Bringing AI into your department when you are not the boss

Bringing AI into your department when you are not the boss

Asking permission to introduce something new before a result has been shown is a losing move: you will be told "not now". Below is an order in which permission is not required: start with your own tasks, gather figures and arrive not with a request but with something finished.

Asking permission to introduce something new before a result has been shown is a losing move: you will be told "not now". Below is an order in which permission is not required: start with your own tasks, gather figures and arrive not with a request but with something finished.

"I hold a junior position and I work in a public body"

A phrasing from my survey. Behind it is the typical situation: a person is already using AI and getting results. A document analysis that took a day is done in ten minutes. A report is assembled in an evening.

And then a dead end. He cannot introduce it in the department because he decides nothing. Proposing it to management is frightening: either they will not understand, or they will forbid it, or they will decide he is amusing himself instead of working. Continuing quietly is also bad: there is a result and no way to explain where it came from.

A separate annoyance: the larger and more formal the organisation, the slower any change goes through — and the stronger the sense that trying is pointless.

Trying is not pointless. You simply have to approach from a different side than people usually do.

Why "let us introduce AI" fails as a proposal

The root cause: you are proposing a tool and the decision is made about risk.

In the head of the manager you bring this to there is no "interesting or not". In his head is what happens if it breaks, who will answer for it, whether anything will leak and whether he will have to explain himself upwards. Your ten minutes instead of a day do not enter that frame at all.

The misconception: "I need to describe more convincingly how good it is". The more vividly you describe it, the stronger the sense of risk. A junior employee's enthusiasm reads to a manager as "he does not understand the consequences".

A second misconception: "I need permission from above to start". You do not. You need one finished result, shown at the right moment. Permission is granted for what already works rather than for an idea.

And a systemic layer. In a large structure no single person can say yes to a big change — but almost anyone can say yes to a small one. So the size of the first step matters more than any argument.

The order that works

Step 1. Do it on yourself, without announcements

No announcements. Choose a task of your own that takes a lot of your time and is visible to nobody: reconciliation, preparing a summary, drafting a document, sorting incoming enquiries.

Do it the new way for three weeks. Without announcements, but within your company's rules on data and tools. There is one task at this stage: gather facts. How long it used to take, how long it takes now, how many errors were found.

Obligatory: everything that goes out is verified by you against sources. An error at this stage will close the subject for years.

Step 2. Gather a figure, not an impression

After three weeks you should have a one-line proposal containing no mention of AI:

"Preparing the weekly summary took four hours, now it takes forty minutes. Over three weeks that is twelve hours freed up."

That is the language decisions are made in. "I tried out some AI" is not that language.

Step 3. Arrive with a result rather than a request

The difference is fundamental.

Not this: "Can I try using AI for the reports?" — that is a request to permit a risk, and the default answer is no.

This: "Here is the weekly summary, I assembled it differently and it took forty minutes instead of four hours. I checked it against sources, here is the reconciliation. If it suits, I can do it this way regularly."

In the second case the manager is assessing finished work already in front of him. The risk is invisible, because the result already exists.

Step 4. The first pilot: small and reversible

If permission is given, do not expand immediately. Take one task, one person, one month and name the rollback condition without fail: "if anything goes wrong we return to the old order and lose nothing".

Reversibility is the main argument for a cautious manager. A change that can be cancelled does not require courage to approve.

Step 5. Hand the result upwards

An unpleasant but necessary point. State the effect so that your manager can own it — it is his department that started working faster.

Implementations fail not on the technology but on the author of the idea looking like an upstart. Somebody who brought a result and did not grab the credit gets a second yes far more easily than the first.

How to answer three standard objections

They will almost certainly be raised, so prepare the answers in advance.

"And what if it makes a mistake?" Do not argue and do not say "it does not make mistakes". Answer with process: "Everything that goes out I check against sources — here is the reconciliation for this summary. An error is caught at the same stage as before, it simply reaches that stage faster."

"That is not permitted here." A clarifying question rather than an argument: "Where can I look at the policy — I want to check I am working within it." Half of prohibitions turn out not to be prohibitions but an absence of practice. The other half are a real constraint, and it is better to know about it in advance.

"Have you got nothing better to do?" The most common and the most galling. Answer with a figure and nothing else: "The summary now takes forty minutes instead of four hours, and the freed time went on sorting enquiries." Freed hours are hard to argue with.

What must not be done

Do not send out what must not be sent out. Personal data, internal information, restricted documents. One such case cancels everything, and rightly. If in doubt, treat it as forbidden.

