The work between systems
Most business automation is not clever. It is reading a file from one place, changing its shape, and putting it somewhere else — repeatedly, reliably, without anyone doing it by hand.
Python is unusually well suited to that: good libraries for every file format, every API and every database, and code that is readable by whoever inherits it.
What to automate first
- Whatever someone does every morning — high frequency, well understood
- The report assembled from three systems
- The file that arrives and gets reformatted
- The data re-keyed between two systems
- The check somebody performs manually every week
Ask who in the business would be most relieved by a specific task disappearing. They will tell you immediately, and they are usually right about which one.
Do the arithmetic first
Frequency multiplied by duration multiplied by hourly cost gives the annual value. Compare it against build plus three years of maintenance and hosting.
- Under about two hours a week is hard to justify unless errors are expensive
- Count only time that would actually be redeployed
- Include the cost of errors where they are real and countable
- Be conservative on the saving and generous on the cost
Automate the routine, not the exception
The exceptions are where the pain is loudest and where automation is least reliable and most expensive. Automate the routine eighty per cent and route the rest to a person.
That division is where the return is, and it is where most automation projects go wrong by aiming at the hard cases first.
Build it to be trusted
| Requirement | Why |
|---|---|
| Logs what it did | Somebody will ask |
| Reports on completion | Silent failure is the risk |
| Handles bad input | There will be bad input |
| Can be run manually | For catch-up and testing |
| Has an owner | Unowned automation drifts |