A faster AI process does not make the business faster

I see more and more discussion about changing processes for AI rather than simply automating tasks with AI. That's a good sign.

AI cannot deliver much business value by optimising an isolated task inside an unchanged process.

But the next question is:

What exactly should the process become?

Making one operation 50% faster does not necessarily improve the overall outcome. The bottleneck may move:

• elsewhere in the process;

• into another process;

• to a supplier or partner.

This can also create material operational risk.

Existing processes depend on a balance among business requirements, people, controls and technology platforms. Change one element and the others may need to change too, or may break.

So, AI-enabled process redesign cannot be:

Find a bottleneck → remove it → done.

It needs a systemic, iterative approach:

Understand the system → identify the constraint → redesign → assess the effects across business, people and technology → observe what changes → address the next constraint.

The solution is therefore not a better AI model, technical integration and a redesigned process. It is a multi-factor redesign of the system, supported by a comprehensive integration strategy.

AI itself may not be the most complex or expensive component of that change.

Each step raises a harder question.

The next, harder question is:

How do we redesign the system so that removing one constraint does not create another and/or introduce a new operational risk elsewhere?

That, I think, is where AI integration starts becoming a systems-design problem rather than an automation problem.