Incorporating AI Into the Workplace: A First-Month Plan for Small Teams

Start with one job, one tool and one owner. Choose a task your team repeats at least five times a week, that a person can check in under a minute, and that involves no sensitive records. Give it to a named owner, time how long it takes today, and agree a one-page use policy before anyone opens the tool.

The usual failure runs the other way round: six people try six tools on twenty tasks, and a month later nobody can say what helped. This plan is narrow on purpose. It takes about two hours of set-up in week one and roughly an hour a week after that, and it ends on day 28 with a decision to keep, widen or stop. Expect a modest first result: one task that reliably takes less time is a good month, and it gives the team a checking habit before AI touches anything riskier.

Follow me on Instagram@sagnikteaches

Three decisions on day one: the task, the tool and the owner

Write the three decisions on one card and pin it where everyone can see it. A card forces choices that a vague "let's try AI" conversation never does, and in week four it gives you something concrete to review.

Connect on LinkedInSagnik Bhattacharya

• The task is a named job with a start and an end, such as "write the first reply to a new enquiry". "Marketing" or "customer service" is a department, not a task. If you can't describe the input and the finished output in one sentence each, the task is still too big.
• The tool is whichever assistant your team can already sign in to with a work account. The best-reviewed assistant is the wrong answer if it means a new login, a new bill and a new set of data rules in week one.
• The owner is one named person with a few hours of calendar time, the authority to pause the trial, and a backup. The owner isn't necessarily the most senior person and doesn't need to be technical.

Here is a filled-in card for a florist with two full-time staff and a Saturday assistant. The task is the dullest one in the shop, which is why it works: orders arrive by phone note, message and email, and someone retypes each one into a single order sheet.

Subscribe on YouTube@sagnikteaches

TASK       Turn each incoming order (phone note, message or email) into one order-sheet row:
           customer first name, delivery date, delivery slot, card message, budget, allergies
TOOL       Microsoft Copilot Chat, included with the shop's Microsoft 365 business subscription
OWNER      Shop manager (about 1 hour a week). Backup: the Saturday assistant
BASELINE   5 minutes per order, about 40 orders a week = 3 h 20 min (timed on 10 orders)
STOP RULE  After week 2, if more than 3 rows in 10 need correcting, pause and fix the instructions
REVIEW     Fridays, 4.30 pm, 20 minutes, owner plus whoever entered orders that week
PARKING    Ideas for later (not this month): stock-take notes, supplier emails

Every line on that card can be checked later. The baseline came from a stopwatch, the stop rule says in advance what failure looks like, and the review slot is already in the diary. What the card leaves out matters as much: there is no second task, no list of future uses and no vendor comparison. Anything else goes on the parking line and waits for week four.

Scoring candidate tasks: a driving school's shortlist

Say a driving school has four instructors and two office staff. At a Monday huddle each person writes down the repeated job they like least, and the office manager, who will become the owner, collects five candidates. Each job gets a score from 1 to 3 on four tests, with thresholds so the scoring doesn't depend on anyone's mood.

• Frequency: 3 for five or more times a week, 2 for two to four times, 1 for once a week or less.
• Time drain: 3 for two or more hours a week in total, 2 for one to two hours, 1 for under an hour.
• Easy to check: 3 if a person can judge the output in 30 seconds, 2 if it takes a few minutes, 1 if the checker has to go back to the original source.
• Low risk: 3 for no personal data, 2 for names and contact details only, 1 for anything about health, money or children.

Replies to enquiries win with 11 out of 12. About 35 arrive a week, and each takes roughly six minutes once someone has checked the diary, so the baseline is 210 minutes, or 3.5 hours. The output is a short email the office manager can judge at a glance, and the only personal detail involved is a name and an address the reply is going to anyway.

Progress notes look tempting because they eat two hours a week, but they score 1 on checking and 1 on risk. The notes are about named learners, some of them under 18, and the only way to verify a note is to compare it with what the instructor actually said. That makes them a sensible third task once the team has a habit of reviewing, and a poor first one. If you want a longer test than these four, choosing your first AI project in seven tests goes further.

Matching the tool to what your team already uses

In most small workplaces the tool question has a dull answer: use the assistant that sits next to the work. Prices below are USD list prices as of October 2026, so confirm them on the vendor's page before you buy anything.

The driving school runs on Google Workspace, so its enquiries land in a shared Gmail inbox and the owner picks Gemini. The florist is on a Microsoft 365 plan and picks Copilot Chat. Neither spends anything extra in month one. A team with no suite at all can start with two seats of a Team plan, and both ChatGPT and Claude let a business team share a project that holds the instructions and examples, so everyone starts from the owner's prompt instead of their own version of it.

