Generative AI for companies: where to start without an endless project

30-second summary

• Start with a boring, repetitive, measurable process. Not the most impressive one.
• If you cannot say how much time it consumes today, it is not ready to automate.
• Projects that take more than six to eight weeks to show value usually die.
• The hard part is rarely the AI. It is company information being organized.
• Generic tools solve little. The gain comes from connecting AI to your data and processes.

Why most projects stall

It is not a technology problem. It is a scope choice.

Companies start with what looks most transformative: an assistant that "knows everything", a copilot for everyone, a support overhaul. Six months later there is a nice demo and no number.

The pattern that works is the reverse. One process, one owner, a short deadline, a number at the end.

How to choose the first case

Look for a process that meets four criteria:

• Repetitive. It happens many times a week.
• Boring. Nobody will argue it is creative work.
• Measurable. You can say how much time or money it consumes today.
• Error tolerant. If it goes wrong once, nothing serious breaks.

Common candidates: message triage, meeting summaries, first response to leads, organizing information that lives in spreadsheets, preparing reports. We listed several in 7 processes to automate today.

What actually blocks it

In most companies the obstacle is not AI, it is the state of the information.

Knowledge sits in three people's heads, in chat threads and in folders with four versions of the same document. No model fixes that on its own: it can only answer well if it has something to read.

Organizing that base is usually half the project, and it separates an assistant that gets things right from one that invents. That is the concept behind RAG.

Off-the-shelf tool or custom solution?

Generic tools solve generic tasks: writing, summarizing, translating. That already helps and costs little, and it is worth giving the team access with guidance, as we discussed in training your team on AI without creating dependency.

The big gain, though, appears when AI talks to your data and your systems: the CRM, the sales spreadsheet, the support history. That is when it stops giving generic advice and starts answering about your operation.

How to measure whether it was worth it

Define the number before starting. It is usually one of these:

• Hours per week returned to the team
• Response time to customers
• Percentage of tasks resolved without a human
• Errors avoided

Without a number defined upfront, evaluation becomes opinion, and opinions about new technology tend to swing between excitement and disappointment without touching reality.

How much time to allow

If nothing is running and being used within six to eight weeks, the scope was too large. Cut it in half and ship.

One small working case unlocks more inside a company than any presentation, because the team stops debating whether it works and starts asking for the next one.

The mistake of starting with the biggest problem

The company's biggest problem is usually the most complex, the most political and the least measurable. It is the worst possible place for a first project.

Start with the second or third problem. Once it is solved and measured, you will have internal credibility and experience to tackle the first.

If you want that design done with someone who has been through it, that is the work of area next, the AI consultancy of the area one hub.

Frequently asked questions