The Future of Insurance Technology: Predictions Worth Holding Us To

On 22 July 2022, a Swiss company called B3i Services AG filed for insolvency. It had no creditors chasing it and no scandal attached. It simply ran out of belief.

B3i was the insurance industry’s official future. Fifteen of the world’s largest insurers and reinsurers built it together to move the market onto blockchain, raised tens of millions of dollars, incorporated in 2018, and shut four years later when the shareholders declined to fund it again.

John Dacey, Swiss Re’s group chief financial officer and an original investor, gave it an honest obituary: “I think it was a very quality effort, but at the end of the day, we did not see the volumes in the demand that would have justified continued investment in this platform.”

The industry’s record of predicting its own future is poor.

Better still, the record is checkable. So before adding new predictions to the pile, this piece grades the old ones.

Grading the last decade’s predictions

Blockchain was the loudest bet, and it failed in the most instructive way possible: not because the technology broke, but because nobody needed it enough to pay for it. After the insolvency, B3i’s codebase was offered as open source and pieces were absorbed into ACORD’s standards work. The future arrived as a file format.

The second-loudest bet is still open, and it is losing. McKinsey’s widely cited Insurance 2030 work predicted that manual underwriting would have “ceased to exist” for most personal and small-business products by 2030, with more than 90% of pricing and underwriting tasks fully automated and claims headcount down 70 to 90% against 2018.

Four years out, here is where production actually stands. Camunda finds 11% of agentic AI use cases reached production in the last year. Sedgwick puts carriers at 7% “scalable success”. Bain finds 4% of insurers have scaled AI in claims. Three studies, three definitions, one order of magnitude - and we went through why in our piece on the pilot-to-production gap.

The quiet bets are the ones that paid. Nobody held a summit for APIs, and integration speed quietly became the thing distribution partners select carriers on. Parametric insurance escaped catastrophe modelling and now pays on cloud outages and sensor data. Embedded distribution grew into the default for whole product categories without ever being anyone’s keynote.

Other graded entries run the same way. Chatbots were supposed to replace service teams; Klarna famously tried outside insurance, reversed course, and the insurers who watched learned to put humans back at the escalation points. Telematics was supposed to make usage-based motor cover the default a decade ago; it became a useful rating input instead of a revolution. Direct-to-consumer digital carriers were supposed to disintermediate the industry, and the one usually cited as proof, Lemonade, spent ten years building technology before its loss adjustment expense ratio finally justified the story, a decade its imitators did not have the capital to copy. The predictions that failed all imagined a new industry. The changes that stuck made the existing one faster at specific, unfashionable tasks.

Score the decade honestly and a pattern falls out.

The centralised, consortium-scale visions failed. The boring plumbing compounded.

Follow the budget, not the keynote

Money is the more reliable forecast. Gartner puts global insurance IT spending at $256.8bn for 2026, up 9.4%, growing at 8.6% a year through 2029. Inside that total, software is growing at 13.4% a year, nearly twice the pace of IT services.

Forrester describes the 2026 shift in US insurance technology budgets as moving “from modernization to intelligence”. That is the polite version. The unpolite version is that PwC found roughly 70% of insurer IT budgets still go to maintaining existing systems, which means the intelligence is being bolted onto estates that consume most of the money just standing still.

Both things are true at once.

Together they say where the pressure lands: the fastest-growing spend category is software that promises change, purchased by organisations whose budgets are mostly consumed by software that resists it.

What is already real, as opposed to trending

Strip out the projections and 2026 already contains most of the future worth discussing.

Regulation went live. The EU AI Act’s high-risk obligations for insurance pricing took effect on 2 August 2026, putting duties on deployers, not just vendors. In the United States, a court in the Lokken case ordered UnitedHealth to produce the governance record around its care-duration algorithm while leaving the source code alone. The regulatory unit of account is the decision record, not the model.

The market began insuring and excluding the same technology simultaneously. Verisk’s generative AI exclusions became attachable to general liability renewals in January 2026, while a new market of AI-specific covers - Armilla, Munich Re’s aiSure, AIUC, Testudo - sells the risk back. Cyber wordings were rewritten three times in three years, for war, accumulation and AI, a churn we documented in the cyber statistics report.

The white space is visible too, and it is niche-shaped rather than platform-shaped. The covers growing fastest from small bases are the ones built on new measurable events: parametric triggers on cloud outages and sensors, AI performance warranties, embedded protection inside someone else’s checkout. Each is small. Each rewards whoever can launch and revise a product quickly rather than whoever has the largest balance sheet. That is where the opportunity actually sits for mid-sized carriers and MGAs, because scale is not the entry ticket - product speed is.

And capital made up its mind. Gallagher Re counted $2.44bn of insurtech funding in the second quarter of 2026, the most since 2022, with 99.1% of it going to AI-focused companies while early-stage funding fell 51.8% in a single quarter. There is no longer an insurtech investment category in any meaningful sense. There is an AI category with insurance use cases.

Five predictions you can hold us to

Each one is dated, grounded in a number above, and states what would prove it wrong.

1. McKinsey’s 90% will not happen by 2030. Production rates of 4 to 11% do not become 90% in four years in an industry where 70% of the IT budget is maintenance. The gap will close from the unglamorous end - document intake, submission triage, renewal processing - exactly where Zurich’s rollout with Cytora already works. High confidence. Wrong if any tier-one carrier reports majority straight-through underwriting on small commercial business before 2030.

2. By 2028, the audit trail becomes something insurers buy, not something they assemble. The EU AI Act demands logging and human oversight; American courts are ordering production of governance records; capacity providers already ask. Decision records will be a procurement line item the way document management became one. High confidence. Wrong if AI-related regulatory penalties and discovery orders stay rare enough that carriers keep treating audit as an afterthought.

3. By 2028, most new SME and embedded premium will arrive through an API rather than a portal or an email. Distribution partners already select carriers on integration speed, and every embedded programme defaults to it. Medium-high confidence. Wrong if API-delivered premium stays a minority of new SME business in the markets that publish channel data.

4. Wordings volatility spreads beyond cyber. At least one more major line rewrites its exclusions twice within a 24-month window before the end of 2028, with AI liability the likely candidate as the Verisk exclusions and the affirmative AI covers collide in court. Medium confidence. Wrong if AI liability wordings stay stable through 2028.

5. The differentiator stops being model quality and becomes time-to-change. Models are converging into utilities; WTW’s Magdalena Ramada Sarasola already argues orchestration beats raw intelligence. By 2029 the platform question in RFPs will be how fast a product, rating table or exclusion can change in production, with evidence. Medium confidence. Wrong if carriers keep buying on model benchmarks rather than change cadence - observable in what vendor marketing leads with.

Hold us to these.

This page carries its publication date, and the graded section above is what we expect someone to do to us in 2030.

The future, minus the waiting

Look back at the five predictions and they are one prediction wearing five hats: the winners will be the organisations that can change their insurance products as fast as the wordings, regulations and channels around them change, and can prove what they changed and when.

That is the bet Openkoda makes now rather than in 2028. Products, rating and rules live as configuration a business user can change with a reviewable change-set instead of a release cycle. Rates and eligibility sit in effective-dated tables, because prediction four means your exclusions will move again. Every decision writes its own audit record, because prediction two means someone will ask for it. And distribution is API-first, because prediction three is mostly already true.

The industry’s last official future went insolvent on a Friday in Zurich. The one that actually arrived was built out of unglamorous parts: interfaces, tables, logs and the ability to change them quickly. We think the next one looks the same, and the predictions above are where we have put that on the record - watching a product change in production is the fastest way to test whether we mean it.