How to get started with AI in your company

You know artificial intelligence can help your company, but you are not sure where to start. That is the position of most SMEs: interest and a touch of vertigo in equal measure. The good news is that getting started with AI needs neither a large investment nor technical knowledge. It needs method. Here are the steps to take the first one properly.

Most companies wanting to start with AI get stuck in the same place: they do not know where to begin. They try scattered tools, someone on the team opens an account, some text gets generated and there it dies. This is the sequence that does lead somewhere.

Step 0: stop looking for tools

The most common starting mistake is beginning with the catalogue. Six applications get tried, each solves a fragment, none fits how you work and three weeks later nobody uses them.

AI is not a tool you install, it is a capability applied to a specific problem. Without the problem in front of you, the tool is useless.

Step 1: inventory what repeats

For a week, have each person note the tasks they do more than once that do not require much thought. Nothing formal is needed: a note will do.

Things nobody had ever put into words will appear. That is normal: repetitive work is invisible precisely because it has been normalised.

Ask for two figures per task: how many times a month and how many minutes each time. Without those two numbers, the rest of the process is guesswork.

Step 2: sort by hours, not by annoyance

Multiply times × minutes and sort. The resulting list almost never matches the one for “what bothers us most”.

The process that irritates most is usually the most visible. The ones that really consume time are silent: five minutes here, ten there, forty times a day.

To put figures on this you have the savings calculator: with times per month, minutes and hourly cost, it tells you how much each task costs per year.

Step 3: filter by the three conditions

From the top of the list, keep the ones that meet all three:

• Volume. It repeats a lot. Something happening twice a month rarely pays off.
• Pattern. It is done the same way almost always, and the variations can be listed.
• Explainable rules. Someone can say what is done and why.

If a process is missing the third, do not discard it: note it as a candidate for later. But it has to be documented first, and that work comes before.

Step 4: measure the starting point

Before touching anything, note three things about the chosen process: how long it takes today, how many times a month it runs and how often something needs correcting.

Without that initial number, any later improvement is an impression, and impressions get argued about. Five minutes that save you months of debate over whether this helped.

Step 5: automate just one

One. Not three.

There is an underlying reason: every automation changes how the team works, and teams absorb one change at a time. Going faster than people can take in is the surest way for the system to end up switched off “temporarily”.

And there is a practical reason: if you build three at once and the overall result is mediocre, you do not know which one failed.

Step 6: run alongside the manual process for a few days

Do not switch off the old way on day one. Let both run for a week. That way, if something does not fit, nobody is blocked and you have something to compare against.

This step is almost always skipped out of haste, and it is the one that prevents disaster when the first odd case appears.

Step 7: decide with data

After a month, compare with what you noted in step 4. And look above all at one figure almost nobody measures: how often someone has had to step in and fix it.

An automation that runs on its own 60% of the time and requires reviewing the rest is not saving time: it is moving it somewhere else.

What NOT to do in the first months

• Signing up for an expensive platform before having a clear use case. You will pay for capacity you do not use.
• Starting with the most delicate part. If the first project touches sensitive data or decisions with consequences and turns out mediocre, you burn confidence in everything else.
• Announcing it as a revolution. It creates expectations no first project meets, and fear in anyone who thinks their job is at stake.
• Leaving it all to one person. If that person leaves, the project leaves with them.

How long this takes

The inventory and prioritisation: a calendar week and a few hours of actual work. A specific automation: about seven days from closing the design. The first measurable result: a month.

If someone tells you they will transform your company in two days, be sceptical. And if they tell you it is a six-month project before you see anything, be sceptical too.

The shortcut

Steps 1 to 3 are exactly what we do in the free diagnosis: 45 minutes looking at your processes and leaving with a list prioritised by hours saved and implementation effort. You keep it even if you do not continue with us.

If you would rather do it yourself, here are the processes that work best and the mistakes that come up most.