AI projects at companies usually get stuck in one of two places. The first is the start: once someone says "it can do anything," nobody can settle on what to begin with. The second is the pilot: a small trial runs, everyone likes it, then nobody takes it into their daily work and the project quietly fizzles out.
To avoid both traps, this post splits the first 90 days into three phases. At the end of each phase, you should have a concrete deliverable in hand. If you don't, you don't move on to the next phase. I explained what AI transformation is and where to start in the AI transformation guide; this post is that guide laid out on a calendar.
90 days, three phases
Discovery
What's being done, where the time goes, which data can go where. Deliverable: a pilot task and a baseline measurement.
Pilot
A trial on a single task, with a small team, using real data. Deliverable: a weekly measurement table.
Decision and rollout
Continue, adjust or stop. Deliverable: a written decision and the plan for the next 90 days.
In this phase, no tool gets bought and nothing gets automated. The goal is to pick the right task.
Weeks 1-2: the task list. Ask each team to write down the five recurring tasks that take up the most time each week. For each task, three pieces of information: how many hours a week it takes, what data it's done with, and what happens if there's a mistake. While you're at it, also ask how the teams already use AI today; the answers are often surprising.
Week 3: data rules. Put in writing which data can go into which tool. Customer data, personal data and trade secrets are handled separately. If personal data is involved and you operate in Turkey, the guide from KVKK (Turkey's Personal Data Protection Authority), Generative AI and the Protection of Personal Data Guide (in 15 Questions) (in Turkish), is a good starting point.
Week 4: pilot selection and baseline measurement. Pick a single task from the list and measure how it's done today: how many minutes it takes to do once, how many times a week it's done, how many errors turn up in review.
A bad pilot
A good pilot
The pilot is kept small: one task, a team of two or three people, real data.
Week 5: setup. A tool is chosen for the pilot task and set up on a company account. Sometimes it's a few prompt templates prepared in a chat tool, sometimes an automation workflow, sometimes an assistant that answers from a document. The task determines the tool, not the other way around. If the task is an automation, see the "first workflow" rules in the n8n automation examples post.
Weeks 6-8: trial and weekly measurement. The team does the task the new way every time and records the same three numbers every week: time, number of repetitions, errors caught in review. Plus a short note: what didn't work this week?
| Week | Time per task | Weekly repetitions | Errors caught in review | Note |
|---|---|---|---|---|
| Baseline (week 4) | Old method | |||
| Week 6 | ||||
| Week 7 | ||||
| Week 8 |
Week 9: the decision. Put the baseline measurement next to the last week of the pilot. There are three possible decisions:
| Decision | When | Next step |
|---|---|---|
| Continue | Time dropped noticeably, errors didn't increase | Write down the method, roll it out to everyone who does the task |
| Adjust | There's a gain, but errors or friction are high | Change the template or the tool, extend the pilot by two weeks |
| Stop | No gain, or the risk outweighs the gain | Write down why, pick the second task from the list |
Stopping is a successful outcome too. Learning in two months, with a small team and with measurement, that a task isn't a good fit for AI is far cheaper than learning the same thing after rolling it out to the whole company.
Weeks 10-11: rollout. If the decision is "continue," the method is put in writing (which tool, which template, who reviews) and the other employees who do the task get a short training. I explained how to plan training by role in the AI training for employees post.
Week 12: the next 90 days. Go back to the task list from Phase 1 and pick the second task. This time it goes faster, because the data rule, the measurement method and the team are already in place.
| Phase | Deliverable | Who keeps it |
|---|---|---|
| Discovery | The teams' task list | Project lead |
| Discovery | Data rule (which data, which tool) | Management, all employees |
| Discovery | Baseline measurement of the pilot task | Pilot team |
| Pilot | Weekly measurement table | Pilot team |
| Pilot | Templates and workflows used | Pilot team |
| Decision | Written decision and its rationale | Management |
| Rollout | Written method and short training | All employees who do the task |
| Rollout | Plan for the next 90 days | Project lead |
This table is also useful when you're evaluating an AI consulting proposal: which of these deliverables does the proposal include, and which does it leave out? Apart from a presentation and a "strategy document," will you be left with anything concrete?
Is my 90-day plan ready?
0/7Compare the seven items with your own plan; every item you don't check is a job for the first two weeks.
In our AI consulting service, we run these 90 days together with you: discovery interviews, a draft of the data rule, choosing and setting up the pilot task, weekly measurement and the decision meeting in week 9. At the end of each phase, we hand over the deliverables in the table above in writing; we don't guarantee results, but we write down from the start what we'll measure and how.
The calendar stays the same; the workload shrinks. In a five-person business, the task list comes together in an afternoon and the pilot team is one person. What really matters isn't the phase boundaries but each phase's deliverable.
A task that repeats often, whose output a person reviews first, and whose time is easy to count. Proposal drafts, meeting summaries, product descriptions or sorting incoming emails are commonly chosen examples.
Write down why it failed: was the tool not good enough, was the task a poor fit, or was data missing? Then pick the next task from the list. The notes from a failed pilot make the second pilot much faster.
The first 90 days don't require a big investment: for the pilot, a subscription for a few users and the team's time are usually enough. The real cost is time; set aside the project lead's and the pilot team's weekly time from the start.
Yes. That's what the plan was written for. A consultant helps most in two places: choosing the right pilot from the task list, and making the week 9 decision impartially.
When you train employees on AI, giving everyone the same seminar doesn't work. We cover three role-based levels, the modules, a time plan, data rules and how to measure whether the training worked.
According to Comscore's September 22, 2026 report, ChatGPT's share of AI assistant conversations fell from 70% to 50% in six months; the same day, Claude Opus 5.5 launched at a lower price. The risk of an AI setup tied to a single vendor, and what to do today.
October 4, 2026: starting October 9, Google is limiting the Gemini app on free personal accounts to the Flash-Lite model only, and Pro is also going away on AI Plus.
We use cookies to improve the site and measure advertising performance. Analytics and advertising cookies run only if you accept them. Details are in our Privacy Policy and Data Protection Notice.
