Five chairs. One thing on the table.
Part 1 was the problem: an industry that prices everybody else's risk has never priced its own, and the floor and the mould it has never measured now decide how much of the market it can reach. This is the solution — the five people who have to agree to it, and the one engine all five could. Not a system to buy: a way of building and running products in which a new shape is assembled rather than built. And, at the end, one page you can lay over your own operation before you speak to anybody.
You have heard the number. Seventy per cent of transformations fail.
It gets quoted in steering committees, in board papers, in the careful email explaining why the pilot will not be extended this year.
It is also not true.
The figure entered the world in 1993, in a book on business re-engineering, where the authors described it as their "unscientific estimate" that somewhere between 50 and 70 per cent of efforts fell short. A journal article the following year restated it as fact and rounded it to seventy. One of the original authors later clarified that it had been a descriptive observation, not a finding. In 2011 a peer-reviewed study went looking for the evidence behind five published versions of the claim and found none — no valid, reliable empirical basis for any of them.1
Thirty years, thousands of citations: a number nobody measured, repeated until it became a reason not to act.
Which rhymes with Part 1. I cannot tell you how many decisions this figure has changed — nobody can, and I am not going to claim it. What I can tell you is that I have heard it quoted in the room, more than once, by people who had no idea where it came from and no reason to doubt it. An unmeasured number is doing work in an industry that has not measured its own. The blind spot and the excuse are at least neighbours.
Part 1 ended on three layers — structure, legacy, and the time it takes people to decide — and said the third was the subject of this one. So into the room where those decisions get made.
Every meaningful decision about how an insurer works passes at least these five chairs. There are usually more — a group function, a joint-venture partner, a distribution head with a view — but these five recur, and the same five sit at every administrator, every claims operator and every reinsurer in the chain.
I have sat across all of them. Not one is a blocker. Each is protecting something that needs protecting, and each has an answer that is correct when judged by what they are measured on. The right-hand column of each card is not a wish list. It is a specification — and the orange line beneath it says, in mechanism rather than in branding, what meets it. An interest to declare: my firm builds a layer of this kind, which is why what follows is written as configurations and use cases rather than as a product. If the mechanism is wrong the vendor does not matter; if it is right the vendor is a detail.
Put those five in a room and something quietly illogical happens. Each protects their own exposure, the union of five reasonable cautions is a veto, and the outcome nobody chose becomes the outcome everybody gets.
Not a failure of character — a well-documented feature of how people decide, and research older than most of the systems under discussion.
In 1988 two economists showed that people stay disproportionately with whatever is presented as the existing arrangement — and that the advantage of the status quo grows as the number of alternatives increases.4 The more options a committee is offered, the likelier it chooses none of them. Losses from switching weigh heavier than equivalent gains, and harm from doing nothing feels less culpable than harm from acting.
Insurance knows this intimately. When two US states changed the default motor cover presented to drivers in the 1990s, only about a quarter moved off the default despite materially different costs.4 We have relied on that behaviour ever since — it is why renewal books are as sticky as they are.
We have been monetising the status quo bias of our customers for thirty years while being governed by it ourselves.
A study of 6,000 executives and employees across fifteen countries found the decisive factor in failed change is neither strategy nor money but the false consensus effect: leaders assume their own view is widely shared. The finding worth stopping on — around 70% of executives report feeling positive about a change they know nothing about.5
Read that beside the five chairs. Everyone leaves believing the others are broadly on board; each waits for one of the others to move. Nothing is refused and nothing happens — the most expensive outcome available, and the only one that never appears in a set of management accounts.
So the question is not how to overcome five objections. It is what single piece of evidence would answer all five at once — and, before that, what would have to exist for the evidence to be worth having.
Measure the floor and the next question arrives: what lowers it, and what breaks the mould?
One thing does both, which is the whole reason it can get past five people at once. Read the five cards above again and every "what would change the answer" is a property of the same engine — beside the core, one workflow, weeks, governed, reversible. Not five arguments. One machine, seen from five chairs.
Not a core replacement. Twenty years of evidence says the core is where transformation programmes go to die, and the first person in the room has the scars. What has changed is that lowering the floor no longer requires touching it: three layers of capability now sit on top of whatever is already there.
Part 1 ended on a promise — that an engine built in modules could maximise customisability without raising the cost of a policy, because a new shape is made by assembly rather than rebuild. This is that engine, and the honest way to describe it is not by its name but by what can be configured in it.
"Configuration" has been used loosely enough in this industry to mean nothing. Concretely, these are the surfaces that stop being code and become parameters — each editable, versioned, and validated against the filed product before it goes live.
Unlike the use cases below, this list is meant to be complete. Between them the nine cover the life of a policy from definition to reporting. If something in your operation falls outside all nine, that is worth knowing on its own — an unconfigurable step is exactly where a launch stops.
Configuration matters only for what it lets an insurer sell and run that it could not before. Six examples — chosen because each closes a gap named in Part 1, not because the list ends there. The better ones usually come from inside a book somebody already knows.
