Blog

Where to start with AI: the first 90 days in an Icelandic company

Breytt 6 min read
A man writing on a flip chart

The question I hear most from leadership teams isn't whether to use AI. It's where to start. The answer is rarely a big system or a thick consultancy report. It's a structured start that takes about three months and leaves the company with two things: results it can measure and knowledge it keeps.

Most Icelandic companies are small or mid-sized, and that works in their favour. Lines of communication are short, decisions get made quickly and it's easy to see whether something is working. The flip side is that few people have time to follow every development in AI while running the business day to day. That's why a simple plan everyone understands matters more than an ambitious one.

Here's the approach I recommend: three phases of roughly thirty days each.

Before you start: one person owns it

AI projects that nobody owns die quietly. Before anything else happens, one person needs to be accountable for the work. They don't need to be technical. It matters far more that they understand the business, have the authority to make decisions and get real time for the project, not whatever is left at the end of the week.

That person also needs backing from leadership. Not a nicely worded statement, but clear answers to two questions: What problem are we trying to solve? And how will we know in ninety days whether we solved it?

Days 1–30: Map what's happening and set the ground rules

The first month is about understanding where you actually are. In many companies people are already using AI, often without management knowing. They try chatbots on personal accounts, paste text into free tools and find their own workarounds. That's no reason to punish anyone. It's a reason to take stock.

Ask the team, anonymously, which tools they use, for what, and what data goes into them. The answers tell you where the enthusiasm is and where the risk is.

At the same time, you need simple ground rules. They don't have to be a long document. One page is enough if it answers three questions: Which tools are approved? What data must never go into them? And who reviews the output before it leaves the building?

Think about data protection from day one. In Iceland, the Data Protection Act (No. 90/2018) applies just as much when data is processed with AI, and it's far easier to set the rules up properly at the start than to fix them later.

By the end of the month you should have a list of repetitive, time-consuming tasks. We explain how to build that list in our article on zombie tasks.

Days 31–60: One well-defined project

The second month is about picking one task from the list and solving it properly. One, not five. It's tempting to spread the effort, but half-finished experiments add up to very little.

Five half-finished pilots deliver less than one that makes it into daily use.

A good first project meets four conditions:

  • It comes up often, ideally daily or weekly.
  • It relies on data you already have access to.
  • Mistakes are cheap and easy to catch, because a person reviews the result.
  • The benefit is visible, both to the people doing the work and to management.

Typical first projects include summarising meeting notes or reports, drafting replies to common enquiries, sorting incoming requests or producing a first draft of a quote. None of these needs a custom-built system. Often the tools the company already pays for will do the job, once they're set up properly.

Measure before you change anything. How long does the task take today? How often does it need redoing? Without that baseline you can't show results at the end of the period, and without visible results it's hard to win support for the next step.

Days 61–90: Make it stick and decide what's next

The final month is about turning an experiment into a way of working. This is where many projects stall. The solution worked in testing, but nobody wrote down how to use it, nobody is responsible for it, and it's forgotten the moment things get busy.

Write simple instructions. Train the people who will use the solution, not just the ones who tested it. Decide who looks after it from now on and what happens when it gets something wrong, because it will.

Then measure again. Compare the result with your baseline and share it across the company, including what didn't go well. An honest summary builds more trust than a polished presentation.

At the end of the ninety days you should be able to answer three questions: What did we learn? What do we want to do next? And can we carry on ourselves, or do we need help?

A checklist for the first 90 days

  1. Appoint an owner with both the authority and the time.
  2. Define the problem and how success will be measured.
  3. Map current AI use with an anonymous survey.
  4. Write one page of ground rules covering tools, data and review.
  5. List the repetitive, time-consuming tasks.
  6. Pick one task and measure the baseline.
  7. Test the solution in a real workflow for a few weeks.
  8. Write instructions and train the users.
  9. Measure again and share the results internally.
  10. Decide on next steps and who owns them.

Three common traps

Starting with the tool. Licences get bought for everyone before anyone knows what problem they're meant to solve. Usage becomes random, and nobody can say whether the investment paid off.

Starting too big. A project meant to transform the whole business in its first quarter takes too long, costs too much and delivers too little, too late. An early small win is worth more.

Forgetting the people. The technology is rarely the bottleneck. Trust, training and time are. A team that understands why the change is happening, and gets to shape it, adopts it much faster.

Next steps

If you want to move faster, two routes suit most companies. Our one-day workshop covers the first month in a single day: current AI use is mapped, ground rules are agreed and the list of time-consuming tasks is put together. Your AI Roadmap then builds on that work and turns it into a clear plan for what comes next.

And once the first solutions are in use, our always-on partnership helps you keep track of what matters, without spending your week reading about new models.