Skip to content
Inside the Boundary

All notes  /  Putting it in

How These Deployments Fail

The recurring failures, each visible in the first month, each with a specific correction.

Putting it in · Analysis

Geofenced attendance fails in recognisable ways, and the diagnosis matters more than the product choice.

Radius set from a default

The symptom: a refusal rate that varies wildly between sites.

The cause: one number applied everywhere, chosen by the vendor.

The correction: measure each site and set per-site radii.

No fallback

The symptom: people standing at the door, and a grievance about unpaid time.

The correction: accept with a flag, plus a non-phone route for anyone without a working device.

Refusals not logged

The symptom: a dispute you cannot answer, because only successful punches were stored.

The correction: log failures with their accuracy figures, which costs nothing at procurement.

Punch time treated as arrival time

The symptom: lateness conversations about minutes that were queueing and fix delays.

The correction: a punch window, and checking the site's refusal rate before any conversation.

Background location enabled

The symptom: a workforce reaction out of proportion to a clock-in rollout.

The correction: request only while-in-use, verify what arrives when the app is closed, and say so in the notice.

Coordinates kept indefinitely

The symptom: a subject access request that produces a map of someone's year.

The correction: delete coordinates on a short schedule while keeping the attendance record.

Shift change not tested

The symptom: a queue, and people punching from the car park.

The correction: a terminal or a beacon at high-volume doors, and a punch window.

Personal phones assumed

The symptom: someone without a suitable phone is marked absent.

The correction: an alternative route, stated when the system is introduced.

Corrections piling up

The symptom: pay errors at period close.

The correction: a named approver with a deadline, and fixing the refusal rate that generated the volume.

The common thread

Every one of these is the employer's system failing and the worker carrying the cost.

Reversing that — the system fails, the worker is paid, the site gets fixed — is the whole design principle, and each correction above is an application of it.

The recovery move

For a deployment already generating complaints.

Publish the refusal rate by site.

Fix the worst one and say you did.

Turn on accept-with-flag if it is not on.

Approve outstanding corrections without scrutiny and pay them.

One fortnight of that, and a resented system becomes one people use as intended.

A concrete product reference

When translating this principle into a buying test, the design-team scenario provides a concrete workflow reference. Verify current behaviour in a trial and judge it against the purpose and limits above.