Don't buy seats for the whole team in month one. Buy for the owner and the one or two people who do the task. If you'd like a side-by-side view of the assistants before choosing, ChatGPT, Claude, Gemini and Copilot compared for small businesses covers the differences.

What the owner does in an hour or so a week

The owner is the difference between a trial and a pile of logins. The job is smaller than people fear and none of it is technical. A good owner is organised, trusted by the people doing the task, and senior enough to say "pause". In a six-person team that is usually the office or shop manager rather than the founder. Choosing and supporting an AI champion goes deeper on who suits the role; in month one the duties are these.

1. Keep the facts sheet current (10 minutes). One document holds every price, opening hour, policy and "we don't offer that". It carries a date at the top, and any change goes in the same day. The assistant is only as accurate as the facts it is given.
2. Read the drafts (30 minutes). In weeks two and three the owner reads every draft. After that, a sample of five a week is enough.
3. Keep the log (10 minutes). Each run is marked "sent as it was", "small edit" or "rewrite", with a few words on why.
4. Answer "can I use it for this?" (5 minutes). The answer is yes, no, or ask first, within one working day. Slow answers push people back to personal accounts.
5. Run the Friday review (20 minutes). Four numbers, two questions, one decision about what changes next week.

That adds up to about 75 minutes in a normal week, and about two hours in week one when the owner also writes the facts sheet and the rules. Name a backup who can run the Friday review if the owner is off. Two missed reviews in a row is the signal to pause the trial, not to carry on without it.

Week one: time the task, write the rules, set up the accounts

Week one is groundwork, and nobody uses the tool on real work until week two. The order matters, because each step makes the next one easier to judge.

1. Monday, 30 minutes: time ten real examples. Do the task exactly as you do today, with a stopwatch running, and write down the minutes for each. The driving school's office timed ten enquiries and got a spread from 3 to 11 minutes, with a mean of about 6. Without this baseline, "it feels faster" is the only evidence you will have in week four, and feelings follow whichever tool was most recently exciting.
2. Tuesday, 45 minutes: write the facts sheet. For the school it holds lesson lengths, weekday and Saturday availability, that the first lesson is an assessment, that both automatic and manual cars are offered, and how to book. It also lists what the school doesn't do, and what the reply must say when a question falls outside the sheet: a person will confirm today.
3. Wednesday, 30 minutes: set up the accounts. Work accounts only. Check whether the tool has a model-training setting and look at what it says rather than assuming. Ask the admin to switch on whatever the plan includes. Then put the instructions and the facts sheet into a shared project, or a pinned document if the tool has no projects, so everyone starts from the same prompt.
4. Thursday, 45 minutes: the rules and a 15-minute huddle. The owner drafts the one page below, the team reads it together, and each person says aloud one thing they would never paste into the tool. If anyone has been using a personal account for work, say so without blame at this point; bringing that use into the open is the quickest way to move it onto the work account. Shadow AI in a small business explains how to ask.

Here are the rules as the driving school wrote them. Nine lines, deliberately.

AI USE RULES, MONTH ONE
1. Work accounts only, and only the tool on the card. No personal accounts for work.
2. OK to paste: first names, the customer's question, and our facts sheet.
   Never paste: full contact details, payment or ID numbers, health or welfare
   information, staff or HR matters, or anything covered by a confidentiality clause.
3. A person reads every AI draft before it leaves. Whoever sends it owns it.
4. AI drafts, people decide. No refund, complaint, hiring or discipline decision
   is made by the tool.
5. Any fact in a draft (a price, a date, a rule) must match the facts sheet.
   If it isn't on the sheet, the reply says a person will confirm.
6. If a customer asks whether AI was used, answer honestly.
7. Log each run (30 seconds). Report mistakes straight away; nobody is blamed
   for a mistake they report.
8. New tool or new task? Ask the owner first. Answer within one working day.
9. These rules are reviewed on day 28.

A policy that fits on one page gets read, and a policy that gets read gets followed. Rule 2 does the heaviest lifting, because it tells people what is safe to paste instead of listing everything that isn't. Rule 6 is there because someone will be asked, and an honest answer given calmly costs nothing. The fuller version, with sections for suppliers and record-keeping, is in how to write an AI usage policy for your small business.

Week two: ten live runs, each one read before it goes out

On Monday the first real enquiries go through the tool. The owner goes first, on their own, so the instructions get tested by the person who wrote them before anyone else relies on them. The prompt below carries the facts sheet in the same message. Once it works, save it as the project instructions so nobody retypes it.