Do not submit anything unverified. Your reputation here is worth more than the hours saved.

Do not promise staff reductions. Even if it is true. The first mention turns every colleague into an opponent and you into the person who came to sack them.

Do not start with the most important process. Start with something dull and your own.

Do not argue with "that is not how things are done here". That is not an argument to be refuted — it is a state of affairs changed by one result shown.

What to do if the answer is no

That happens too, and not rarely. It does not mean the end.

Clarify what exactly the no covers. A ban on the tool, a ban on this process, or a ban right now are three different things. Usually it is the third.

Carry on with your own tasks. A ban on changing a department process is not a ban on how you personally prepare your own draft. That boundary is worth clarifying rather than assuming.

Keep gathering figures. In a quarter you will have four months of observation rather than three weeks, and the conversation will go differently. More often still, what changes is not your argument but the setting: a neighbouring department acquires a similar practice and the question resolves itself.

And the most practical point: this skill stays with you. Even if introducing it in the department does not work out, you have acquired something worth money on the market — and that is no longer about your job title.

Which tasks to take first

What works is where you supply the material yourself — there is nothing to invent and verification takes minutes:

Do not take first: any task involving provisions, references and figures that were not in the source material. They can be done, but only with full verification against sources, and that is not the case to start a pilot with.

Where to start

Ready specialists for most departments are already assembled: over 200 AI assistants across 14 categories — legal, HR, marketing, sales, analytics, e-commerce, engineers, media. These are not prompts needing refinement: you open a chat on your subject, describe the task, get a document.

Plus automation templates for recurring processes — once the pilot has passed and the task has become regular.

Registration is free and opens three days of full Basic access — all the assistants, the prompt texts, the skill files and six foundational express courses.

Today, one task. Take something you do by hand every week and time it before and after. In three weeks you will have a figure rather than an opinion. With a figure, the conversation with a manager looks entirely different.

You have to start not with permission but with a result nothing can be said against.

If you have a colleague doing the same thing quietly and afraid to say so, forward this to them.

Which task to start with so as not to argue

There is a class of task requiring nobody's permission and giving a measurable result immediately.

The “Automator” course is eleven lessons on working with files, and the first is on report analysis: any analytics, cost documentation, accounting exports, advertising reports.

Why that is the right first step. You work with your own working files, involve nobody, introduce nothing. In three weeks you have a figure: it took this long, it now takes this long.

And the most reliable property of that class of task. You supply the material — which means the model retrieves nothing from memory and has nothing to invent. An error is almost excluded, and that matters when you are showing a result to a manager for the first time.

Then upwards by degrees: site performance analysis, summarising books and documents, transcribing meetings. All of it your own tasks rather than the team's.

The course is included in the Plus plan.

What to take to your manager

Ready roles instead of "buy us a subscription". The assistant catalogue, Basic plan — over 200 specialists across fourteen directions. That is a strong argument: not "buy a tool and we will work it out" but "here is a role for our task, and it works today".

There is almost certainly a category for your department: 26 marketing, 13 legal, 10 sales, 8 HR, 13 e-commerce, 9 engineering, 12 strategic analysis methods.

Figures for the justification. The “Automator” course, Plus plan, walks through working tasks with files: reports, contracts, product cards, transcription. Count how many hours they take now and what a contractor would cost. Your argument: the same thing can be done internally.

What a pilot looks like. Take one process, assemble an assistant for it and show the result before the meeting. A phrasing that works inside a company too: while others promise an analysis in a week, you show a finished table in twenty minutes.

What to show on money. In Gen AI the estimate is visible before generation, money is returned to the balance for anything not delivered, top-ups start at $5. For a manager that answers the main question: how much will this cost and can we stop.

Skills for the conversation with a manager

The article is about going with a figure rather than an idea. Below is what that figure is calculated with. All six skills are advanced: from the Creator plan, or $9 one-off, or 1000 XP.

What to show first. The first item. Not an implementation plan but a readiness assessment: it shows you understand the company's constraints rather than having arrived full of enthusiasm from the internet.

What to take on the platform for this task

What to read next

AI for business owners — how this looks from the other side of the desk.

A company knowledge base as a service — the next step once it works.

Local models for sensitive data — if the company has data requirements.

How to verify a model's output — without which an implementation falls apart on the first error.

AI API pricing in 2026 — what a request costs and how to work out your own costs.

Also on AI SKILLS