How to Implement AI in Your Company: The 90-Day Roadmap

Implementing AI in a company doesn't start by buying tools: it starts by measuring processes. This is the 90-day roadmap we use to take a company from zero to its first AI agents in production — with milestones, deliverables and the mistakes to dodge in each phase.

The premise that orders everything: in 90 days the goal is not to "transform the company", it's to have 2-3 processes operating with agents, measured against their baseline, and a team that knows how to supervise them. That — small, real and measured — is what unlocks everything else.

Days 1-30: diagnosis and baseline

Weeks 1-2 — Process inventory. Map the real operation, not the org chart: what gets done, how many times a month, how many hours it consumes, which systems it touches, how much judgment it requires. The source of truth isn't the manuals, it's the people who execute — half an hour with each key role beats any documentation.

Weeks 3-4 — ROI prioritization. Score each process on three axes: current cost (hours × fully-loaded rate), automatability (high volume + clear rules + manageable exceptions) and risk (impact of an error, compliance requirements). The month's deliverable: a prioritized list where the top 2-3 candidates have a measured baseline — hours/month, cost per operation, error rate.

Mistake to dodge: starting with the flashiest process instead of the most measurable one. The first project must prove its ROI beyond argument; there'll be time for the ambitious ones.

Days 31-60: first agents, supervised

Weeks 5-6 — Design and build. For each chosen process: define what the agent decides alone, what requires approval and what is off-limits; connect it to the systems where the work happens (email, CRM, ERP); and set up traceability — every agent action, logged.

Weeks 7-8 — Supervised production. The agent operates on real work while a person reviews its outputs before they take effect. This phase has a double purpose: calibrating the agent's accuracy and — just as important — building the team's trust, which shifts from skeptic to supervisor.

Mistake to dodge: the eternal pilot. Set the graduation criterion on day 31: "when the agent is right X% of the time for Y weeks, standard cases go autonomous". Without a written criterion, supervision becomes habit.

Days 61-90: bounded autonomy and measurement

Weeks 9-10 — Widen autonomy. Standard cases move to direct agent execution (with thresholds: amounts, operation types); exceptions keep going to people, each time with better context attached. This is where the saved hours start to really show.

Weeks 11-12 — Measure against the baseline and decide. The final deliverable: hours/month before vs now, cost per operation before vs now, error rate before vs now, and the annualized projection. With those numbers, the decision on the next wave (more processes, more autonomy, more functions) makes itself — and is funded by the savings.

Mistake to dodge: not measuring. Without the day-30 baseline, on day 90 you'll only have feelings — and budgets don't get renewed on feelings.

What you need to start (and what you don't)

You need: a sponsor with real authority, an internal project owner (they don't need to be technical), access to the systems where the work happens, and the willingness to measure honestly.

You don't need: a new ERP, a data lake, a data science team or an 80-page "AI strategy". Today's agents work on top of the tools you already use — as we explain in what an autonomous AI agent is — and the real strategy emerges from the data of the first 90 days.

This roadmap is the condensed version of phases 1 and 2 of the full transformation we describe in what an AI-First company is, and the heart of our consulting service.

Frequently asked questions

Can an SME without a technical team follow this roadmap?

Yes — it's the ideal case. The technical phases are covered by the partner; what the company brings is process knowledge, the internal owner and the discipline to measure. SMEs often reach day 90 faster than large companies: fewer committees.

How much does a 90-day program like this cost?

Diagnosis + 2-3 agents in production usually ranges €25,000-70,000 depending on complexity — detailed ranges in our breakdown of how much an AI agent costs. With well-chosen processes, the annualized return exceeds the investment.

What if the numbers don't work out at day 90?

If you measured well on day 30, by day 60 you already know whether it will — and you correct: another process, another scope, or honestly, stop. That option to stop cheaply mid-way is exactly what the phased approach buys.

What happens after day 90?

The second wave: more processes from the (already prioritized) inventory, more autonomy for the existing agents, and the start of the real organizational change — supervision roles, per-area training. Day 90 isn't the finish line; it's the end of the beginning.