# The First 90 Days of Leading AI Change: What to Do First

Canonical URL: https://blog.theaieducator.io/posts/first-90-days-of-leading-ai-change
Publication: Dan Fitzpatrick Insights
Author: Dan Fitzpatrick
Topic: AI Leadership and Strategy
Published: 2026-09-15T14:16:58.000Z
Modified: 2026-09-15T14:18:36.809Z

Most new AI leads start with a policy or a training day. Both answer questions the organization has not asked yet. Here is what your first 90 days should actually produce, and the order that makes it hold.

## In brief

In your first 90 days of leading AI change, do three things in this order: find out what is already happening, make the four decisions only you can make, then change one visible thing and name what stopped. The order matters more than the speed. A policy or a training day written first describes an organization you have not met yet, because your people started without you: an IBM study published on September 2, 2026 found 76 percent of U.S. middle school educators report weekly classroom AI use, while Gallup found in May 2026 that just 18 percent of teachers receive any formal guidance on how AI should be used at work.

## Key takeaways

- The first 90 days of leading AI change should produce decisions, not documents: four sentences a member of staff can act on beat a twenty-page strategy nobody reads.
- You are not starting from zero. An IBM study published on September 2, 2026 found 76 percent of U.S. middle school and 73 percent of high school classroom educators report AI is used in their classroom at least weekly.
- Gallup found in May 2026 that just 18 percent of U.S. K-12 public school teachers receive formal guidance on how AI tools should be used at work, with 34 percent receiving none at all.
- The gap between high use and low guidance is a permission problem rather than a training problem, so a training day scheduled first produces confident staff who still do not know what they are allowed to do.
- Dan Fitzpatrick's suggested sequence, The Ninety-Day Order, runs: days 1 to 30 find out what is already happening, days 31 to 60 make the decisions only you can make, days 61 to 90 change one visible thing and name what stopped.
- Four decisions belong to the leadership team alone: what AI is for here, what it will not be used for, who approves a new tool, and who answers when it gets something wrong.
- Waiting for clearer guidance misreads what has already been published: England's Department for Education states that schools and colleges are free to make their own choices about suitable use cases, as long as they comply with their wider statutory obligations.

You have been handed AI. Not a team, not a mandate: the subject. Somebody said at a leadership meeting that the organization needs to get a grip on this, the room looked at the person most likely to say yes, and now it sits on your list between the data return and the parents' evening.

Before you write anything, understand that you are not starting. Your people started without you. In the United States, 76 percent of middle school and 73 percent of high school classroom educators say AI is used in their classroom at least weekly, according to [an IBM study of 1,019 K-12 education professionals published on September 2, 2026](https://newsroom.ibm.com/2026-09-02-new-ibm-study-finds-ai-adoption-is-outpacing-K-12-readiness). Whatever you decide in your first week, you are deciding it about a practice that is already running.

So here is the answer, before the detail. In your first ninety days, do three things in this order: find out what is already happening, make the decisions only you can make, then change one visible thing and name what stopped. The order carries more weight than the speed. Almost every stalled AI effort I am asked to look at got the three roughly right and the order exactly backwards.

## What your first 90 days should actually produce

Your first ninety days should produce decisions, not documents.

That is an uncomfortable target, because a decision is small, it is hard to show at a governors' meeting, and it does not look like ninety days of work. A written strategy looks like ninety days of work. A launch event looks like ninety days of work. Neither of them tells a teaching assistant on the second Monday of November whether she may put a parent's email into a chatbot to help her draft a reply.

Judge yourself at day ninety on four things. Can you say, in a sentence each, what AI is for in this organization and what it is not for? Can a member of staff answer the question above without asking you? Is there a named person, not a committee, who answers when AI gets something wrong? And has one thing actually stopped, because the machine now does it?

If the answer to all four is yes, you have had a strong quarter, even if you have produced nothing anyone could laminate.

## What is already true before you start

Two things are already true in almost every organization: people are using AI, and nobody has formally told them what they may do with it.

