a deadline and a rollback plan. If you cannot draw the plan on the back of a beermat — feature, owner, date, replacement, support route — it is not ready. This is not bureaucracy. It is the difference between a graceful exit and a lock-in that ends with someone crying by the cigarette machine.

The lock-in that never should have happened

Some features get reprieved forever because nobody trusts the data. So make the threshold explicit before the argument starts. If a feature is used by less than half a percent of active accounts, generates more support tickets than sign-ups and has no strategic owner, it goes. That is not cruelty. That is queue management.

The loudest account in the room should not get a permanent reprieve just because they shout. A good landlord does not keep a flat pint on tap because one regular claims to like it. They thank them, offer them something better and clean the line.

A practical script for calling time

  1. Name the feature and give it an owner.
  2. Pull the usage data and the support tickets.
  3. Decide: keep, fix or retire. Write the decision down.
  4. Publish the notice with a date, a replacement and an export route.
  5. Migrate the people who need migrating. Help them.
  6. Turn it off in stages. Watch the tickets.
  7. Hold a short feature funeral. Say thank you. Delete the code.
  8. Add it to a deprecation log so nobody rebuilds it by accident in eighteen months.

That last step matters. British software teams have a habit of deleting things and then reinventing them badly because the original decision lived only in a Slack thread that has since been archived.

Sunsetting is a service, not an apology

We treat removal as a failure. It is not. A product that never removes anything is not stable; it is cluttered. Users do not thank you for keeping a feature they never use. They thank you for making the thing they do use faster, clearer and less likely to break.

So call time. Wipe the bar. Put something better on. The regulars will grumble for ten minutes and then order the new thing. That is how pubs work. It is how software should work too.

All Articles

Related Articles

The Buffering Village: Designing for the Places the Fibre Forgot

The Buffering Village: Designing for the Places the Fibre Forgot

The Council Contract Gauntlet: A Survivor's Field Guide to Selling Software to Local Government

The Council Contract Gauntlet: A Survivor's Field Guide to Selling Software to Local Government

Scribble It Down or Throw It Out: The One-Coaster Rule for British Founders

Scribble It Down or Throw It Out: The One-Coaster Rule for British Founders