You draft first replies to enquiries for a small driving school.
Use only the facts in the FACTS block. If the learner asks for something that
isn't in FACTS (a price, a date, a test slot), say a person will confirm it today.
Never guess. Friendly, plain, under 120 words, no exclamation marks.
Sign off as [your name].

FACTS
- Lessons are 1 hour or 2 hours, weekdays and Saturdays.
- The first lesson is an assessment: the instructor checks experience and plans a course.
- Automatic and manual cars are both available; the learner chooses when booking.
- Booking: reply to this email or use the booking page [link].

ENQUIRY
"Hi, I've never driven before and I'm 17. How much are lessons and can I start
in January? Do you do automatics?"

An illustrative reply from the assistant:

Hi [first name], thanks for getting in touch. Starting from scratch is fine: the first lesson is an assessment, so your instructor will see where you are and plan the course around you.

We have both automatic and manual cars, and you choose when you book. I don't want to guess at prices or January dates, so one of us will confirm those today. In the meantime you can book a first lesson here: [link].

Kind regards, [your name]

It is a good draft, and it still needs three fixes before it goes out. It never asks which days and times suit the learner, which is the one thing the office needs to give a real answer. The promise that one of us will confirm today is only honest if someone is on the hook for it, so the owner adds the enquiry to the day's call list. And "Starting from scratch is fine" is a nice touch that came from the model rather than the facts sheet, so a person should decide whether the school wants to say it.

Tally every draft one of three ways: sent as it was, small edit, rewrite. After ten runs the school's tally was five, three and two (illustrative). The stop rule on the card is more than three rewrites in ten, so two passes, but the edits are worth reading. Three of the five edits were the same one: the draft never asked when the learner wanted to start. That is an instructions problem, not a tool problem, so one line goes into the prompt.

End every reply by asking which days and times suit the learner.

Ten more runs after that change needed two small edits, neither about a missing question. The habit to build is exactly this: change the instructions, run again, count again. Swapping tools because a draft was disappointing teaches the team nothing, because the same weakness follows you to the new tool.

Week three: a 20-minute review built on four numbers

Hold the review in the same slot every Friday, with the slot in the diary before the trial starts. It is short because it only needs four numbers, taken from the log the owner has kept.

• Runs: how many times the task went through the tool.
• Minutes per run: timed on five runs, using the same stopwatch method as the baseline, so the comparison is fair.
• Rewrite rate: drafts that needed a full rewrite, divided by runs.
• Incidents: anything that went out wrong, or any rule that was broken, including near misses.

Add two questions for the people doing the work: what did you catch that the tool got wrong, and what have you stopped trusting it with? The second answer is the most useful thing in the meeting, because it points at exactly where the instructions need tightening. Here is the school's log (illustrative figures).

Week three is also when a second person should join, and that is a better test than anything in week two. Somebody who didn't write the instructions finds out quickly which lines assumed knowledge that lived in the owner's head. The school's second office worker asked what the draft should do when a learner requests a particular instructor. The instructions had no line for it, and the assistant had been cheerfully promising a named instructor. The owner added one sentence: don't promise a named instructor, say a person will check availability.

Week four: keep, widen or stop, with the sum

On day 28 the review has a different job, which is to decide. Three outcomes are legitimate, and stopping is one of them.

• Keep the task as it is if rewrites stayed under three in ten, nothing went out wrong, the owner's time stayed near the hour a week, and the people doing the task would be annoyed if it were taken away.
• Widen by changing one thing only: a second task on the same tool, or the same task for more people, never both at once. A second task gets its own row on the card and its own baseline.
• Stop if two weekly reviews running show no time saved, or the task turned out to need more checking than doing it by hand. Write down why. A clear "not this task" is worth more than a vague "it was fine".

The sum for the driving school, all figures illustrative: 35 replies a week at six minutes before and three minutes after saves 105 minutes a week, or about 7.6 hours over a 4.33-week month. At a loaded cost of $20 an hour for office time, that is roughly $150 a month. The tool costs nothing extra on Gemini or Copilot Chat. On a two-seat Team plan at $25 a seat on monthly billing it would cost $50, leaving about $100 a month.

That is a modest saving in cash, and you should be suspicious of any plan that promises much more from a single task. What the school gets beyond the arithmetic is a first reply inside the hour and an office that finishes the inbox before the afternoon's lessons begin. The next candidate on its table is rebooking cancelled lessons, which scored 9, and which brings in diary details that the rules will need a line for.

