How to Implement AI in Your Company: a 90-Day Plan

The plan is split into three stages of about 30 days, each matching a phase of Encorp AIM: foundations (Enablement), a roadmap with PoCs, and implementation with fine-tuning. This way the company does not jump to technology before it knows where and how AI is used, and does not roll out anything that has not been tested on a real process.

Each stage ends with a decision: continue, change direction or stop. That is why the 90 days are the first cycle of the transformation, not its end — a larger scope is planned in packages of 3, 6 and 12 months.

The plan is for a company that is starting out; if you already have experience, skip the stages you have passed. The 30/30/30-day split is a planning guideline — real timelines depend on team size and scope.

The first task is an inventory: who in the company uses which AI tools and systems, for what, and from which vendor. The Commission names exactly this as the first step when building an AI literacy programme (Commission Q&A). It is worth including use outside approved tools, if there is any.

The second task is rules: which tools are allowed, what data must not be entered and who is responsible. Also check that no use falls under the prohibited practices in Article 5 of the EU AI Act — the prohibitions apply from 2 February 2025, except a few new points that apply from 2 December 2026 (Regulation 2026/1744).

The third task is training by role. Article 4 requires AI literacy measures (see the Article 4 guide); the Commission does not require a certificate but allows an internal training record. In Encorp AIM this is phase 1 — see AI training for companies.

In the second stage the company chooses the processes on which it will test AI. The criteria are practical: the process is repetitive, data is accessible, the risk is limited, and there is a person who knows it and will answer for the result. The processes need not be many — it matters more that the choice is deliberate.

Before the start the baseline is measured — how long the process takes, how many errors it has, what it costs. Without such a measurement a PoC cannot prove anything. Results are assessed against it, and the measurement method is recorded so every number can be explained.

The stage ends with a roadmap: which PoCs proved useful, which did not, and what comes next. Here AI business process automation is used; if the solution lives in Microsoft 365, also see the Microsoft Copilot track.

In the third stage the solutions that proved their value move into daily work. Implementation includes fine-tuning: instructions, data and review rules are adjusted to how people actually work with the solution.

Every rolled-out solution has an owner in the company and a plan to run it: who watches quality, who reacts to an error, and how the solution changes when the process changes. This is the essence of running AI systems — see AI Ops.

It must also be clear who is responsible for the rules. If the company has no one with enough experience, an AI Director is useful here — Encorp AIM has one lead every package. Rules and compliance are the subject of AI governance and the EU AI Act.

Solutions that have not proven their value are not rolled out — they stay on the roadmap as ideas for a later stage.

Progress is measured at three levels. First, training coverage: how many of the people in scope completed training — a number computed from the internal record. Second, use: which teams actually work with the approved tools. Third, the effect on the process: the difference from the baseline, measured with the recorded method.

Numbers without a method are not used in decisions. If you cannot explain how a number was obtained, it must not justify either expanding or stopping a project.

At the end of each stage leadership reviews the three levels and takes the decision: continue, change or stop. So the 90 days end with a reasoned recommendation for the next scope — for example a package for one team or several.

Indicators are few and clear — one for each level and each process — so leadership can follow them without additional reporting.

One of the main traps is starting from the technology: if the choice of tool comes before the analysis, the company pays for a solution to an undefined problem. The second is training “for the record” — the Commission notes that merely relying on a system’s instructions for use may be ineffective (Commission Q&A).

Other common problems are organisational. They are avoided with a few simple rules, listed below.

Training delivers results when it is tied to a real task: people should leave it with at least one thing they will do differently tomorrow. That is why role-based training — for leadership, sales, finance, HR and IT — is more useful than one general course for everyone. It also helps to limit the number of simultaneous pilots: teams have limited attention, and each pilot needs someone to follow it.

The plan is built along the phases of Encorp AIM: analyses and training, a roadmap with PoCs, implementation with fine-tuning. The 90-day period is close to the shortest package — 3 months — but the scope of a package is defined by the described results, not by the calendar.

If you first want to find out where you stand, start with the free AI readiness test; if you want a strategy before a package, start with the AI Strategy Sprint. One package covers a team of up to 20 people and is led by an AI Director.

The path of the method is described on AI transformation for companies as a funnel: free test, training, sprint, package, automations and running the systems. The 90-day plan is a practical way to cover the first part of it — even before you have chosen a package.