Opens in a new tab
needsahuman.

How to implement AI in business: your first 90 days

How to implement AI in business without wasting money or trust? Start small, measure, write the rules down, then extend. This lesson turns the course into a 90-day roadmap. In this lesson A plan in four stages, from learning and listing to extending what works. What each stage produces, and who owns it. How the…

How to implement AI in business without wasting money or trust? Start small, measure, write the rules down, then extend. This lesson turns the course into a 90-day roadmap.

In this lesson

  • A plan in four stages, from learning and listing to extending what works.
  • What each stage produces, and who owns it.
  • How the plan changes depending on how far you choose to go: assisting your team, reshaping roles, or the far end.
  • A readiness checklist to run before you start.

How to implement AI in business: the plan at a glance

Thirteen weeks, four stages. Each stage ends with something written down, so you can see progress and decide what comes next.

Four phases over 13 weeks, each with its output.

  1. Weeks 1–2Learn and listOutput: everyone has tried one assistant's business plan on real work; a list of the team's repeated tasks.
  2. Weeks 3–6Pilot two tasksOutput: two tasks chosen with the task test, each with a named owner and a before and after time log.
  3. Weeks 7–10Review and write the AI use policyOutput: keep, change or stop for each pilot; a one-page AI use policy.
  4. Weeks 11–13ExtendOutput: the next tasks started, perhaps one agent or bot, and a date for the next review.
Thirteen weeks, four phases, one output each. On wider screens, segment widths follow the weeks in each phase.

Weeks 1–2: learn and list

The goal is for everyone to get a feel for what AI can and cannot do, on real work, and to list the tasks worth trying.

  • Choose one assistant on a business plan for the people who will try it. A business plan matters because of how it handles your data (Lesson 2). One tool is enough to start.
  • Set two simple rules straight away: what must never be pasted in (customer and staff personal details, passwords, anything confidential), and that every output is a draft a person checks. You will write a full policy in week 7.
  • Everyone tries it on their own work for two weeks: drafting emails, summarizing documents, planning, research. Ask people to note what worked and what went wrong.
  • List the repeated tasks. Each person lists the tasks they do every week, roughly how often, and how long they take. Combine the lists. This is the list from Lesson 4’s “Try this”.
  • Look up your roles in the AI team planner and compare its task lists with yours.

Output: a list of repeated tasks with frequency and time, and notes on what people found.

Weeks 3–6: pilot two tasks

The goal is to test two tasks properly, with a before-and-after measure.

  • Pick two tasks using the five-question test from Lesson 4: frequent, mostly on a screen, quick to check, and cheap to get wrong. Avoid anything customer-facing for the first pilot unless a person checks every output.
  • Name an owner for each: usually the person who does the task today. They set it up, check the output and keep the log.
  • Write the instructions down: what the AI is given, what it should produce, what it must never do, and how the output is checked.
  • Measure before and after with the two-week log from Lesson 11: time per task including checking, and mistakes caught.
  • Hold a short weekly check-in of fifteen minutes: what went wrong, what to change.

Output: for each pilot, the measured time and error figures before and after, and the owner’s view.

Weeks 7–10: review and write the policy

The goal is to decide what to keep, and to put the rules in writing.

  • Keep, change or stop each pilot, based on the measurements, not on how it felt. Stopping is a good result if the numbers say so.
  • Write your AI use policy, using the template in Lesson 10: approved tools, what data may go in, who checks what, approvals for actions, decisions about people, telling customers, and who to ask.
  • Share the results with the whole team: what was tried, what it saved, what went wrong and what happens next.
  • Update your team chart from Lesson 5 with the AI tools you are keeping and the name of each checker.

Output: a written policy, an updated chart, and a decision on each pilot.

Weeks 11–13: extend

The goal is to build on what worked, carefully.

  • Pick the next two or three tasks from your list, using the same test and the same measure.
  • Consider one agent or bot, if a pilot showed a good fit: for example, an AI phone answerer out of hours, or an agent that drafts quotes for approval. Start with approval on every action (Lesson 7) and a clear hand-over to a person (Lesson 9).
  • Plan training for anyone who has not yet used the tools, led by the people who ran the pilots.
  • Set a review date in three months for the whole setup: tools, costs, policy and results.

Output: the next round of pilots, with owners and measures, and a review date.

How the plan changes with how far you go

Our AI team planner has three settings. The 90-day plan above works for all three, with these changes.

Low: assist the team

Stop at assisting. Every task stays with the person who does it today, and AI helps them do it faster or better. The plan above is complete as it stands; the result is time freed in people’s weeks. Decide what that time is for: more customers, faster replies, the work you never get to.

Medium: hand over and reshape

Add a reshaping step in weeks 7–13. Some tasks will be handed over to AI with spot checks, and some roles will end up with much more free time or mostly checking work.

  • For each role that changes, agree with the person what they will do with the freed time. Write it down.
  • Retrain where the role is moving towards new work: checking and supervising AI, more customer time, or skills the business lacks.
  • Use natural turnover. When someone leaves, ask whether the role should be refilled as it was, reshaped, or combined with another, before advertising it.
  • Keep the checkers. Every handed-over task needs someone who knows the work well enough to spot mistakes.

Scorched earth: the far end

If you are considering the planner’s far setting, “replace everyone you can”, the plan needs the longest runway. Even here, the planner shows the work in your team that our data says still needs a person, and every AI’s output needs someone to check it.

  • Run AI and people side by side until you have measured the AI’s error rate on your own work over weeks, not days, including the unusual cases. The research in Lesson 3 shows how differently AI can perform on tasks that look alike.
  • Keep a fallback for every task: the people, the instructions and the data to do it without the tool if it fails, changes its terms or raises its price.
  • Read the “Cutting too far” section of Lesson 10, including the company that announced an AI assistant was doing the work of hundreds of agents and later said it was hiring people again so customers could always reach a person.
  • Take the legal and consultation steps set out in Lesson 10 before any role changes. They take weeks, and in some cases months; they are not optional. Take advice.

AI readiness checklist

Before you start, check you have these in place. (The boxes are there to tick as you read; they are not saved, so print the page or copy the list if you want to keep track.)

Where to go from here

  • Look up your team in the AI team planner, and try it at each setting.
  • Find tools by role on our AI tools pages, with prices as the vendors list them and the tasks each tool helps with. We have not tested them and do not rank them.
  • Look up any job in the rankings to see its task list and how much of it still needs a person.
  • Look up a word in the glossary.
  • Go back to any lesson from the course contents.

Key terms

Every term is explained in the glossary.

Try this

Put the four stages in your diary now: a start date, the end of week 2, week 6, week 10 and week 13. Next to each, write the output that stage produces and the name of the person who owns it. A plan with dates and names happens; a plan without them usually does not.

Sources

This lesson uses no outside figures. It draws on the studies and rules cited in Lessons 3, 9, 10 and 11.