Where the plan bends: a bookshop and a photography studio

An independent bookshop with five staff has no office suite beyond email, so it buys two seats of a Team plan, about $50 a month at monthly list prices, for the events lead and the owner. Its first task is the Friday newsletter, drafted from the events sheet and the staff-picks list. The rules change in two places. Rule 2 names the customer-order list and the subscriber list as never-paste, because both hold names alongside phone numbers or email addresses. And a new rule says that whoever recommends a book has read it and signs off the blurb, because an assistant will happily invent a plot detail. In an illustrative first draft, a novel was described as set in the 1920s when it isn't. Only the bookseller who had read it noticed, which is the whole point of the rule.

A photography studio with three photographers and a manager starts with replies to wedding and portrait enquiries. Its facts sheet was pasted from last year's price list, and in week two a client quoted the old price back. That is how a stale fact usually shows up: late, and from outside. The fix had three parts. The sheet gets a date at the top and a single editor. Every price or date in a draft is a placeholder such as [price] that the sender fills in by hand during month one. And the Friday review gains a line asking whether anything on the sheet changed that week. For the wider picture of where a studio should begin, starting with admin, not editing is the sibling tutorial.

Six ways month one goes wrong, and how each one shows up

• Six people, six tools. It shows up when two staff paste the same enquiry into different assistants, get different answers, and nobody can say which instructions work. The fix is one tool per task for month one, plus an amnesty on personal accounts.
• No baseline. It shows up in week four as "it feels quicker", with nobody able to defend the subscription. The fix is ten timed examples in week one, before the tool is switched on.
• Drafts sent in a hurry. It shows up when an email goes out beginning "Here is a friendly reply you could send:", or with "[first name]" still in it. The fix is rule 3, plus the owner reading five sent items a week during month one.
• A stale facts sheet. It shows up as a reply quoting last season's hours or a package that no longer exists. The fix is a date on the sheet, one editor, and placeholders for anything that changes.
• A busy owner. It shows up as a skipped Friday review, then another. The fix is the pause rule: two skips and the trial stops until you change the owner or shrink the task.
• Task creep. It shows up as "while it's there, can it also draft the supplier emails?" The fix is the parking line on the card. Everything goes there until week four.

None of these needs a better tool. Each one needs a smaller promise, a clearer owner or a date on a document, which is why the card, the facts sheet and the Friday slot do more for a small team than any feature list.

Questions teams ask before they start

How many people should use AI in the first month?

Two or three: the owner plus whoever does the chosen task most often. That is enough to find the gaps in your instructions without turning the trial into a company-wide change. Add the rest of the team in month two, once the card, the facts sheet and the rules have survived a few weeks of real use.

Is it safe to start on free personal accounts?

For the month-one task, use work accounts instead. Consumer accounts have their own model-training setting that each person has to find and check, and the chats belong to the individual. Business plans from the major vendors don't train on your content by default, and a tool already included in your office suite adds no new login at all.

What if someone on the team doesn't want to use AI?

Don't force it in month one. Ask what worries them; it is usually job security, quality or being blamed for a mistake. Offer the reviewer role, checking drafts against the facts sheet, which is useful work and keeps them involved. Sceptics often find the gaps in your instructions faster than enthusiasts do.

Do we have to tell customers we use AI?

It depends on how you use it and where your customers are. A draft that a person reads and sends under their own name is ordinary business writing. A chatbot talking to customers directly is different, and some laws, such as the EU AI Act for customers in the EU, require saying so. If a customer asks, answer honestly, and check specifics with a qualified adviser.

Further reads

• How to Get Your Staff to Actually Use AI Tools — For the staff who nod at the huddle and never open the tool.
• How to Talk to Staff Who Fear AI Will Take Their Job — Deal with job-security worries before the first huddle.
• AI Change Management for Small Teams: A Practical Plan — The wider change plan once month one has worked.
• How to Survey Your Staff Before an AI Rollout (With Questions) — Questions that surface the repeated jobs people like least.
• Is My Business Too Small for AI? What Works for Teams of 1 to 10 — Reassurance and limits for teams of one to ten.
• How to Keep Customer Data Private When Your Team Uses AI — The detail behind rule 2: what never goes into a prompt.
• Starting a Business Using AI: A First-Month Plan for a Shop or Cafe — The same month for someone opening a shop or café from scratch.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft Learn's Copilot Chat overview (standard access, Outlook grounding), the Google Workspace pricing page (Gemini features by edition), Anthropic's Claude pricing page and OpenAI's ChatGPT pricing page, checked October 2026. Worked figures are illustrative.