AI as an operating change, not a software purchase.

Straight answers

Questions I get asked before the first call.

What is an AI operating model?

It is the set of decisions that make AI show up in results rather than in a subscription line: which two or three processes you are changing, who owns each one, what metric proves it worked, and how often leadership reviews it. Tools are the last decision, not the first. Without owners and a metric, adoption stays flat regardless of the software.

Where should a company start with AI agents?

Start with the highest-volume, lowest-judgement step in a workflow you already understand — quote assembly, first-response email, data entry between systems. Automate that one step, measure cycle time before and after, then move to the next. Starting with a broad strategy or a company-wide rollout is how pilots die without a number to defend them.

Do I need engineers to run AI agents in my business?

No. I operate products and systems with agents without an engineering team. What you need is someone who can describe a workflow precisely and who owns the output. The tooling is now assembly rather than development. Engineering becomes necessary when you are integrating deeply into custom internal systems, not for communication, sales follow-up or pricing work.

What should AI never be allowed to do in a business?

Anything where judgement or relationship is the product: final pricing concessions, quality claims, hiring and firing, and any legally binding commitment. Agents draft, route and prepare; humans decide. The failure mode I see most is an agent given authority without a named human owner and a visible way to see when it got something wrong.

How do you measure whether AI is actually working?

Pick the metric before the tool. Usually cycle time, first-response time, or cost per unit of output. Take a baseline for two weeks, deploy, then compare the same measure. If the number has not moved after a month, kill it. Licence counts, seats activated and prompts sent are activity metrics, not results.

How long before an AI change shows results?

For a single well-chosen workflow, four to six weeks to a measurable change in cycle or response time. Operating-model change across a leadership team takes one to two quarters, because the constraint is decision cadence and ownership rather than technology. Anyone promising transformation in a fortnight is selling a tool, not a change.

What does an AI audit involve?

A short review of how work actually flows today, including the exceptions people handle silently, then a shortlist of the two or three processes where throughput is the real constraint. You get named owners, the metric for each, a sequence, and an explicit list of what not to do this year. It is an operating document, not a technology report.

Do you implement, or only advise?

Both, and I prefer to start by implementing one workflow. Advice that has not survived contact with a real operation is guesswork. I run agents daily across my own products and systems, so the recommendations come from things that broke in production rather than from a vendor deck.