From Pilots to Performance: What McKinsey’s 2025 AI Report Means for Insurers

Article by Kevin Gaut, Chief Technology Officer at INSTANDA.

AI is everywhere, its promise dominating boardroom conversations and conference stages alike, but here’s the reality: impact is uneven, progress is patchy, and most insurers still grapple with the same stubborn foundations. McKinsey’s latest cross-sector analysis puts this into context: 88% of organizations report using AI regularly, yet only a third have managed to scale it across the enterprise, and fewer than half can tie any significant Earnings Before Interest and Taxes (EBIT) to those efforts.

The next phase isn't about chasing hype or adding shiny new pilots. As I shared at this year’s Insurance Innovators Summit, it's about ‘rewiring’ core business systems so that AI becomes a source of real, durable value.

The Agents are Arriving—But Foundations Are Shaky

Agentic AI, systems that plan and execute multi-step work, has made the leap from flashy promise to in-the-trenches trials. The buzz is palpable, but what truly matters is not if insurers use agents, but how they embed them into processes for maximum value.

Roughly 62% of McKinsey respondents are experimenting, and 23% are scaling agents somewhere in the enterprise, most often in IT and knowledge management. But breadth is limited: deployments tend to live in one or two functions. That pattern should feel familiar to insurers—complex, regulated environments demand safety, observability, and auditable handoffs.

Though here’s what I see time and again: deployments that struggle to escape the gravity of their legacy platforms (COBOL, anyone?). To use an analogy from my Back to the Future themed presentation, we can’t ride into the future on hoverboards if the tracks keep leading us back to the past.

Success isn’t going to come from a silver-bullet “do everything” agent. It will come from composable, embedded micro-journeys—those narrow but crucial workflows where AI assembles data, reasons about actions, and hands off smoothly to human expertise. Think: intake to quote triage in commercial lines; claims FNOL to settlement recommendations; portfolio repricing sprints where underwriting, actuarial and finance collaborate with the model.

What 6% of Leaders Do Differently

McKinsey’s “high performers”—about 6% of respondents—follow a consistent playbook. They set ambitions beyond efficiency, explicitly targeting growth (new revenue, better cross-sell) and innovation (new products, new operating models). They redesign workflows rather than wrapping AI around legacy systems, scale across more functions, invest heavily, and institutionalize human-in-the-loop, leadership ownership, agile product delivery, and strong data/tech foundations.

For insurers, that resonates. Many AI proofs-of-concept deliver speed or cost gains locally (document extraction, call summarization), yet stall before policy, billing, claims, and finance are truly connected. Leaders break through by treating AI as a business transformation, and not simply an IT enhancement.

Practical Lessons: Start Small, Build Momentum

Rather than trying to “boil the ocean,” my advice is: Don’t make it grand—make it real:

1. Start Small, Choose High-Value Use Cases. Choose specific, frequent workflows where time-to-decision and quality matter: broker submissions; straight-through small-commercial underwriting; subrogation identification; retention risk management. Use agents to coordinate retrieval, reasoning, and action across systems—and bake in checkpoints where underwriters or claims handlers can accept, edit, or reject. It’s about unlocking value where it matters most, so that adoption sticks.
2. Keep Humans in the Loop. High-performers know exactly when and where humans should step in to validate AI outputs. For insurers, this means setting role-based thresholds (e.g., payments over £X require review), using confidence bands to flag uncertainty, and making it easy to reverse or appeal decisions. This isn’t just risk control; it’s about building trust. When people feel empowered, not replaced, adoption takes off.
3. Understand and Address Data Limitations. Although great strides have been made with AI point solutions in areas like fraud detection and claims solutions, there’s a bigger question to consider: Do you have the right data for AI to have a transformative impact right across the insurance value chain? Do you know precisely why an underwriter decided to apply a loading in one case but not another? Or how decisions are being made across your portfolio? These aren’t just operational questions; they’re the foundation of making AI work in the real world. For AI to be effective and reliable, it must be tightly coupled with systems of record.
4. Measure for Growth, Not Just Cost. Enterprise EBIT impact remains rare because many programs chase efficiency alone. Add growth metrics to your scorecard: faster quote turnaround, bind ratio lift, upsell conversion, new product launch cycle time.
5. Put Leadership and Operating Models to Work. AI success starts at the top. Companies that thrive have a single leader accountable for AI value, regular reprioritization, and a cross-functional team that ships updates weekly, not quarterly.
6. Treat Model Risk as Product Risk. AI risks—inaccuracy, cybersecurity, and compliance—need the same rigor as product risks. Insurers should have controls in place from day one: prompt/tool inventories, input/output logging, IP filtering, red-teaming, model lineage, and auditable justifications.
   
   

The INSTANDA Perspective: Back to the (Pragmatic) Future

As a policy-administration platform, the most powerful thing we deliver for insurers isn’t a flashy, killer demo (though we can deliver those too!)—it’s supplying the plumbing and patterns that make AI safe, repeatable, and fast to scale with:

• A composable foundation: Enabling you to build AI-powered products and customer journeys and connect them to your existing systems, without disruption.
• Agent templates for insurance: ready-made embedded underwriting, product configuration, and policy management “AI Agents” that plug into product schemas, rating, and business rules.
• Data-driven architecture for AI readiness: Every piece of data generated through policy administration is structured, accessible, and ready for AI consumption. Data insights are surfaced directly within the platform to inform underwriting decisions, while real-time data access enables dynamic pricing, risk assessment, and customer engagement.
• Human-in-the-loop by default: review queues with rationale, confidence, and action history so experts remain in control while throughput climbs.
• Outcome dashboards that prove value: track cycle time, loss-adjustment expense, bind ratio, NPS, and revenue uplift per use case—because the board will ask.
• A control plane for risk & compliance: patterns for data residency, explainability snapshots, and per-LOB guardrails; a library of mitigations for inaccuracy and IP leakage.
  
  

Actionable Questions for Insurance Leaders

If top of your agenda is ‘rewiring’ for the Agentic AI era, you’ve got to start asking the right questions. Not the theoretical ones, but the kind that really get you moving. Here’s where to begin:

1. Where will an agent remove the most friction this quarter?
   Don’t overthink it—pick one micro-journey that’s been a pain point forever. Nail it down, ship it, and make sure you’ve got human-in-the-loop (HITL) controls baked in.
2. What growth metric will you tie to AI?
   Efficiency is great, but let’s aim higher. What’s one revenue KPI—conversion rates, cross-sell, you name it—that you can tie directly to AI? Put it on the scoreboard next to cost savings.
3. What must change in your operating model?
   AI is a team sport. Who’s owning the value? How often are you reprioritizing? Are you shipping updates weekly, or are you still stuck in quarterly cycles?
4. What risks will you mitigate on day one?
   Document your checkpoints, logs, and artifacts so you’re not scrambling when someone asks, “How does this actually work?”
   
   

The Inflection Point—Taking the First Step

McKinsey’s findings are clear: while most organizations are experimenting with AI, only a few have rewired their operations to realize value at scale. The leaders aren’t just automating tasks—they’re reimagining how work gets done, setting bold growth ambitions, and building the governance and operational frameworks to support sustained success.

What sets these leaders apart is their commitment to continually asking the right questions. They don’t stop at the first success—they keep identifying friction points, tying AI to growth metrics, refining their operating models, and proactively managing risks. This iterative approach ensures that AI evolves from a series of disconnected pilots into a powerful, scalable engine for performance and growth.