How to get started with AI in your business

Getting started with AI transformation can feel expensive and uncertain. Leaders can see that the technology matters, yet demonstrations and headlines give them little help with the first practical decision.

That hesitation is reasonable. AI projects affect processes, responsibilities and budgets. A rushed purchase can waste money and tire the team before useful work begins. The way forward is to choose a small, important problem and study it properly.

Start with the work, not the AI tool

A common first move is to buy AI subscriptions for the whole company. A few curious employees find useful ways to draft, research or summarise. Others try it twice and stop. Six months later, the company has more licences but cannot point to a faster process or a better customer result.

The missing step is deciding which work should change. Before comparing products, name a problem in ordinary business language:

• invoice checks take two weeks and delay month-end
• sales enquiries wait in a shared inbox
• staff copy the same order details between two systems
• customer answers depend on one experienced employee being available

This gives the project a purpose. It also makes it easier to decide whether AI is needed at all. A fixed rule may need ordinary automation, while varied emails or documents may benefit from AI.

McKinsey's 2026 State of AI reports that 88% of organisations use AI regularly in at least one function. Use is widespread. The more useful measure for your company is whether one complete piece of work has improved.

Check the four parts around the problem

Once you have a candidate, look at the people, process, data and technology around it. A weakness in any one of these areas can stop the first phase.

People

Speak with the employees who do and review the work. Ask where time disappears, which cases cause rework and what they would be nervous about handing to a system.

Their involvement matters from the beginning. They know the exceptions, and they will be the first to notice a poor result. A system designed without them often creates an extra checking job rather than saving time.

Process

Ask someone to show the last five cases on screen. The written procedure may skip calls, spreadsheets and judgement that have become part of the real method.

Look for a clear beginning, a recognisable result and a manageable number of exceptions. If every employee follows a different route, document and improve the process before trying to automate all of it.

Data

List the records people use to complete the work. Check whether the information is complete, whether teams agree on the meaning of important fields and whether the proposed system may use it.

You do not need to clean every record in the company. You need reliable enough data for the chosen job. Our AI data readiness checklist shows how to test 50 recent cases in one workflow.

Technology

Only now should you examine tools and connections. Check what the company already owns, which systems can exchange data and where a manual handover is acceptable for the first version.

The simplest suitable technology is usually the best place to learn. A spreadsheet and an approval screen may be enough for an early phase. A broad custom platform rarely needs to be the first purchase.

Run a useful diagnostic this week

Bring together three or four people who understand different parts of the business. Keep the discussion focused on real work rather than general ideas about AI.

Start by listing three workflows that repeatedly cause delay, rework or missed opportunities. For each one, answer:

Then review five recent cases from each promising workflow. A process may sound simple in a meeting and look very different when staff open the actual files.

One workflow will often emerge as a sensible first phase. It may be less exciting than the idea that started the discussion. Invoice preparation, stock reporting or enquiry routing can be a better beginning than automated sales proposals because the method and result are easier to check.

The guide to choosing your first business process to automate provides a more detailed test if several candidates still look viable.

Write a one-page brief for the first phase

The brief should be understandable to the employee doing the work, the person paying for the project and anyone helping to build it. Include:

• the workflow and its owner
• the current delay or cost
• the records involved
• the first group of users
• the output they will review
• the actions the system may take
• the cases the first version will refuse
• one main measure of success

For overdue invoices, the first phase might prepare a review-ready list with account history and draft reminders. It may leave disputed invoices and changed bank details with the finance team. Success could mean reducing preparation time from two days to half a day without increasing errors.

This scope is small enough to test and useful enough to matter. The team can see what changed, build confidence and decide whether adjacent work deserves a second phase.

Aim for a result the team can absorb

An early phase should leave a working improvement behind. It should also teach you where the process, data and controls need more attention.

Start with one job, measure the old way and involve the people who know the exceptions. That creates a firmer basis for investment than another round of general AI demonstrations.