The second half of that is measured. [Gallup found that just 18 percent of U.S. K-12 public school teachers receive formal guidance on how AI tools should be used at work](https://news.gallup.com/poll/710534/teachers-receive-no-formal-guidance.aspx), with about a third (34 percent) receiving no guidance at all and roughly half (48 percent) receiving only informal guidance. That survey reached 2,069 teachers between February 9 and March 2, 2026, and was published on May 26, 2026. Gallup's earlier work in the same program found six in ten teachers use AI for their work, including three in ten who use it at least weekly.

Hold those two numbers next to each other. Six in ten are using it. Fewer than two in ten have been told anything formal about it. The gap between those figures is not a training problem. It is a silence, and people have been filling it with their own judgment for about two years.

The same IBM study found that only 20 percent of K-12 educators say they have received extensive AI training, and that lack of training or professional development is the top barrier educators name, at 42 percent. Only 24 percent of educators said the education system is adapting very well to advances in AI.

Here is where I part company with the obvious reading of that data. The barrier people name is training, so the move that looks responsive is a training day. But staff who have had no formal word from their employer about what is permitted do not primarily need to be taught how to write a better prompt. They need somebody to answer a question they have been quietly answering for themselves. Training a workforce that has not been given permission produces confident people who still do not know whether they are allowed.

## The Ninety-Day Order

The Ninety-Day Order is the sequence I give anyone who has just been made responsible for AI in an organization: first find out what is already happening, then make the decisions only you can make, then change one visible thing and name what stopped. The order is the whole of it. Each phase is close to worthless out of turn, because a decision made before you know what people are already doing is a guess, and a change made before a decision is a pilot nobody has agreed to.

It is a suggested sequence rather than a tested method, drawn from the pattern I see repeated across the leadership teams I work with. Take it as a way of ordering your attention, not a standard anyone is holding you to.

### Days 1 to 30: find out what is already happening

Spend the first month finding out, not deciding.

You are looking for the real practice, not the reported one. Which tools are open on people's screens. Which jobs they are using them for. Which of those jobs used to be done a different way, and who noticed. Where the personal accounts are, because there will be personal accounts. What people say they would never put into one, and whether that matches what they do at half past four on a Thursday.

Ask individuals, not the room. In a staff meeting you will hear what people believe they are supposed to say, which is usually a mix of enthusiasm and disclaimers. In a corridor conversation you will hear what they actually do. Both are useful, and only one of them is information.

The diagnostic question for this phase: could you name, today, three things people in this organization are already doing with AI that nobody formally approved? If you cannot, you have not finished looking. The answer is never none, and a leader who believes it is none is describing an organization they do not currently work in.

### Days 31 to 60: make the decisions only you can make

In the second month, make the small number of decisions that no consultant, vendor or national framework can make for you.

There are four, and they fit on one side of paper. What AI is for here. What we will not use it for. Who approves a new tool. Who answers when it gets something wrong. Those four sentences do more work than a twenty-page strategy, because each of them settles an argument that would otherwise be had again every fortnight by people with no authority to settle it.

Write them in the words your organization already uses, and write them so that they could be wrong. "We will use AI to reduce the time staff spend preparing materials, so that time goes back into contact with students" is a decision, because in six months you can look and see whether it happened. "We will harness the potential of AI to enhance teaching and learning" is not a decision. It is a mood, and nobody can fail it.

The test to apply here is [The Monday Test](https://blog.theaieducator.io/posts/how-to-create-an-ai-strategy-people-actually-use), which is a single question to ask of any AI strategy before it is signed off: could a member of staff read it and know what to do differently on Monday morning? If the four sentences pass that, you have a working strategy and you did not need ninety pages for it.

The diagnostic question for this phase: could a member of staff read what you have decided and know what they may do on Monday, without asking you? If they would have to ask you, you have written a set of principles and given yourself a new job answering questions about them.

