AI in your organisation: start with what you want it to make possible

The demonstration is impressive.

A document appears. A presentation takes shape. Something that used to consume an afternoon seems to have happened while you were finding the right mug.

It is very easy, at this point, to buy a subscription.

The question I would sit with a little longer is what you want the organisation to become able to do.

Give somebody more time with a customer? Let a small team try an idea that used to need a department? Help the person with a good instinct turn it into something others can understand?

Those are interesting beginnings.

What do you mean by AI adoption?

For me, useful AI adoption means changing some actual work so that people can do something worthwhile more easily.

An account, a training session and a rather enthusiastic announcement can help. The test comes afterwards, in the ordinary working day.

Can someone point to a task that became easier? Can the person relying on the result trust it? Has any of the saved time reached the thing it was meant to serve?

You can have a great deal of AI activity without answering those questions.

Look at the work before improving the machinery

Imagine a report that takes three people most of Friday to prepare.

You could experiment with producing a first draft more quickly. Before doing that, I would ask who reads it, what decision it informs, and whether all of it still needs to exist.

Perhaps the report is essential. Perhaps two pages carry the useful part. Perhaps it has become a weekly reassurance that everyone is busy.

Each answer leads somewhere different.

The order in eliminate, simplify, automate, delegate still matters. A faster version of an unnecessary task remains an unnecessary task. It may now produce more of itself.

Keep people able to think

Human agency sounds rather grand. I mean something quite ordinary by it.

The person doing the work can understand what is happening, question the result, and choose a different course when the situation calls for it.

If a tool suggests an answer, someone still needs to decide whether that answer belongs in this particular conversation, with this particular person, on this particular day.

Give people permission to say that the new arrangement is making things worse. Listen when they do. The receptionist and the newest member of staff may notice a problem long before the person who commissioned the system.

And be honest about the intention. If roles may change, that is a conversation with people whose lives are involved. Cheerful talk about freeing everyone for creative work cannot carry it for you.

Try one change you can learn from

Choose one recurring piece of work with someone who actually does it.

Describe what a useful improvement would look like. Try a small version. Compare it with how the work happened before, including checking, correcting and explaining the result.

Then ask what became possible.

Perhaps you keep the experiment. Perhaps you simplify it. Perhaps you stop, having learnt something worth knowing.

The learning is useful even when the tool isn't.

I am excited by the possibilities. I also know how easily the possibilities can become another job, which is part of how I accidentally became the helpdesk.

I would quite like these tools to help us enjoy the future, with enough room left to live in it.

If your organisation has an important question about what AI could make possible, bring me that thing.