AI can brief you before you meet anyone and sort your notes afterward. It can't do the listening in between, and that's the part the job is built on.
Don’t ask AI for a 90-day plan built from your team’s files. Ask it to brief you before you meet people, draft questions for your one-on-ones, and sort your notes into themes afterward. Then confirm every theme with two people, pick two small reversible changes, and check in month four whether they’re still in use and whether the work improved.
Say you were the strongest customer support specialist on an eight-person team, and on Monday you became its lead. By Wednesday you’ve got forty-one files in the shared drive, a six-page note from the person who had the job before you, and a dashboard with fourteen metrics on it. It’s tempting to drop the lot into an AI tool and type, “Build me a 90-day plan for this team.”
You’ll get one, and it will look good. It’ll have phases, priorities and a tidy list of quick wins, all built from what the documents say. What the documents don’t say is that two of your best people are quietly covering a vacancy nobody has mentioned, or that the Thursday ticket cut is treated as a hard deadline by half the team and as a suggestion by the other half. A plan built from paperwork tells you about the paperwork.
Here’s the idea this guide is built on. In your first 90 days, use AI to prepare you for conversations and to organize what you hear in them, and don’t let it stand in for either one. Before you meet anyone, it can brief you so your first questions are good ones. After you’ve met them, it can sort a pile of messy notes into themes. The listening is the job, and it’s the one part that can’t be handed over.
Below you’ll find a phase-by-phase plan, three prompts you can paste, a table that splits the work between you and the AI, and a way to tell in month four whether any of it stuck. It’s written for first-time managers, and most of it holds if you’ve been promoted from inside the team, hired in from outside, or stepped up to manage other managers. The places where those cases differ are marked.
Most people searching for a 90-day plan want a list of deliverables. The first three months of a management role are better described by what you’re collecting than by what you’re producing: how work really moves through the team, who people trust, what they’d keep, and what they’d change if somebody asked. You’ll make decisions in that window, and they land better when they rest on those answers.
It also helps to be realistic about where you’re starting. Gallup estimates that about 70% of the variance in team engagement comes down to the manager, and it finds that only 44% of managers worldwide have had any manager training.[1] Gallup sells engagement measurement, so hold the 70% as a strong signal and not a law of nature. The practical reading is simpler: most new managers learn the role by doing it, and their teams feel the difference. AI can make that self-directed learning faster and less lonely. In our view, a structured program, with practice and feedback on real management situations, shortens the curve further than reading alone, and the two work well together.
Here’s the plan this guide follows, filled in for the support team lead from the opening. Change the contents for your role. Keep the order.
| Days | The job of this phase | What you have by the end | First AI task |
|---|---|---|---|
| 1 to 30: Learn the ground | Understand how work moves through the team and what the words mean | You can explain the team’s workflow in the team’s own language, and you’ve met every person once | Brief you from the documents: a glossary, dated commitments, and the questions the files can’t answer |
| 31 to 60: Listen | One proper one-on-one with each person, then look for patterns | A short list of themes, each confirmed by at least two people | Draft a conversation guide for each person beforehand, and sort your notes into themes afterward |
| 61 to 90: Choose two changes | Make two small, reversible changes and say why | Both changes running, each with an owner and a review date | Stress-test each change: who it touches, what could go wrong, how you’d undo it |
An illustrative plan, not a prescription. The phases and the “confirmed by two people” line are the parts that travel between roles.
The plan is deliberately light on outputs. A manager who ships five changes in month two is often acting on the first plausible explanation, and teams tend to notice.
Let AI prepare you for the conversation, and don’t let it hold the conversation for you.
The first month is where AI earns its keep, because the job is absorbing a lot of written material quickly. The method is to gather a small set of documents, have AI pull out what you need to know, and then use its gaps as your first questions. Start with these:
Leave out anything that’s about a named person: performance reviews, pay, health, conduct. That isn’t squeamishness. It’s the line between material you can reasonably ask a work tool to help with and material that belongs in a conversation with your manager or HR partner.
Anything about a named person stays out of any tool your company hasn’t approved for that kind of information.
Our guide to using AI at work without leaking confidential company data covers what to check with IT. Once you know what’s allowed, here’s the briefing prompt. The documents go first and the instructions after them, and it asks for quotes before any summary. Anthropic’s own prompting guidance for Claude recommends both moves for long documents.[3] That’s vendor advice, not independent research, but it fits the job: a quote-first summary is harder to smooth over.
| [Paste each document here, labelled with its title. Individual names removed.] I’ve just become the lead of an eight-person customer support team. The documents above are our team goals, team-level results for the last two quarters, my predecessor’s handover note, and the meeting calendar. Before you summarize anything, quote the five passages you think matter most to someone new to this team. Then give me: 1. A glossary of team shorthand, acronyms and internal terms, with what you think each means and how sure you are. 2. What the team is measured on, in plain words, and where two measures might pull against each other. 3. Commitments with dates in the next 90 days. 4. Ten questions these documents can’t answer that I should ask people in my first two weeks. Mark anything you inferred as “inferred.” Don’t guess at anyone’s motives or performance. |
Swap in your own role and documents. The “inferred” label and the list of questions the files can’t answer are the two lines worth keeping.
