Before You Ask for an AI Strategy, Define What Good Looks Like

“We need an AI strategy.”

It’s an understandable request from a board or CEO. The opportunity is significant, the tools are moving quickly, and nobody wants to be the business left behind. My first question would still be a fairly simple one. Why?

What are we actually trying to achieve? A better customer experience? Fewer errors? More capacity? A service we couldn’t offer before? Until we can answer that, choosing the technology feels a little premature.

I’ve spent around 17 years working with business process automation in one form or another. The tools have changed considerably, but some of the most useful questions have stayed remarkably consistent. Where does work get stuck? What causes rework? What should people be spending their time on? And what would better actually look like?

Start at the end

My preference is to get clear on the business outcome, then jump forward and describe the finished state. We can work through the middle once we know where we’re going.

Imagine a customer request has made its way through the business. What has the customer received? Which decisions have been made? Who owns the result? What record remains in the system that the next person can trust?

That last question matters. A useful answer in a chat window may still leave somebody copying information, checking it, chasing approval and updating the customer record. If we haven’t thought about those steps, we may simply have moved the work around.

For me, “good” needs to be observable. The request is resolved, the record is accurate, the customer knows what happens next, and there is a clear owner when something falls outside the normal process. That gives us something concrete to design towards.

Then work through the middle

This is where architecture, people, process, data and technology come together. Some steps might suit a simple rule or an existing workflow. Others might benefit from AI’s ability to interpret information, generate content or support a decision. Some will need an experienced person to make the call.

AI opens up genuinely new possibilities. It can also introduce uncertainty into places where we’ve previously expected a predictable result. Understanding both is part of choosing the right combination.

The people doing the work need to help shape it. They know the awkward exceptions, the unofficial workarounds and the details that rarely make it onto a process map. They also need to know what they are accountable for, when to intervene and how to challenge an output.

Security belongs in that conversation from the beginning. What information can the system access? What can it change? Who can approve an action, and what evidence will remain? Those choices become much harder once a pilot is already handling live work.

ADAPT’s April 2026 analysis of Australian CIO priorities points to data foundations, governance and workforce readiness as barriers to scaling AI. That feels very familiar from an operational perspective. A better tool still needs a business that is ready to use it.

Be sensible about value

I don’t think every AI-assisted email needs its own business case. We don’t calculate a return for every phone call or every task completed on a laptop either. We still expect those investments, taken together, to help the business perform.

The same practical thinking should apply here. Everyday assistance, a redesigned business process and a completely new capability will have different costs, risks and timeframes. Gartner made a similar distinction in March 2026, arguing for different approaches to assessing productivity uses, process improvements and larger transformational investments.

We should be clear about the value we expect and use a sensible way of testing it. Time saved is useful, but where does that time go? Does it create capacity, improve service or reduce rework? What does the whole process cost once checking, exceptions, licences and ongoing support are included?

Start with a baseline. Understand how the work performs today, test a small, bounded change, and agree what would justify expanding it. Include the people who will run it and the cases where it gets things wrong. A promising demonstration is a starting point; the day-to-day result is what earns the next investment.

A better starting question

In my earlier post on human judgement and AI, I wrote about the importance of people owning the consequences. I think the same principle applies before we even select the technology.

If a board asks for an AI strategy, I’d start by asking what they want the business to be able to do, and what they want customers and staff to experience differently.

Get that clear, describe the end state and give somebody ownership of it. Then we can have a much more useful conversation about how AI helps us get there.

Ken Hickey

October 2026

The views expressed in this and all of my blogs are my own and not necessarily endorsed by my employer(s).