The tasks you should not automate

Most automation projects fail before they have begun, and always in the same way. Someone sees a demo, the company buys a licence, nobody knows what to plug into it. Six months later the subscription is still being charged and the spreadsheet is still filled in by hand.

The mistake is not having chosen the wrong tool. It is having started with the tool. The useful question comes earlier: which tasks are worth attending to, and which should be left alone.

The frequency calculation, which rules out the most candidates

A ten-minute task repeated forty times a month consumes eighty hours of work a year, which is ten working days. It deserves a look. A one-hour task done three times a year consumes three, and no automation pays for itself over three hours.

The calculation looks obvious written down like that. It is far less obvious in a meeting, because the tasks people mention spontaneously are the painful ones, and painfulness has nothing to do with frequency. The annual accounts close is hateful and does not automate. Copying orders from an email into a spreadsheet annoys nobody enough to be mentioned, and it is the one costing ten days.

Rules that move

The second criterion is harder to see: the stability of the rule.

An automation encodes a decision. As long as the decision does not change, it holds. When the rule changes every month, someone has to go back into the flow every month, and that someone costs more than the person who was doing the task by hand. The work has not disappeared, it has moved to a rarer profile.

This comes up often on commercial discounts, on price lists negotiated customer by customer, and on anything depending on a recent human judgement. The right answer is then to automate the part that does not move, the gathering and the formatting, and to leave the decision with the person.

The cost of an error

The third criterion is the one people forget, and it is the most expensive. On some tasks, one error costs more than all the time saved in a year.

A bank transfer, a regulatory filing, a send to an entire contact list. An automation that gets it wrong once in a thousand on operations like those wipes out months of gains in an afternoon. Those tasks can be tooled, checked, prepared automatically, but the execution stays triggered by someone who has looked.

Why we charge for the audit

These three criteria serve as much to rule out as to keep, and that is what makes the position tenable. We look at what the teams do during an ordinary week, how often, at what cost, and we deliver a document listing the candidate tasks, their current cost, the price of automating them and how long before that pays for itself. Whatever is not worth doing is written down as such, with the reason.

The audit is paid for and it ends. You are committed to nothing afterwards, and the document is yours: you can hand it to your own team or put it out to tender. That is the condition for it to tell the truth. A free audit always concludes that you should buy something, because that is the only way it pays for itself.