A single thread of light that calmly builds a surface, step by step

Starting an AI project without turning your organisation upside down ​

The biggest mistake is starting big straight away. A broad project in an organisation that is not yet ready for it becomes expensive and slow, and the support disappears.

It can be done differently: small, focused and measurable. Six steps. Want to check first whether a task is suitable? See which repetitive work you can already have AI do today.

1. Choose one well-defined task ​

Not your whole administration, not a department. One task that comes back often. If you doubt, take the most boring and most repeated one: that is where the gain and the smallest risk sit.

2. Let AI run alongside your team ​

A few weeks alongside your people, not in their place straight away. Your team keeps checking, errors surface early, and people accept a new way faster when they first see it work.

3. Agree up front what success is ​

In three figures: hours saved, errors avoided, output delivered. Write it down. Without that agreement, "it works fine" stays a feeling.

4. Work in phases ​

Each phase gives a measurable result on its own. If it works, continue. If it does not, adjust or stop, with little loss. That way you only scale what has proven itself.

5. Involve your people ​

Technology is half of it. The other half is that your team understands and wants the new way. Explain why, show it together, leave room for questions. Count that time as part of the project, not beside it.

6. Decide on figures ​

At the end you look at the agreed figures, not at the feeling. If it is worth it, expand. If it is not, you lost little and learned a lot.

What you avoid this way ​

  • A big budget for something unproven.
  • Months of building before the first result.
  • A team that does not carry the new way of working.
  • Making an unclear process faster, not better.

Where possible, pay for a result, not for a big project up front. That way the risk sits in the right place.

Frequently asked questions ​

How long does a first phase take? Often a few weeks. Long enough for real figures, short enough to adjust quickly.

What if the first task does not work? Then you stop or choose a better one. Because you started small, the loss is small. That is the point.

Do I need to have everything in order first? No, but the chosen task must have a clear process. If not, read when AI is not worth it and put the process straight first.

Do I need technical people? Mostly someone who knows your process and follows the figures honestly. A partner can supply the technology.

Further reading ​

Coronis works in phases: one task, a measurable result, and only scaling up when it is worth it. You pay for what works.

Written with AI, read and approved by a human. Here is how that works.