Half of what looks like an AI job is a software problem from 2019

There is a construction management platform, used by plenty of firms, where updating your cost catalogue works like this.

You cannot overwrite the existing prices. The import creates duplicates instead. So before you can bring the new list in, you have to delete the old one.

There is no bulk delete.

One line at a time.

In practice people export the lot, change it in a spreadsheet, delete everything by hand, and import it fresh. That is the workaround, and it is what the people using it do every single time.

Why this matters before you buy anything

Nobody in that business would describe this as an AI opportunity. They would describe it as a Tuesday.

But when somebody eventually goes looking for workflow automation services, that is the sort of job that comes up. And it is worth being honest about what the answer is, because it is often not AI.

Sometimes the answer is AI. Sometimes it is a better platform. Sometimes it is a twenty line script that a competent person writes in an afternoon and nobody ever thinks about again.

An honest consultant tells you which. One who only sells AI will find an AI shaped answer, because that is what is in the bag.

Where AI genuinely is the right tool

The Federal Reserve Bank of St Louis surveyed nearly 14,000 workers about what they actually use these tools for. The three most common tasks were reading documents to gather technical information, at 61.3 per cent, preparing research reports at 60.7 per cent, and analysing data for trends at 57.5 per cent.

Notice what those have in common. They are all jobs where information exists somewhere, in a form that is awkward to get at, and a person has to go and extract it.

That is the shape. A PDF that needs to become rows. Twenty invoices that need checking against a list. Six months of enquiry emails that nobody has ever summarised.

If your job has that shape, AI is genuinely good at it now, and it is worth a fortnight of somebody's attention.

If your job is "the software makes me delete a hundred rows one at a time", that is a different conversation and a cheaper one.

The order to do this in

Look at your existing tools first. Not because it is virtuous, but because it is free.

Half the workflows I get shown were designed around a limitation in a piece of software that has since been fixed, or that has an export button nobody knew about. People build habits around awkwardness and then stop noticing the awkwardness.

Then look at the jobs with the shape described above.

Then, and only then, look at building something.

What you could do on Monday

Pick the task in your week that annoys you most. Not the biggest one. The most annoying one, because annoyance is a decent proxy for how often it happens.

Write down what you actually do, step by step, like you are telling a new starter.

Two things usually fall out of that. One or two steps exist purely because a piece of software cannot do something obvious. And one step is you moving information from one place to another by hand, which is the bit worth automating.

You will have done this in fifteen minutes and it costs nothing.

The honest limit

Automating a workflow does not shorten it. It moves who does it.

That is fine when the freed up time goes somewhere useful. It is not fine if the honest answer is that the forty minutes gets absorbed and nobody could tell you afterwards what changed.

So decide in advance what the time is for. If you cannot say, the automation will still work and you will never be sure it was worth it, which is how these things quietly stop being maintained.

And be careful automating something you have not done manually enough times to understand. If nobody in the business can explain why a step exists, that is not a step ready to be automated. That is a step ready to be questioned.