What thirteen years of this actually taught us.

We have written reports that were technically excellent and completely useless. The cause is almost never method. It is calendar.

A budget bilateral happens on a date. A board meets on a date. A cabinet memorandum closes on a date. An evaluation that lands two weeks later is not late by much, it is late by everything, because the decision it was built for has already been taken on whatever evidence was in the room at the time.

So we design mandates backwards now. We find the decision, we find its date, and we work out what can be defensibly known by then. Sometimes that means a rapid evaluation with stated limits instead of a full one with none. A rapid implementation evaluation delivered inside a single campaign cycle changes the next cycle. A perfect one delivered afterwards changes a filing cabinet.

The second half of this is naming who acts. A recommendation addressed to government is addressed to nobody. A recommendation that names the institution, the instrument it will use, and the cost, has somewhere to go.

Between 2016 and 2025 we completed ninety-one technical spending reviews under a single programme. Some patterns repeat with almost boring reliability.

The first is that the largest savings are rarely in the line item everyone is looking at. They are in the contract structure behind it: a services agreement priced per transaction rather than per outcome, a fleet held rather than leased, a licence bought per seat for an institution that has since halved.

The second is that nobody owns the cost of a shared service. When one agency provides technology, fleet or facilities to everyone else, its price is a transfer, and transfers are the least scrutinised numbers in any budget.

The third is more uncomfortable. A spending review that produces a savings figure and stops has done a third of the job. The figure has to be carried into a budget bilateral by someone with the standing to hold it there, and then into an adjusted allocation. Reviews that are not owned inside the treasury process become interesting reading.

A sustainable finance taxonomy is a classification: what counts as green, what counts as transitional, what does not count at all. Large markets built theirs after the instruments existed. Several of the markets we work in are being asked to build theirs first.

That inversion is not necessarily wrong, but it changes the design problem. A taxonomy written for a deep market is a screening tool for thousands of instruments. A taxonomy written for a shallow one is mostly a signalling device, aimed at development finance institutions, at concessional lenders, and at the domestic banks deciding whether a green loan book is worth the systems cost.

The failure mode is copying a large jurisdiction's thresholds wholesale. Sector coverage that ignores the two industries actually constituting the economy produces a document that is technically compliant and commercially irrelevant.

What works better, in our experience, is a narrow taxonomy anchored to the country's own climate commitment, tied to the budget classification the finance ministry already uses, and built with the central bank and the non-bank regulator in the room from the first workshop rather than the last.