Where value is won, and where it is quietly lost.
Most businesses do not lose value in one dramatic decision. They lose it in a hundred small ones: a customer segment that never covered its cost to serve, a price that was never tested, a technology programme that was approved on enthusiasm and never held to a payback period. These notes are about finding that leakage before a buyer does — and fixing it while the business still owns the outcome.
Margin is a design decision, not a residual
Gross margin is usually treated as an outcome — what is left after pricing, procurement and delivery have each done their own thing. It is better read as a design decision. Which customers, products and contracts are structurally profitable, and which are profitable only on average, subsidised by the rest of the book? Answering that honestly usually moves more value than any efficiency programme, and it takes weeks rather than quarters. The discipline is to put cost-to-serve beside price on the same page, per segment, and then make a decision the organisation can live with.
AI-to-EBITDA: holding every case to a number
The question is not whether AI can do something impressive. It is whether, in this business, a specific deployment changes cost, revenue or capital efficiency enough to repay its build, its integration and its supervision within an acceptable period. That means each candidate case gets a baseline, a measured effect, a payback period and an owner. Cases that cannot clear the bar do not enter the plan — however good the demonstration was. In most engagements the honest finding is that the data, process or pricing fundamentals have to be fixed first, and that is worth saying plainly rather than selling a programme around it.
The EU AI Act turns governance into an operating requirement
Governance used to be a policy document. Under the EU AI Act it becomes an operating requirement: knowing which systems you deploy, how each is classified by risk, what evidence exists for the claims made about it, and who is accountable when it fails. Boards are increasingly asked to sign for that. The organisations that handle it well do not build a parallel bureaucracy — they map AI use to the risk tiers, attach documentation to the existing change process, and give one named owner the authority to say no.
For funds, the same logic runs through diligence: technology and data questions asked early enough to price them, and a first-hundred-day plan that assumes the thesis has to be proven in the operating numbers, not in the investment memo.
Discuss a diagnostic