Why ChatGPT is not a strategy
Using AI everywhere isn't a strategy any more than not using it is. The businesses getting it right know exactly where variability lives — and how quickly they can move downstream into something predictable.
Why ChatGPT is not a strategy
Ask a room of business owners whether they're using AI and most hands go up. Ask what that means and you'll hear variations of the same answer: someone on the team uses it to draft emails, write social posts, or summarise meeting notes.
That's not a strategy. That's a productivity hack. An occasionally useful one, but a hack nonetheless.
The mistake most businesses make next is overcorrecting. Having decided that ad hoc AI use isn't a strategy, they conclude that the answer is to embed AI into every workflow they can find. That's not a strategy either. That's a different kind of enthusiasm.
Where AI actually belongs
Good technology strategy is about picking the right tool for the right situation. AI is not the right tool for every situation.
Deterministic workflows — the kind that need to be predictable, auditable, and consistent — are where you want automation, not AI. Approvals, compliance checks, structured reporting, regulated processes: these benefit from rules-based systems that do the same thing every time. Introducing AI into processes that need to be reliable and predictable adds variability where you explicitly don't want it.
AI earns its place where you're dealing with unstructured variability. Interacting with customers who express the same need in a hundred different ways. Making sense of legacy data that was never captured consistently. Processing documents that don't follow a standard format. These are problems that defeat deterministic rules and where AI's tolerance for inconsistency is genuinely valuable.
The direction of travel
The goal isn't to use AI everywhere. It's to use AI to handle the variability — and then, wherever possible, to move the output into something deterministic.
The faster you can convert unstructured input into structured, rules-based process, the more predictable your technology costs become. AI is expensive to run at scale and harder to audit. Deterministic systems are cheaper and more controllable. A good AI deployment often has a finite job: absorb the mess, produce something consistent, hand off to a workflow that can take it from there.
That's what a strategy looks like. Not a question of whether to use AI — but a clear view of where variability lives in your business, whether AI is the right tool to handle it, and how quickly you can move downstream into something you can predict and control.
If you're trying to work out where AI actually belongs in your business, that's a conversation we can have.
Robin Carswell
More on
AI doesn't know what day it is
AI will tell you the answer whether it knows it or not. Understanding the structural blind spots — the things it genuinely can't know — is the difference between using AI well and being misled by it.