MARK’S SIGNAL
Focus on the targeted business outcome, not the solution. Use AI where it is needed, and make clear who recommends, decides and executes.
KEY POINTS
- AI activity and utilisation are not the same as measurable business value.
- Insurers should move from many pilots to a focused set of priorities that can scale across end to end processes.
- Modern AI deserves particular attention where insurance work depends on large volumes of unstructured information.
- AI should be used selectively within a wider architecture, alongside workflow, rules engines and established controls.
- Agentic AI requires explicit authority, accountability, testing, monitoring and intervention.
Creating business value from AI in insurance starts with familiar management disciplines: a defined outcome, measurable targets, committed business ownership and clear controls. We need to be specific about the value we intend to create, how we will measure it and what evidence will justify moving to the next stage.
The scope must be focused enough to execute and broad enough to make a meaningful difference. These foundations apply to both generative AI and agentic AI. The challenge is to keep the business outcome in view while accounting for what these technologies can do, where they can fail and how they will fit into the wider process.
Utilisation is not a measure of business value
Giving employees access to large language models and room to experiment can help organisations learn. Measures such as the number of users or queries are useful indicators of adoption.
Those measures do not, however, show whether AI is creating tangible value. Usage can affect cost without necessarily improving revenue, customer outcomes, decision quality or operating performance. We now need to tighten the connection between AI activity and the value we intend to capture.
Move from AI pilots to focused, scalable priorities
Early AI committees and pilot projects helped organisations learn what modern AI could do. That exploration had a purpose. The next requirement is focus.
We should select a manageable set of important business outcomes, commit ownership and resources, and move the capabilities that work into scaled operations. A long list of pilots can demonstrate activity while leaving the organisation short of material change.
Design across the whole insurance process
An initial question for many organisations was: how can AI improve individual productivity? That question is useful, but it can limit the scope of the opportunity.
We should think bigger and bring wider value into scope. How could AI help transform new business processing from point of sale through policy issue? How could it improve claims from submission through settlement? Looking across the process helps us identify where information, decisions, handoffs and controls can work together more effectively.
Use AI to make unstructured insurance information useful
Insurance is inherently document intensive. Commercial underwriting, complex individual underwriting and claims all depend on varied material, including medical reports, police reports, bank and investment records, engineering assessments and contractor quotations.
A major part of insurance work is distilling this information into a standardised view that supports decisions which are consistent, repeatable, explainable and aligned with company policy. Modern AI can help extract, summarise and interpret unstructured information. This capability deserves attention because it addresses a substantial part of insurance work. Its usefulness depends on the accuracy of the resulting information and the controls around its use.
Combine AI with workflow, rules and established controls
We should not insert AI into every step simply because it is available. For some tasks, established technologies may offer lower cost and more predictable control. The stronger design uses each technology for what it does best and integrates AI where its capabilities are genuinely required.
Illustrative example: a high net worth insurance submission
In a possible design, workflow coordinates the submission. AI extracts and summarises medical reports and financial evidence for review. Existing interfaces support know your customer checks, while an underwriting rules engine applies defined policy and routes exceptions for review. The example illustrates how the components could work together; it is not a reported deployment or measured result.
Define who recommends, decides and executes
AI introduces a different risk profile. Probabilistic models draw on data from a world that continues to change. Depending on the permissions granted, agentic AI may take actions, make decisions or enter commitments on behalf of the company. The potential value is significant, and so are the responsibilities.
We need to make it clear who or what recommends, decides and executes. If underwriting authority is delegated to an AI agent, the organisation should understand in advance where accountability and liability sit when the agent makes an error or acts outside its intended purpose. We should not wait for a disputed claim or a challenge from a reinsurance partner to find out.
We also need robust, tested controls to identify potential bias and tightly manage the authority granted to AI agents. Those controls need to remain effective after deployment through monitoring, escalation and the ability for people to intervene.
What leaders need to get right
The number of AI tools, users or pilots is not the measure of success. Better business outcomes are. Insurers that define value clearly, focus their priorities, redesign important processes, use AI selectively and establish explicit accountability will be better placed to turn innovation into durable business value.AVER SOLUTIONS PERSPECTIVE
At AVER Solutions, we see Mark’s contribution as a practical test for AI investment: can the organisation explain the outcome it intends to improve, the evidence it will use and the person accountable for the result?
For insurance and banking leaders, that means connecting process design with a baseline and review criteria. Depending on the objective, useful measures might include processing time, rework, decision quality or customer outcomes, assessed alongside cost and risk. These are possible measures, not results claimed for any organisation.
As AI moves from providing information to taking action, the same discipline should connect each permission to an accountable owner, a control and a route for intervention.
This reflection represents AVER Solutions’ views. Mark’s participation in AVER Signals does not imply endorsement of AVER Solutions products or services.
ABOUT AVER SIGNALS
AVER Signals, published by AVER Solutions, brings together perspectives from leaders across insurance, banking, technology and transformation, alongside clearly attributed company reflections. Each contribution explores a practical question: as AI reshapes our industry, what should leaders prepare for, and how can we use AI to work smarter, serve customers better and strengthen trust?
The aim is not to predict the future, but to help our industry prepare for it through clear, practical and responsible leadership perspectives.
Questions leaders should consider — AVER Solutions perspective
What should insurers measure when evaluating AI business value?
How can insurers move beyond AI pilots into practical implementation?
Where can AI help insurers work with complex information?
What governance should insurers establish before an AI agent acts?
How can insurers connect AI visibility with customer trust and business value?
