The product block
Every product line, run as one loop - margin, complaints and suitability read together, acted on before the regulator or the P&L forces it.
The pain
“Nobody can tell me which lines actually make money, or prove the product is fair when asked.”
A product line is judged by whichever function is looking at it. Finance sees margin, service sees complaints, compliance sees suitability, sales sees the mix - and no one sees the line. So the evidence that a product has stopped working arrives late and in pieces: complaints clustering here, a claims ratio moving there, margin eroding somewhere no one reconciled until the quarter closed. The line that should have been repriced or retired runs on, and the fair-value case for it cannot be assembled until a regulator asks. In insurance and wealth this is product governance and fair value; in legal, service-line profitability; in distribution, range and pricing.
What the block is
The object is each product or service line - one continuous picture per line, from design and pricing through performance and suitability to retirement. Everything it produces feeds one live record: design and pricing, margin, complaints, suitability outcomes, claims ratios, fair-value assessments. A model reads it permanently. Your product, finance and service systems stay exactly where they are - the block reads from them, it does not replace them.
The loop in action
Each output states what it saw, why it matters, what it recommends, and how confident it is - every fact linked back to its source, in an order a product owner can read and a regulator can examine.
You set the dial
Every capability has four positions: observe - it reads silently and measures its own accuracy; recommend - it surfaces the case and a proposed action to a named owner; draft - it prepares the action for a human signature; act - it executes within agreed bounds and logs everything.
Everything starts at observe. Promotion is earned on the measured record and reversed instantly if it doesn’t hold. Repricing, withdrawal and anything a regulator relies on stay behind a human signature wherever your policy demands one. The mechanics are the immune gate described in our framework.
How it lands
It starts narrow: one product line or one segment of the range, the block at observe against your own past outcomes. Nothing is replaced, nothing is migrated - the first signals flow from the systems you already run. From there it is a bounded engagement - blueprint, build, handover - and your team owns the result outright: the signal map, the judgement rules, the dial settings, the working loop.
The backtest
Take three past lines: one you retired too late, one whose margin drifted for a year unseen, and one you could not prove was fair when you were asked. What did your own record know, and when?
That is the first question the diagnostic answers. Take the ten-minute diagnostic, or write to hello@somai.studio.
Runs on our framework · related reading: Most AI spend never reaches the P&L