Where the value actually sits

When I am not busy delivering AI transformation projects that drive EBITDA, I spend some time on stage at AI events, or in the audience.

I noticed something.

The companies that listened to Microsoft and Salesforce pitches and rolled out fleets of agents are quietly realising the same thing: no value was created.

It is not a surprise. At best, you made one task a little faster. In most cases, all you really added was licence cost and a dip in productivity while everyone learned the tool. As I keep saying, rolling out Copilot and expecting the benefits of AI is like rolling out Outlook and expecting the benefits of digital transformation.

So I am glad the agent hype is finally cooling. At least I hope it is.

And to be fair, I am hearing better advice now. It is not about the tools, it is about redesigning the workflow.

True?

Not quite.

When you actually redesign a workflow, the real work is not rearranging the tasks. A workflow is a sequence of decisions, with steps in between to feed them. The enquiry is read so a decision can be made. The data is gathered so a decision can be made. The document is drafted to record a decision already taken.

Redesign the tasks but leave the decisions where they were, and you get a faster version of the old workflow. The economics do not move.

The economics move when the decision itself moves. From a person making it slowly, a few times a day, to a system making it consistently, hundreds of times, at near-zero marginal cost.

And here is the irony. That is exactly what an agent is. An agent reasons over memory, data, tools and instructions to reach a decision and act on it. It is a decision engine. The problem was never the agents. It was pointing them at tasks instead of decisions.

So “redesign the workflow” really means “redesign and automate the decisions inside it.” I call it Decision-Centric Transformation.

This is also why AI transformation needs a different approach from the one most teams are running. You do not start with the tool, or even the workflow. You start with the decision: what is being decided, by whom, on what data, at what cost. Design the decision first, then build the agent architecture around it. The digital transformation playbook will not get you there.

And it is how you unlock the significant returns. Do this across the decisions that matter and the gains are not incremental. This is how AI moves EBITDA instead of adding licence cost, and why it belongs in the value creation plan, not the IT budget.