The one-button billing robot

Everyone wants the robot that does everything. Ours starts as a scheduled job pressing one button.

Not because we cannot build the ambitious version. Because the ambitious version fails in a specific and expensive way: it works for 95% of cases and quietly mishandles the rest, and nobody notices until the exceptions have compounded for a quarter.

Phase 0

The first version of a billing automation we ship does exactly one thing: it presses a button the billing software already trusts, on a batch small enough that a human can check every line of the output.

That is it. No branching. No cleverness. A person reviews every run.

It looks like a waste of an engineer. What it actually is, is a month of free discovery, because the exceptions surface in a context where they are cheap. Every unrecognised code, every payer quirk, every “oh, we always handle that one manually” gets found by a human who is already looking.

Earning permissions

After a month of being right, it earns the next permission. Bigger batch. Then a second button. Then a category of exception it is allowed to resolve on its own.

Each step is written down, and each step is reversible. Nothing widens because someone felt confident; it widens because the previous scope ran clean for a defined period.

The part that actually matters

Anything the process does not recognise goes to a person. Automatically. Immediately.

Not to a log file. Not to a retry queue. Not to a dashboard somebody checks on Fridays. The design goal is not “handle everything”, it is make the exceptions loud.

It is easy to build automation that appears to work. The unglamorous engineering is in the routing of what it cannot do.

Slower on paper

A Phase 0 robot delivers maybe half the value of the ambitious version, at something like a hundredth of the risk. On a Gantt chart that looks like underachievement.

In practice it ships sooner, breaks less, and, the part that compounds, it teaches you what the ambitious version would have got wrong. We have never regretted starting boring. We have regretted the other thing.