New KPIs for the AI age
The following article was originally published on Insurance Day on 08 May, 2026.
The insurance leaders must stop treating AI as a tool that assists the existing operating model.
The long-term winners will those that treat AI as the foundation of a new operating model
08 May 2026 | OPINION by Max Richter
Re/insurers must promote artificial intelligence from mere ‘assistant’ to the foundation of operating models
THE BIGGEST MINDSET SHIFT FOR INSURANCE LEADERS IS TO STOP TREATING AI AS A TOOL THAT ASSISTS THE EXISTING OPERATING MODEL. (Source: Miha Creative – stock.adobe.com)
The working heart of insurance lives in the broker emails, the PDFs, the bordereaux, the spreadsheets, the handoffs and the judgement calls between them.
The industry has for years digitised the islands of this process, but it has left the swamp between them largely untouched, which is why so many transformation programmes have produced disappointing returns.
Automating individual tasks inside a fragmented workflow leaves the fragmentation in place and the operational burden stays where it always was – in the gaps between systems and steps.
Agentic AI changes the question. Instead of asking whether artificial intelligence (AI) can extract a data field or draft an email, the right question is whether AI can own the journey from intake to outcome.
When AI ingests, interprets, validates, decides, communicates and updates systems across multiple steps, the work shifts from task automation to operational redesign. Humans move into exception handling, control and judgement. Scale stops coming from headcount and starts coming from the percentage of work that flows straight through.
The breakthrough sits in AI doing the work while people handle the exceptions, the negotiations, the broker relationships, the coverage calls and the judgement calls where experience matters most.
For this model to hold in insurance specifically, the AI has to understand insurance, but generic large language models, however capable, were built for general use.
The systems that handle complete insurance journeys reliably are the ones built on language models trained on insurance data, layered with knowledge structures that encode how insurance entities, relationships and concepts connect.
“The breakthrough sits in AI doing the work while people handle the exceptions, the negotiations, the broker relationships, the coverage calls and the judgement calls where experience matters most” | Max Richter, mea Platform
Without that foundation, agentic systems break down at the points where the work becomes hard.
Building this kind of intelligence also requires teams that have run underwriting desks and claims operations, because the points of breakdown are rarely visible from a process diagram.
Most of the delays in insurance processing sit in waiting time and thinking time is rarely the bottleneck.
Inboxes, queues and handoffs are the largest hidden cost driver in most operations. When that delay disappears, what looked like complexity often turns out to have been fragmentation.
The speed-gain matters beyond productivity. When turnaround times collapse, carriers see more of the right business. In-appetite risks surface quickly enough to win. In claims, earlier triage produces cleaner data and fewer avoidable delays. Operational change starts producing top-line and combined ratio effects.
For underwriters, claims handlers and operations teams, the day changes. They stop rekeying data, chasing documents, and managing work from inboxes.
Cases arrive classified, enriched, validated, summarised and routed, with the next best action already suggested. More time goes to negotiation, broker relationships, coverage decisions, and true exception handling. That is where humans add value.
The industry’s instinct to wait for the perfect integration solution now functions, often unintentionally, as a way of preserving the status quo. Operational impact requires removing manual work safely and quickly. Architectural perfection is optional.
Carriers and brokers measure efficiency through familiar KPIs: workflow automation, accuracy and reductions in routine staffing hours. These are useful for tracking improvement within the existing model, but they are insufficient for measuring whether the model itself has changed.
The real prize is combined ratio improvement and margin expansion across the organisation, which requires KPIs designed for the AI age. KPIs that measure outcomes rather than activity.
Five metrics matter most.
Straight-through completion rate: The percentage of work that finishes without human intervention. This is the clearest signal of whether AI is operating or merely assisting.
Elapsed turnaround time: The total time from intake to outcome, measured as a combination of agent processing time, human-in-the-loop time, and waiting time. Strong AI operations move this from days to hours, sometimes even to minutes.
Exception rate: The frequency with which workflows escalate to humans. This shows how well the system handles real-world variability across documents, formats and edge cases.
Rework rate: The proportion of completed transactions requiring correction. High rework erodes cost savings and signals quality problems at scale.
Cost per completed transaction: This connects operational performance directly to combined ratio and margin.
A sixth metric distinguishes platforms that improve from those that plateau: learning velocity. This is how quickly a system absorbs a new exception so that it stops appearing as one. Platforms that learn become harder to compete with over time, but platforms that do not, lose ground.
The biggest mindset shift for insurance leaders is to stop treating AI as a tool that assists the existing operating model. The long-term winners will treat it as the foundation of a new one. They will ask which jobs humans should never have to do again. That is the question that separates the operators using AI to speed up the old model from those using it to replace it.
Max Richter is EMEA chief executive and global growth leader at mea Platform
Read the originally published version in Insurance Day: https://www.insuranceday.com/ID1155896/New-KPIs-for-the-AI-age
Read our framework for measuring KPIs in the AI era here: https://www.meaplatform.com/measuring-kpis-in-the-agentic-ai-era/
