The fastest way to waste time on artificial intelligence is to begin with the instruction: “Find somewhere we can use AI.”
That question encourages demonstrations, generic use cases and disconnected pilots. It makes the technology the centre of the decision. Established businesses usually achieve more by beginning with operating friction.
Look for recurring loss
Useful opportunities often sit where the business repeatedly loses one of four things:
- Time — people re-enter data, search for information or produce the same first draft.
- Quality — output varies because knowledge and checks are inconsistent.
- Insight — information exists but arrives too late or in a form that is difficult to use.
- Resilience — an important process depends on one person remembering how it works.
Ask team members where work waits, repeats, returns for correction or requires unnecessary interpretation. These observations create a more credible opportunity list than a catalogue of AI features.
Choose bounded work first
Early applications should have a clear input, a reviewable output and a manageable consequence if the system is wrong.
Examples might include classifying incoming enquiries, drafting summaries from approved source material, extracting standard fields from documents, preparing a first-pass comparison or helping staff retrieve internal procedures.
Avoid beginning with autonomous decisions that affect safety, employment, credit, legal rights or material customer commitments. Human review should be explicit, not assumed.
Establish a baseline
Before changing the workflow, measure the current state. How long does it take? How often is it performed? What error or rework rate occurs? Where does it wait? What does poor quality cost?
Without a baseline, a pilot can feel impressive while producing little economic improvement.
Design the whole workflow
An AI step rarely creates value in isolation. Decide how information enters, who reviews the output, how exceptions are handled, where the approved result is stored and what happens when the system is unavailable.
The surrounding workflow determines whether the tool saves time or simply moves effort to a new place.
Protect information deliberately
Understand what data the tool receives, where it is processed, whether it is retained and who can access it. Customer, employee, financial and commercially sensitive information requires appropriate controls and, where necessary, specialist privacy and security advice.
Scale evidence, not enthusiasm
Run a limited pilot, compare performance with the baseline and collect feedback from the people doing the work. Continue only if the improvement is meaningful and the controls are proportionate.
The aim is not to become an “AI-enabled business” in the abstract. It is to build a better-operated business—one carefully chosen friction point at a time.