None of this requires the policy administration system to change — only a layer that reads from it, writes to it and runs the work around it, which is why a first programme can be live in about ninety days rather than three years, and why it can be stopped without a migration.
That last property is not a technical detail. It is the single feature that makes "yes" available to a person who has said "not yet" for a decade. And it is what Part 1 asked for: a floor that can be lowered and a mould that can be broken without a three-year programme — affordability and customisability from the same layer, at the same time, because the engine was built in modules.
None of this is a forecast. Every figure below was published by the organisation that produced it, and where a number is company-reported rather than independently audited I say so.
The clearest evidence sits in a listed company's shareholder letter. Between late 2022 and the end of 2025 one digital-native insurer added 1.2 million customers while its headcount fell 6%. In-force premium reached US$1.24 billion, up 31% year on year; premium per customer rose 7%; the gross loss ratio improved from 63% to 52%.6
Read together, that is the thing Part 1 said does not happen in insurance: more customers, better risk selection, fewer people. Cost per policy did not fall because somebody negotiated harder. It fell because the work changed shape.
Asia's largest private insurer reports 93% of policies underwritten within seconds and an average life and health claim processed in 7.4 minutes.7 Company-reported, across a book of 240 million retail customers.
Against Part 1's benchmark — the best year the US property claims industry ever recorded, 40.7 days to final payment. Different lines, different markets, not like-for-like. But no definition of "different" stretches from seven minutes to forty days.
A tier-one European group reports an AI underwriting workbench in its UK commercial business cutting processing times 78% in under two years, a Chilean health claims service settling eligible claims in minutes, and — the one that matters for revenue — voice analytics at a Spanish joint venture lifting retention by 20%.8 Company-reported again, and a direction of travel rather than a guarantee.
That retention figure is Part 1's growth lever with a number attached. Not a cost saving: the same book, kept.
What would that mean here? Take a Singapore carrier administering 500,000 policies with 400 people in operations. Apply the ratio a listed digital insurer achieved — roughly 40% customer growth against a 6% reduction in headcount — and it would serve around 700,000 policies with about 375 people.
At a fully loaded cost of S$120 per policy-year, cost per policy would fall to roughly S$86 — a 28% reduction. And on Part 1's argument, that is not a saving. It is the floor dropping 28%, which brings every household whose premium sits in that band — smaller, more bespoke, or both — inside the addressable market for the first time. Break the mould as well, and the products that reach them can be shaped to fit.
The honest position: the proof that the floor can move is public and undisputed. What nobody can yet tell you is how far it moves in this market, for this product mix, under this regulator — because the starting point has never been measured.
Not by us, and not by anybody. Which is where the next page comes in, and why it comes before any of this.
Everything above is an argument until somebody lays it over their own operation. So here is the instrument, in full, with nothing held back for a meeting.
One page, one line of business, one working session. It is shaped like the canvases businesses already use to argue on a wall, because the blocks are not a checklist and where they sit is the point. It reads left to right along the top: the floor, then the mould, then what the two of them together make of you. Underneath sits what is holding it, and beneath that what it costs you and what you would do about it. Fill it with the five chairs in the room, because the object is not the answer. It is that all five are looking at the same page at the same time.
If the canvas comes back with a low floor, a short time to market, an empty never-shipped list and a small unreached segment, then the argument in these two pieces does not apply to you, and I would like to meet you for an entirely different reason.
If it does not, you are holding something the industry has never had in writing: the shape of your own unpriced risk, on one line of business, on one page.
Fill the canvas for one line of business, and then come and argue with us about what it says. That is the whole ask. The canvas is above in full; it needs nobody's permission and no budget, and everything written on it belongs entirely to you. We would simply like to be in the room when it is read, because the conversation that follows a measured floor is a different conversation from the one that follows a proposal.
Two things follow from that, and both are optional.
Put your page into the market baseline. Four readings from it are ones no market anywhere publishes, and we are building them here: what it costs to run one policy for one year; what it costs to take one claim from notice to payment; how many days pass between a risk being bound and its reinsurer seeing it; and where each dollar of the health claims dollar goes — each with the time it takes to change the thing it measures, because a carrier that is cheap and rigid is no better placed than one that is dear and quick.
They are not built from an opinion survey. Each comes from a fixed method — what counts as a cost, what is in scope, what is excluded — applied to a participant's own ledgers in a working session. You get your own result back, on the same definitions as everybody else and set against the anonymised distribution: where you sit, rather than what you score. Only the aggregate is published, annually, on the same definitions, which is what makes year two worth more than year one. Contribute anonymised, or put your name to it and help set the benchmark. Closed-door sessions run from November, beginning with carrier leaders.
Or test the specification. Take what you wrote in the last block — one product, one channel, one quarter — run beside your core with your own operations team as the judge. If the floor does not move, you switch it off. That is the offer to all five chairs at once.
Five people in a room, each right, is how an industry stands still for twenty years.
One engine that answers all five is how that stops.
Which chair are you sitting in?
→ Start the conversation. If any question in these two pieces stayed with you longer than it should have, that is the one to begin with.
This piece was written by human intelligence. For now.