Price optimisation in regulated environments

In a regulated pricing environment — mortgage rates, insurance premiums, any price a regulator or an ombudsman can ask you to justify — the price is a decision under simultaneous constraint: margin targets, competitive position, portfolio risk, and a regulatory obligation to price fairly and consistently across customer segments. The Australian Competition and Consumer Commission’s Residential Mortgage Pricing Inquiry made this explicit for home loans: discretionary pricing and discounting practices came under direct scrutiny, and price transparency stopped being optional.

That’s the same shape as marketing-spend allocation under an inventory constraint, or any commercial decision currently made by a rate card, a rule of thumb, or a relationship manager’s discretion. The lever is different, but the underlying problem of optimising a decision under constraints that keep moving is the same one we work on across every engagement.

A pricing model that performs well in a backtest and never reaches a live decisioning system is a common outcome, not a rare one. The gap is usually the governance path around the model. Getting a price optimiser into production in a regulated environment means the model has to survive the same scrutiny as any other lending or pricing decision: documented sensitivity analysis showing how the price responds to each input, evidence that the model doesn’t produce disparate outcomes across protected segments, a champion–challenger rollout rather than a full cutover, and sign-off from risk and compliance functions who need to be able to explain the price to an ombudsman, not just to a data scientist.

Practically, that means building the evidence trail alongside the model, not after it: every input to the price is logged, every override is captured and reasoned, and the model’s behaviour at the boundary of each constraint is tested before it ever sees a live customer.

The programmes that stall tend to fail in one of two places. Either the model is built in isolation from the people who have to sign off on it — so the governance conversation starts after the model is “done,” and every question sends it back for rework — or the optimisation is treated as a one-off analysis rather than a decision process, so it goes stale the moment the competitive or regulatory environment shifts, which in pricing is constantly.

The fix for both is the same: build the governance conversation and the retraining or re-calibration cadence into the engagement from day one, rather than treating them as a hand-off at the end.

Doryan wrote about this problem in an article for FICO’s decisions blog in 2019, while consulting to retail banks across APAC on exactly this kind of rate-setting problem. The same constrained-optimisation approach (price or allocate a scarce resource under moving constraints, with the governance to keep it trustworthy) underpins the marketing-spend optimiser Anneal built for a DTC beverage business.

Get in touch if you’re looking at a similar problem.