### Days 61 to 90: change one visible thing, and name what stopped

In the third month, make one change people can see, and say plainly what has stopped as a result.

One is deliberate. A first quarter that changes eight things changes none of them, because nothing gets the attention it needs to survive contact with a busy term. Pick the change with the highest ratio of visibility to risk: something that a lot of people touch, where a mistake is easy to spot and easy to undo, and that nobody will mourn.

The second half of the sentence is the half people skip. If nothing stopped, nothing was adopted: it was added, and added work is how AI becomes the thing everybody resents. So name it. The old template is retired. That form is no longer collected. That report is now drafted by the machine and checked by a named person, and the four hours it used to take are going back into something you can name.

The diagnostic question for this phase: can you name one thing that has stopped, with the date it stopped and the person who agreed it would? If you cannot, you have run a pilot, not a change. There is a longer account of why that difference decides everything in [what makes AI pilots go nowhere](https://blog.theaieducator.io/posts/why-ai-pilots-go-nowhere).

## Why not start with a policy or a training day?

Because both of them answer questions your organization has not asked yet.

A policy written in week one describes a school you have not met. You will write it from the sector's anxieties rather than your own practice, it will ban things nobody was doing and permit things nobody understands, and within a term the one member of staff who read it will quote it back at you. Write the four sentences first. The policy, when it comes, is then a record of decisions you have already made rather than a prediction of decisions you might.

The training day is the more tempting error, because everybody wants it and it is genuinely popular. But training moves a skill into a person. It does not move a decision into an organization. Send thirty people on a course while the permission question is still open and you get thirty people who are better at something they remain unsure they are allowed to do.

There is also a version of this problem that dresses itself up as prudence: waiting for clearer guidance. In England, the Department for Education's policy paper [Generative artificial intelligence (AI) in education](https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education), last updated on August 12, 2025, states that "schools and colleges are free to make their own choices about the most suitable use cases for generative AI tools in their settings, as long as they comply with their wider statutory obligations." Read that again if you are waiting. The guidance arrived, and what it says is that the decision is yours. Guidance in most jurisdictions now works the same way: it draws the edges of the field and leaves you to choose where to stand in it.

## What I Tell Leadership Teams

What I tell leadership teams, before anything else, is that their AI lead does not have an authority problem, they have an order problem.

In advisory work with education systems, including the Department for Education in England, the KHDA in Dubai and the Ministry of Education in Kazakhstan, the same pattern turns up at every scale. The instinct at system level and at single-school level is identical: publish something, then train people, then find out what is happening. Run in that order, the finding-out stage becomes an audit, which is to say people tell you what they think you now want to hear, and you have lost the only honest month you were ever going to get.

The other thing I tell them is about the person, not the plan. If you have just appointed somebody to lead this, look at what has come off their plate. A new responsibility with nothing removed is not a delegation, and the ninety days you are expecting will be spent on the job they already had. The question of what is actually being handed over is worth settling before the clock starts, and it is the subject of [who should own AI strategy and why one AI lead is rarely enough](https://blog.theaieducator.io/posts/who-should-own-ai-strategy).

## When this order is wrong

This order is wrong in two situations, and both are worth naming plainly.

The first is a live problem. If personal data is going into an unapproved tool this week, or something has already gone wrong with a student, you do not spend a month observing. You make one decision immediately, in writing, and you go back to the sequence afterwards. Discovery is a luxury you get when nothing is on fire.

The second is the genuinely cold start: a small organization where nothing is happening at all, nobody has an account, and the honest answer to the month-one question really is none. There the risk inverts. Your first thirty days are short, and your problem is not an unmanaged practice but an absent one, so the visible change moves earlier and you use it to create something to decide about. That situation is rarer than leaders think, so check it rather than assume it. [The seven dimensions of AI readiness](https://blog.theaieducator.io/posts/seven-dimensions-of-ai-readiness) gives you the questions to check with, and [what it actually means to be AI ready](https://blog.theaieducator.io/posts/what-does-it-mean-to-be-ai-ready) explains why your readiness is your lowest dimension rather than your average.

