Starting Again After an Automation Project Went Wrong

Second attempts are common and not embarrassing

A substantial share of the work we are asked to quote for is a second attempt. The first one usually stalled rather than exploded: go-live slipped twice, the pilot ran alongside the manual process indefinitely, the champion moved on.

Nobody sends an email announcing that an automation project has failed, which is part of why it takes so long to act.

What usually went wrong

1. The scope described a wish, not a process. “Automate our order handling” is a hope.
2. Exceptions discovered in production because the specification had one sentence about them.
3. Nobody owned it internally, so questions took days and assumptions became rework.
4. The process changed during the build because it was never stable.
5. Built for the demo — no monitoring, no retries, no manual override.

It is almost never a technical failure. The code we inherit is usually competent; what is missing is the answer to what happens when the order arrives as a photograph.

What we do first

Read what exists before proposing anything. Sometimes the previous team produced a decent specification and poor delivery, in which case a great deal is salvageable.

Sometimes the specification was the problem, and rebuilding on it would reproduce the failure. We will tell you which, with reasons, rather than reflexively recommending a rewrite.

What is usually salvageable

• Integration work already done against your systems — often the expensive part
• The data model, if it reflects the business rather than the previous supplier's assumptions
• Anything already in production and being used, however partially
• The knowledge of what did not work, which is worth more than people think

How the second attempt differs

Smaller first phase. Exception list written before anything is built. A named owner confirmed before we start. Working software every fortnight. Parallel running before cutover.

And measurement, so that this time the outcome is a number rather than an impression.

Frequently asked questions