Two parts of the output deserve most of your attention, and they’re the parts people skip. The first is the glossary. Say the team keeps talking about “the BAU queue,” “P2s” and “the Thursday cut.” You could nod through your first team meeting, or you could arrive already knowing which of those you understand and which you need someone to explain.
| Term | The AI’s reading | How sure | What you do with it |
|---|---|---|---|
| BAU queue | Routine, non-project tickets | High: defined in the handover note | Use it as is |
| P2 | Second-highest priority level | Medium: the priority scale isn’t in any document | Ask the person who triages |
| The Thursday cut | Some weekly deadline | Low: mentioned three times, never explained | Ask in the first one-on-one, and listen for whether it’s a rule or a habit |
The low-confidence rows are where your first conversations should start.
The second part is the list of ten questions the documents can’t answer. It’s the most useful thing on the page, because it turns a pile of reading into a shortlist of things only people can tell you. Pick the five you most want answered, and book your first conversations before you read anything else.
A first one-on-one has a job: to find out what this person thinks you should know. Four questions do most of the work, and you can ask them of anyone, in any role:
AI is useful for tailoring those to the person. Give it their role, how long they’ve been on the team, and what you’ve learned so far, and ask for three follow-ups per question and one line you can say when someone answers “it’s fine.” Don’t read its list aloud. Read it once, cut anything you wouldn’t say in your own words, and keep the rest in your back pocket.
Say your first one-on-one is with a specialist who’s been on the team three years, and you open with question one. She answers, and then says, without being asked, “Nobody’s told us whether the Thursday cut is real or a suggestion.” That sentence wasn’t in any document, and it wasn’t on your prepared list. You put the guide down and ask what happens on the Thursdays when tickets don’t clear. That’s the conversation doing what no briefing could.
Afterwards, the AI’s job changes. Once you’ve had the conversations, you’ll have eight sets of messy notes, and patterns are hard to see from inside them. Replace names with letters and use this.
| Below are my notes from eight one-on-one conversations with my new team. Names are replaced with the letters A to H. [Paste notes here] Sort what I heard into themes. For each theme, show which letters raised it, the closest words from my notes, and whether people agree or see it differently. Don’t merge themes that sound alike but mean different things. List anything only one person said in a separate section at the end. Don’t suggest solutions yet. |
The “don’t suggest solutions yet” line stops the tool jumping to fixes before you’ve checked the themes are real.
What comes back is a draft register. You fill in the last two columns yourself, and the register is what you take into your manager’s office in week eight.
| Theme | Raised by | Closest words | Confirmed by | First move |
|---|---|---|---|---|
| Thursday cut is unclear | B, C, E, F, G | “Nobody’s said if it’s real.” | Five people, two different readings | Tell the team what it is, in writing, in week ten |
| Refund escalations stall | A, D, H | “I wait a day for a yes.” | Three people, separately | Ask my manager who can approve, then name a cover |
| A vacancy is being covered informally | C, F | “We’ve split her queue.” | Two people, both independently | Raise with my manager and HR partner. Don’t act alone. |
| Only one person said it | G | “The tools are slow on Mondays.” | Nobody else | Check at the next round. No action yet. |
Invented for illustration. “Confirmed by” is the column that keeps you honest: a theme with one voice goes at the bottom.
Who you are changes how you run this round. A few cases:
Handing the preparation to AI doesn’t hand over the responsibility. The decisions, and everything about people, stay with you. A plain split keeps that honest, and it tells your own manager what you’re doing with the tool.
| What AI does | What you still own | How it gets checked |
|---|---|---|
| Drafts a glossary, dated commitments and a list of open questions from your documents | Which questions are worth asking, and in what order | You find three claims that would change what you do in the source documents before relying on any of it |
| Tailors a conversation guide for each person | How you open, what you say, and when you put the guide down | You read it once and cut any question you wouldn’t ask in your own words |
| Sorts your notes into themes and shows who raised each | What counts as a theme, and what’s one person’s view | A theme isn’t a finding until two people, separately, have confirmed it |
| Stress-tests a change: who it touches, what could go wrong, how to undo it | Whether to make it, and every decision about roles, hours or pay | Your manager or HR partner sees any change that touches people before the team does |
The same split works for any manager. Only the examples change.
