We need AI. But what for?

Begin with work that is difficult

“We should be using AI” is understandable. It is also too broad to guide an investment.

A useful starting point is a specific piece of work: locating an instruction, summarising a meeting, preparing an update or comparing information. Describe the difficulty and the result you need before deciding how to produce it.

That gives the business a way to compare AI with simpler process changes, ordinary automation or better use of existing tools.

The problem: the tool arrives before the use case

A demonstration may look impressive while depending on information the business does not have in a usable form.

An assistant answering process questions needs reliable instructions. A reporting tool needs records that reflect actual work. A summary needs someone to check important decisions and missing context.

If those foundations are weak, the first improvement may concern information ownership or process discipline.

What the engagement evidence supports

In a construction project-management engagement, the team discussed AI-assisted documentation, finding information and reporting within a developing workspace.

The conversations also distinguished rule-based automation from more interpretive AI behaviour and raised the need to check errors. Meanwhile, the core project structure, views and rollout were still being refined.

The records support an opportunity being explored, with the core workspace and operating processes still being refined.

That boundary is useful. An idea can be worthwhile without yet being an achieved outcome.

The approach: define a small, testable opportunity

Write down five things:

• The task and who performs it.
• The information required and who maintains it.
• The output that would be useful.
• The consequences of an incorrect answer.
• The person responsible for checking and using the result.

Consider whether a simpler intervention would achieve the same purpose. A clear template may improve reporting. A searchable guide may help someone find an instruction. A rule may handle a predictable reminder.

If AI still looks appropriate, test it against representative examples and difficult cases. Keep the trial proportionate to the consequence of an error and the information involved.

Define the outcome before the trial

“People liked it” is useful feedback, but it does not establish that the work improved.

For a documentation use case, examine whether the result is accurate, complete enough and easier to maintain. For a summary, check whether actions and decisions are represented correctly. For reporting, check whether the source records support the conclusion.

Include the time spent checking and correcting outputs when assessing value. These are proposed evaluation methods, not measurements from the reviewed engagement.

Give AI a defined job

A practical AI decision connects a business need with usable information, accountable people and a way to assess the result.

Start your free Initial Analysis to consider your operating foundations, or explore In Place Assess.

If you have a use case in mind, discuss what you want it to achieve.

Related case: Making project management usable before adding more automation.