Implementing AI starts with the question of where the company is losing money and time; choosing a model comes last. The sequence is this: an audit of processes and data, one or two areas with a clear effect, a pilot with a measurable result, rules and authority, then expansion. Decisions that carry risk stay with people.
Where to start
- Formulate the business task. “Implement AI” is not a task. A working formulation sounds like this: shorten the time it takes to prepare a commercial proposal, stop losing enquiries, see deviations from plan on the same day.
- Check the data. AI works with what the company has recorded: orders, contracts, correspondence, accounts. If this sits in five systems and ten spreadsheets, the first step is to connect them.
- Choose an area for the pilot. A good one has a lot of repetitive work, a result that can be measured and errors that would not be a disaster.
- Appoint an owner on the business side. The pilot needs a manager who answers for the result of the area. One person from IT is not enough.
- Set the rules. What the system does on its own, what it prepares and hands to a person, which data it can access, and who sees the log of its actions.
- Measure, then decide whether to expand. Compare the area before and after on the metrics you chose in the first step.
Where AI usually pays for itself first
- Documents: contracts, applications, reports, tender documentation.
- Client enquiries: intake, analysis, drafting a reply, passing it on.
- Sales: preparing proposals and supporting deals.
- Control of execution: reconciling plan with actuals and finding deviations.
- Company knowledge: search across regulations, contracts and past decisions.
These areas have something in common: a lot of text and repeated actions, and an outcome that can be checked.
What not to do
- Buy a platform before the tasks are clear.
- Start with the most complex and risky process.
- Hand the subject entirely to the IT department or entirely to a contractor. The owner sets the goals.
- Automate disorder. If a process is not described and nobody answers for it, AI will only speed up the confusion.
What stays with people
Goals, authority and decisions with consequences: money, contracts, personnel matters, everything that concerns security. AI prepares options, calculates, checks completeness and keeps track of deadlines. A person approves. The line between “the system does it” and “the system proposes” is drawn before launch and revised as experience builds up.
How long it takes
A pilot in one area takes weeks. Rebuilding a company around data and AI takes months and is done in stages, so that the business keeps running. The timing depends far more on the state of the data and on the managers’ willingness to take part than on the technology itself.
When you need an external team
When there is nobody inside who understands the company’s economics, its processes and the technology at the same time. In that case it makes sense to start with an audit: it shows where the effect will be visible and where putting things in order is enough without AI. The next level for a group of companies is described on the page “AI management company”.