The executive job: separate signal from hype
The pressure on leaders is to “do something about AI,” and that pressure produces waste — pilots with no metric, tools bought because a competitor announced one, internal builds that should have been purchases. The executive’s real job is not to understand transformer architecture; it is to decide where AI changes your economics and where it does not, then concentrate investment accordingly. Today’s AI is unambiguously good at language tasks: reading, writing, summarising, classifying, and answering questions over documents. Functions dominated by those tasks — support, marketing, research, parts of legal and finance — are where cost structures are already shifting. Functions that depend on physical work, regulated judgement, or trust are far less affected. Clear-eyed mapping of that line is the highest-leverage thing a leader can do, and it requires no technical depth, only disciplined business judgement.
Opportunity mapping and the build-vs-buy decision
Start from the work. Inventory your highest-volume, highest-cost, most repetitive processes and flag the ones built on language. Those are your candidates. Choose one or two with a clear, measurable outcome, run a contained pilot, and prove the number before scaling. On build versus buy, the default should be buy: a commercial tool or a thin internal app over a provider’s API is faster, cheaper, and far lower-risk than a custom model for almost every common need. Reserve building for the rare case where AI is a true source of differentiation and you hold proprietary data and the engineering capacity to exploit it. Most organisations that “built” should have bought, and learned an expensive lesson about the gap between a demo and a production system.
Governance, risk, and measuring ROI
Governance is not a compliance afterthought; it is what lets you adopt AI at all without creating liabilities. The minimum viable framework is four parts: an acceptable-use policy that tells staff what they may and may not paste into AI tools, a list of approved enterprise tools with data-protection and no-training agreements, a mandatory human-review gate for anything customer-facing, and a single named executive accountable for AI risk. On ROI, define the metric before the pilot — hours saved, cost per ticket, cycle time, conversion lift — capture a baseline, and compare honestly. Vanity metrics like usage counts tell you nothing about value. If you cannot point to a process that became measurably faster, cheaper, or better, you have an experiment, not a return — and that is fine, as long as you are honest about which one you have.