## Where this leaves you

Ninety days is enough time to turn an unmanaged practice into a decided one. It is not enough time to transform an organization, and any plan that promises you both is selling the second to avoid doing the first.

So the honest version of the target is modest. At day ninety, the people you lead should know what AI is for here, know what they may do with it, know who to tell when it misfires, and be able to point at one thing that is genuinely different. Nobody is waiting for you to start. They are waiting for you to say something about what they have already started.

If your leadership team is working through this sequence, or has discovered at day sixty that it started at the wrong end, this is the kind of work I support through [AI strategy sessions and advisory work with schools and organizations](https://www.theaieducator.io/ai-strategy-for-schools).

## Sources and further reading

- IBM, [New IBM Study Finds AI Adoption Is Outpacing K-12 Readiness](https://newsroom.ibm.com/2026-09-02-new-ibm-study-finds-ai-adoption-is-outpacing-K-12-readiness), September 2, 2026. Survey of 1,019 K-12 education professionals and 1,029 parents conducted by Morning Consult in July 2026, margin of error plus or minus 3 percentage points.
- Gallup, [Most Teachers Receive No Formal Guidance on AI Use](https://news.gallup.com/poll/710534/teachers-receive-no-formal-guidance.aspx), May 26, 2026. Survey of 2,069 U.S. K-12 public school teachers, February 9 to March 2, 2026, part of the Walton Family Foundation-Gallup K-12 teacher research program.
- Department for Education, [Generative artificial intelligence (AI) in education](https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education), GOV.UK policy paper, last updated August 12, 2025. Applies to England.

*Dan Fitzpatrick is the founder of The AI Educator and works with schools, trusts and organizations on AI strategy, readiness and leadership. [More about Dan](https://www.theaieducator.io/about).*


## Frequently asked questions

### What should a leader do first about AI?

Find out what is already happening before deciding anything. Most organizations already have unapproved AI use running quietly, so a month of honest discovery tells you what your decisions actually have to cover. Deciding first means writing rules about a workplace you have not met, which people then quietly ignore.

### Do we need an AI policy before staff can use AI?

No, and writing one first usually makes things worse. A policy drafted before you know what staff are doing bans things nobody was doing and permits things nobody understands. Make four decisions first: what AI is for, what it is not for, who approves tools, and who answers when it goes wrong.

### Should we train staff on AI first?

Training first is the most tempting mistake. Training moves a skill into a person, but it cannot move a decision into an organization. Staff who have had no formal word about what is permitted do not need better prompting technique; they need somebody to answer the permission question they have been answering themselves.

### How long does it take to lead AI change in an organization?

Ninety days is enough to turn an unmanaged practice into a decided one. It is not enough to transform an organization, and any plan promising both is selling the second to avoid the first. Judge the quarter on decisions made, permissions understood, accountability named, and one thing that genuinely stopped.

### How do you know whether the first 90 days worked?

Ask four questions at day ninety. Can you say what AI is for here and what it is not for? Can staff answer that without asking you? Is there a named person who answers when AI gets something wrong? And has one thing actually stopped because the machine now does it?

### What does the Department for Education say schools must decide about AI?

The DfE's policy paper on generative AI in education, last updated August 12, 2025, states that schools and colleges are free to make their own choices about the most suitable use cases, as long as they comply with their wider statutory obligations. In England, the decision sits with the school.

### Who should lead AI in a school or organization?

Somebody with the authority to make decisions, not just the enthusiasm to research them. Check what has come off that person's plate: a new responsibility with nothing removed is not a delegation, and the ninety days you expect will be spent on the job they already had.

---
Source: [The First 90 Days of Leading AI Change: What to Do First](https://blog.theaieducator.io/posts/first-90-days-of-leading-ai-change)
Publisher: [The AI Educator](https://theaieducator.io)