Where to start with AI in a company of 50 people

Here is the order we would follow for a first AI implementation in a growing company.

1. Start with one job, not one tool

Pick a single piece of recurring work that costs your team real time or causes real mistakes. Good first candidates repeat every week and follow a pattern: putting together the month-end pack, chasing customers who pay late, answering the same customer questions again and again, turning a request into a quote, checking orders before they go out.

Write the job down as a question or an outcome: "Who should we chase today, and for how much?" is a far better starting point than "We should be using AI."

2. Check what the AI would need to see

Every AI tool works from what you give it. Before choosing one, answer four questions about the job you picked:

Question	Why it matters
Where does the information for this job live?	If it is spread across three systems and a spreadsheet, the AI will see whichever part it is given.
Is it entered on time and in full?	Late or half-filled entries give confident, wrong answers.
Do people agree what the key words mean?	"Overdue", "active customer" and "revenue" often mean different things to different teams.
Who looks after it?	Someone has to own the data, or it drifts again.

If the answers are good, carry on. If they are not, that is the first piece of work, and it pays off for every AI tool you use afterwards. For a closer look at what an AI agent can and cannot see, read what Copilot Studio can and cannot see in your business.

3. Agree three simple rules before you switch anything on

What may go into which tool, especially anything personal, confidential or belonging to a client. Check each tool's data-processing terms and where it stores data; for personal data, GDPR applies (UK GDPR in the UK).
Who checks the output before it reaches a customer or a decision.
Where the prompts and results are kept, so good ones are shared and bad ones are not repeated.

One page is enough. It stops ten people signing up to ten different tools with company data.

4. Use what you already pay for first

The tools you already license, such as your office suite, accounts package and CRM, may include AI features. Switch those on for the chosen job before buying anything new. You learn what works on your own information, often for little or no extra cost, and you find out quickly where the gaps are. Some AI features are paid add-ons, so check your licence first.

5. Run a small pilot with the people who do the job

Choose two or three people who do the work today and know where it goes wrong. Give them a few weeks, a clear measure and permission to say it does not help. New systems are rarely the hard part; changing how the work gets done is. A pilot built with the team is the one that sticks.

6. Measure something you care about, and weigh the cost

Decide before the pilot what would count as success, and measure it before and after: time spent on the job each week, errors that reach a customer, how long it takes to answer the question you started with.

Then weigh it plainly: the licences, the hours of the people in the pilot and any clean-up of the data on one side; the hours saved and the errors avoided on the other. Keep it and widen it, change it, or stop. All three are good outcomes, because each is decided on evidence.

An example: the month-end pack

An illustrative job, not a client.

What it needs to see	The ledger, the sales system's monthly figures, and the stock spreadsheet the warehouse keeps.
What the AI does	Pulls the figures into the usual layout, flags the lines that moved most against last month, and drafts a first commentary on why.
What a person still checks	Every figure against the ledger, and every sentence in the commentary that sounds certain but is really a guess.
What would count as success	The pack ready a day earlier, with the same number of corrections from the finance director, or fewer.

7. Know when you have outgrown the first step

The first AI wins usually come from a single job inside a single system. The next questions are harder: which customers really make us money, where is the cash tied up, what needs us today? They cut across sales, finance, operations and stock. That is the point where AI needs connected, checked information underneath it, so that the answers agree with each other and with your reports.

What we would avoid

Buying a large platform before you know which job it is for.
Letting everyone use a different tool with company data.
Expecting AI to fix information that is late, missing or disputed. It will read it as it is.

Where we come in

Most of the steps above you can take yourselves. Where we help is the step after: when the questions cut across your systems, we connect your whole business and put AI to work making sure every entry goes in properly, so the right numbers are in front of you every morning and your team spends its time on the work, not the reports. We build it with your team, a step at a time, and automate the routine work around it.

Common questions

Which AI tool should we choose?

Start with the job, then the data, then the tool. Often the right first tool is one you already pay for. The choice matters less than whether the tool can see the right information.

Do we need a data team first?

No. A company of 50 people can start with one job and the people who do it. You need someone who owns the data for that job, not a department.

When is it worth bringing in an AI consultant?

When the job you want to improve depends on information spread across several systems, or when a pilot has worked and you want to widen it without losing control of the data. Before that, the steps above are usually enough.

How soon should we see something?

A small pilot on one job should show within a few weeks whether it helps. If it is still unclear after that, narrow the job or check the data before you widen it.