The right-hand column exists because of how these tools fail. In a field experiment with 758 consultants at Boston Consulting Group, people using AI on tasks within its reach finished them faster and produced work rated more than 40% higher in quality. On a task deliberately placed outside what the model handles well, they were 19 percentage points less likely to reach a correct answer than people working without it.[2] The practical point is that a wrong answer doesn’t announce itself. Nothing on the screen tells you which side of that line a given output sits on.
That’s a consulting experiment, not a study of new managers, so don’t read it as a forecast for your team. It does explain why a new manager is more exposed than most. You don’t yet know the ground truth, so you’re badly placed to spot a confident summary that’s slightly off. That’s where the second rule comes in.
Nothing AI tells you about your team counts as a finding until two people have confirmed it.
Used this way, the tool makes you better at asking and leaves the answering to people. It also fits our advice in when not to use AI at work: if a task depends on trust or on reading a person, keep the human in the lead role and use AI around the edges.
By week nine you’ll have a register with a few themes that two or more people have confirmed. Pick two, and make each change small enough to undo. Two is a deliberate limit. Each change needs you to explain it, answer questions about it and check on it, and a third tends to get less of your attention than it needs.
Before you announce anything, write each change on a card and ask AI to try to break it. A prompt like “Here’s the change I’m planning. Who could it affect that I haven’t listed, what could go wrong in the first two weeks, and what would I see if it was failing?” does this well. You decide what to do with what it says. Here’s the card, filled in for the Thursday cut.
| Field | Filled in |
|---|---|
| The change | Publish a one-page rule for the Thursday cut: what it is, who can waive it, and what happens to tickets that miss it. |
| Why | Five of eight people raised it, with two different readings of what it means. |
| Who it affects | The whole team, and the two teams that hand us tickets. |
| Owner | Me, for the first month. Then the senior specialist who asked the question. |
| How we undo it | Go back to the current informal practice after two weeks if the rule creates more escalations than it removes. |
| Review date | Four weeks after it’s announced. Ask the team what they’d change. |
Invented for illustration. The “how we undo it” and “review date” rows are the ones people tend to leave blank.
In month four, it’s tempting to count how often you used AI. That tells you about your habits and nothing about the team. A better lens looks at the changes themselves:
If you’re an HR or L&D partner wanting this to be how every new manager in the company starts, it’s a different job from the steps above, and the terms are worth keeping apart. Training builds the skill of briefing, interviewing and sorting with AI. Workflow design gives it a place to live, for example a listening-round template that every new manager receives in their first week. Reinforcement is the manager above them asking to see the register at day sixty. Adoption is what you can see when that happens without anyone chasing it. An individual manager can start on their own with this guide, and structured training shortens the path and builds a shared standard across a cohort. Our corporate workshops are built for that second case.
For a manager working alone, the next step is small. Gather the five documents in the list above, run the briefing prompt on them, and choose the five questions from its list that you most want answered. Then put three one-on-ones in your calendar before you read the output a second time.
Usually not, unless your organization has approved the tool for that kind of information and you’ve checked. Notes on performance, pay, health or conduct are best read by you alone. There’s a second reason to hold back, which is that one person’s opinion of a colleague can shape your view before you’ve met them. Read those notes after your first round of one-on-ones, and treat them as a view to check.
Yes, briefly and specifically, unless your organization’s policy says otherwise. Something like: “I use it to draft my questions and sort my own notes. Nothing with your names goes in, and I’m the only one who sees what you tell me.” In my view, saying it plainly settles the question faster than leaving people to wonder.
Yes, and probably a more careful one. You know the work, but you don’t yet know the job, and your former peers will tell you different things now that you can affect their workload and their reviews. Skim the glossary step, since you know the vocabulary, and put extra time into the conversation guide and the opening line about what changes between you.
Start with whichever one your organization has approved for work documents. If you have two approved options, prefer the one that can already see where your team’s files live, such as an assistant built into your email and document storage, because it saves pasting and usually works within the access you already have. A general chat tool with uploaded files is a fine second choice. The method in this guide doesn’t depend on the product.
Pick three claims from the summary that would change what you do if they were wrong, then find each one in the source. If any fails, rerun the prompt with fewer documents, or ask the tool to quote the passage it relied on. Then test the important ones on a person, because a document can be out of date and sound current. A check like that takes about ten minutes, and it’s worth doing before you act on the output.
The figures on manager influence and manager training come from Gallup’s page “How Managers Impact Team Productivity,” which doesn’t show a publication date. Gallup measures employee engagement commercially, so its estimates are best read as a strong signal and not a settled fact. The consultant results come from the working paper by Dell’Acqua and colleagues, a field experiment with 758 consultants at Boston Consulting Group in 2023, which studied consulting tasks and not management transitions. The prompting advice on placing documents first and asking for quotes comes from Anthropic’s documentation, so it’s vendor guidance and not independent research. The scenes, example tables and quoted lines are illustrative, not real cases. Widely repeated figures about how many new managers fail were left out because we couldn’t trace them to a primary source.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.