Peak3's Insurance AI Delivery Lifecycle Targets 80% Cost Cuts

Peak3, the Singapore-based insurance software provider formerly known as ZA Tech, has put a number on what it believes artificial intelligence can do to the cost of building insurance systems: a 50 per cent reduction now, rising to a targeted 80 per cent within three years. Its new offering, Graphene Harness, is positioned as the first insurance-specific AI delivery lifecycle, a governed environment that applies AI across the entire software build process rather than at the coding step alone.

The claim carries an important qualifier. Peak3 currently uses Graphene Harness to build and maintain its own Graphene platform and is extending it to selected implementation partners, with wider availability to clients planned over the next six to 12 months. In other words, this is a capability entering the market through partners, not yet a product any insurer can buy off the shelf. The 50 per cent figure is company-stated and presented without a named customer case study; the 80 per cent target sits 18 to 36 months out.

What has Peak3 actually launched?

Graphene Harness applies AI across business analysis, architecture, engineering, testing, migration, deployment and maintenance. Peak3 describes it as a virtual technology team rather than an assistant, having modelled the roles of business analysts, architects, engineers, testers and reviewers into AI counterparts that operate inside a single governed delivery environment.

It arrives alongside Graphene v4, the latest version of Peak3's core insurance platform. That release adds an upgraded agent orchestration and run-time layer plus a marketplace of pre-built AI agents covering areas such as medical underwriting, conversational first notice of loss, intelligent document processing, and fraud, waste and abuse in claims.

Peak3 says agents can reconfigure products, calculations and rules from natural-language instructions and unstructured documents, cutting configuration tasks that once took days to hours or minutes, with access management, attribution and logging governing what each agent is permitted to do.

How does an AI delivery lifecycle differ from a coding assistant?

The distinction Peak3 is drawing, AI across the whole lifecycle rather than AI bolted onto the coding step, is not one it originated. The concept of an AI-driven development lifecycle, or AI-DLC, was defined by Amazon Web Services, which published the methodology on its developer blog in mid-2025, open-sourced the supporting workflow later that year, and released a dedicated financial-services version in June 2026. AWS frames AI-DLC as a three-phase model in which AI drafts requirements, architecture, code and tests while humans validate each step.

Peak3 has kept the acronym, substituted "delivery" for "development", and applied the approach to the insurance vertical. Read that way, the "first global insurance AI-DLC" positioning describes the first insurance-specific packaging of an existing methodology rather than the creation of a category. That is a meaningful move in a sector defined by regulated, long-lived core systems, but it is an adaptation, and worth attributing as such.

Why is Peak3 targeting implementation costs specifically?

Because that is where the money in insurance technology actually sits. Industry benchmarks put insurer IT spending at roughly 3 to 6 per cent of earned premiums. For a large property and casualty carrier writing around 5 billion US dollars in premium, a full core modernisation typically runs 250 million to 480 million dollars over five to seven years. Within that, implementation and consulting services, commonly 85 million to 160 million dollars, are usually the single largest line, ahead of software licensing at 45 million to 95 million dollars. Full platform replacements with enterprise vendors routinely take 24 to 36 months.

That is the cost base Graphene Harness is aimed at. Peak3 chief executive Bill Song has framed the shift as moving from selling software and related services to selling the production line that builds them. The editorial implication is direct: the target is not only the customer's implementation bill but the professional-services revenue that insurance software vendors, Peak3 among them, have historically relied on. How the company squares an 80 per cent cost-reduction pitch with its own services economics is a question worth pressing.

Can AI-driven delivery meet insurance governance standards?

Insurance core systems are mission-critical and heavily regulated, which is why Peak3 stresses access controls, attribution and detailed logging around every agent. That emphasis is warranted. Independent commentary on AI-DLC as a method has flagged that generating a plan, its code and its tests within the same working day can erode the very controls, peer review, audit trails and release gates, that regulated workloads depend on, and can do so faster than new controls are put in place. The open question for any insurer evaluating this approach is whether governance scales at the same speed as delivery. Peak3's answer is a single governed environment; the proof will be in audited deployments at client carriers rather than internal use.

Where does this leave the core-platform incumbents?

The competitive frame matters. Guidewire anchors the enterprise P&C market and accounts for roughly 42 per cent of new core implementations, with Duck Creek, Sapiens and Majesco competing across the mid-market and internationally. Their commercial models also lean heavily on implementation, much of it delivered through systems-integrator partners. If AI genuinely halves build and implementation effort, the pressure lands not only on project timelines but on the services margin pool across the whole category.

Peak3 is the smaller player in that field. Founded in 2018 and rebranded from ZA Tech in 2024 alongside a 35 million dollar Series A led by EQT, it reports more than 50 insurance clients across over 20 countries and says more than a billion policies are transacted on its systems each year. Those figures are company-stated. Its strategic bet is to reframe the contest around delivery cost rather than feature breadth, a positioning that only holds if the cost claims survive independent scrutiny.

Why This Matters to FinanceX Readers

For insurers and their finance chiefs, implementation is the largest and least predictable line in any core-system programme, so anything that credibly halves it reshapes the build-versus-buy calculation that has defined technology budgets for a decade. For investors, the numbers to watch are whether the 50 per cent reduction is corroborated by named client outcomes, and whether a model built on selling the production line cannibalises the services revenue that vendors in this space have long depended on. For the wider market, this is a signal that AI competition in insurance software is shifting from individual features to the economics of delivery itself.