Getting started with AI: your organisation’s first 90 days

Start with a problem, not a tool

The most common failure pattern is what I call “buy a tool and hope.” Someone signs an enterprise license, sends an announcement, and assumes adoption will follow. It rarely does. People do not change their daily habits because a new icon appeared in their toolbar.

Begin instead with a specific, annoying problem that real people deal with every week. Good candidates are tasks that are frequent, text-heavy, and low-risk if the first draft is imperfect. A few examples that work well in HR and operations:

• Drafting first versions of job ads, policies, or internal communications.
• Summarizing long documents, meeting notes, or survey responses.
• Answering repetitive employee questions about benefits or processes.
• Preparing interview questions or structuring feedback notes.

Pick one. A single use case that visibly saves a few hours per week is worth more than a vague ambition to “use AI across the business.” It gives you something concrete to learn from, and it gives skeptics something real to react to.

How do you get started with AI in an organisation?

How do you get started with AI in an organisation? Start with one recurring problem, a small pilot group and an approved tool, then run the first 90 days as a loop. Weeks 1 to 2: pick the use case, name an owner, set short guardrails and run a literacy session; two hours of practice on real tasks beats two days of slides. Weeks 3 to 6: use it on real work and note what helps and what needs heavy editing. Weeks 7 to 10: adjust prompts and process, add a second team or use case. Weeks 11 to 13: review results and decide what to keep, drop or expand. Good first use cases: drafting job ads and policies, summarising documents, answering repetitive employee questions. Track time saved, how often people choose the tool and quality after editing. In my State of AI in HR survey, 81% of Swedish HR professionals have employer-provided generative AI but only 33% have had formal training: literacy comes before scale.

Build a small amount of literacy first

People cannot adopt what they do not understand. Before you scale anything, give the team working on your first use case enough grounding to use the tools with judgment. This does not mean a long training program. It means a shared baseline: what these tools are good at, where they fail, what should never be pasted into them, and how to write a useful prompt.

Two hours of hands-on practice beats two days of slides. Have people bring a real task and work through it together. The goal is confidence and caution in equal measure. For a structured way to do this across roles, the AI literacy playbook lays out what different groups need to know.

Set simple guardrails early

You do not need a forty-page policy to begin. You need a short, clear set of rules people can remember: what data is off limits, when a human must review output, and who to ask when something feels wrong. Write it in plain language and revisit it as you learn. A light guardrail that people follow is better than a thorough one they ignore.

Involve leadership and managers, not just enthusiasts

Early AI use tends to cluster around a handful of curious individuals. That is a fine start, but it stalls if managers stay on the sidelines. Managers shape what their teams spend time on, and their silence reads as disinterest.

Leadership involvement does not mean executives need to become power users. It means they sponsor the effort visibly, protect time for people to learn, and ask about progress in normal meetings. When a manager asks “how did the AI draft work out this week,” it signals that this is part of the job, not a side experiment. This is the quiet difference between a pilot that fades and one that spreads.

How do you run the first ninety days as a loop?

Treat your first quarter as a short, repeating cycle rather than a one-time launch. A workable rhythm looks like this:

• Weeks 1 to 2: Pick the use case, gather the small group, set guardrails, and run a short literacy session.
• Weeks 3 to 6: Use the tool on real work. Capture what helps, what wastes time, and where output needs heavy editing.
• Weeks 7 to 10: Adjust prompts and process based on what you learned. Bring in a second team or a second use case.
• Weeks 11 to 13: Review results, share stories internally, and decide what to keep, drop, or expand.

Keep the group small enough that you can talk to everyone. The point of these ninety days is learning, not coverage. You are building an internal example you can point to, along with the habits and judgment to repeat it elsewhere.

Measure early wins honestly

You need evidence, but you do not need a sophisticated dashboard. Pick two or three signals that match your use case and track them lightly:

• Time saved on the specific task, estimated by the people doing it.
• How often people choose to use the tool when they could skip it.
• Quality of output after editing, judged by someone close to the work.

Be honest about the misses too. A tool that produces drafts nobody trusts is a finding, not a failure. The teams that succeed are the ones that talk openly about what did not work and adjust, rather than declaring victory and moving on. Early wins are useful mainly because they fund the next experiment with credibility.

What are the common traps to avoid?

• Boiling the ocean. Trying to transform everything at once spreads attention too thin to learn anything.
• Skipping literacy. Handing people tools without context produces either fear or careless use.
• No owner. If the effort belongs to everyone, it belongs to no one. Name a person.
• Measuring nothing. Without a few honest signals, you cannot tell progress from noise.

Getting started well is not dramatic. It is one real problem, a small capable group, clear guardrails, visible support from managers, and an honest look at the results. Do that once and you will have something far more valuable than a tool license: a repeatable way to bring AI into the work. For the wider picture of how this connects to strategy, change, and capability, see the AI adoption hub.