Before implementing AI, understand the business

AI projects often begin with a catalogue of capabilities.

The business is shown what a model can summarise, classify, generate or automate and is then asked where those capabilities might be used.

That sequence is convenient for selling technology. It is not necessarily convenient for solving business problems.

The better starting point is the business.

What is management trying to achieve? Where is work slow or unreliable? Which information is missing? Which decisions depend on judgement? What errors are expensive? Which tasks are repetitive for a good reason, and which are repetitive because the organisation itself is badly arranged?

The requirement comes first

An AI implementation should have a business requirement that can be explained without using the words AI, model or automation.

If the requirement cannot be stated clearly, the project is probably not ready for a technology decision.

Once the requirement is understood, AI may be a sensible option. It may also turn out that a simpler software change, a data-quality improvement, a clearer responsibility or no new technology at all produces the better result.

Avoid automating confusion

AI can accelerate activity.

That is useful when the activity is valuable and well understood. It is less useful when the underlying process is a symptom of organisational confusion.

Putting AI into that process can make the confusion move faster.

The disciplined approach is therefore straightforward: understand the business, define the requirement, examine the evidence and only then decide what technology belongs in the solution.