The AI Progression Roadmap

The question is no longer whether to adopt AI, but how fast it can be scaled without breaking things. Adoption is nearly universal; scaling is rare; and the measured payoff for crossing that gap is large. This page is the strategy in one sitting.

1 · Where the industry actually is

The binding constraints are not model quality. They are data readiness (the top barrier for 78% of insurers¹¹), change management (~two-thirds of the challenge, per BCG³⁸), and governance fragmentation (68% of insurers say controls exist but are fragmented; only 24% are fully confident in them, per Grant Thornton 2026¹⁰).

The maturity models agree on the arc (Gartner, KPMG "Enable–Embed–Evolve", Deloitte three horizons, Microsoft four stages³⁵): experiment → scale in one domain → horizontal platform → AI-shaped operating model. Consensus timing: GenAI copilots scaling now (2025–26), agentic workflows in core processes 2026–27, AI-native operating models emerging among leaders 2028+.⁵ The six-stage technology ladder maps the same progression from the capability side.

2 · What competitors have already done

Two implications. First, speed-to-quote is becoming the visible competitive weapon in E&S: agencies now rank real-time appetite information as the #1 factor in carrier selection (Ivans 2025: 29% of respondents, up from 12% in 2024²⁸), and wholesalers are building their own AI intake layers; submission flow will route to whoever responds fastest. Second, nothing on this list required inventing technology: the vendors and architectures are proven. The moat will be proprietary data, underwriting judgment encoded into workflows, and adoption speed.

3 · The roadmap: four gated phases

Each phase has explicit exit gates: governance, data, and measurement criteria that must be met before advancing. Skipping gates is how carriers end up in pilot purgatory (the 93% who never scale) or in an examiner's findings letter.

4 · The economics

• Budget shape: ~2–3% of IT budget in year one (data foundation + enterprise licenses + 1–2 vendor solutions; enterprise LLM agreements run ~$250k–$1M/yr at mid-size scale), scaling to 10–15% by year three. Context: carriers' IT spend averages ~4.5% of GWP⁴⁴; two-thirds of insurance CEOs plan to allocate 10–20% of budget to AI (KPMG³⁹).
• The 1:1 rule: for every $1 of technology, budget $1 for adoption: training, workflow redesign, change management (McKinsey⁷). Leaders invest in change management at ~3x the average rate (Capgemini⁶); the industry currently spends 72% on tech vs. 28% on adoption, an inversion to avoid.⁶
• Where the money comes back (measured, specialty-relevant⁴³): underwriter capacity (30–40% of commercial underwriter time is admin, per McKinsey⁷; Markel +113% productivity¹⁸), speed (time-to-quote reductions of 50–99% now common in reported deployments¹⁷²²), claims (20–30% LAE reduction potential, per BCG⁵; early litigation triage; leakage prevention³⁵), and portfolio steering (1–2.5 pts combined ratio, per BCG³⁴).
• Realistic timeline: 6–18 months to first production ROI; 18–36 months to enterprise-level P&L impact. 67% of insurance CEOs now expect returns in 1–3 years (KPMG 2025³⁹).

5 · Why most insurers fail, and the countermeasures

6 · Regulatory non-negotiables, sequenced with the roadmap

• Now (Phase 1): written AIS Program¹², AI inventory (including vendor-embedded AI), board-level oversight cadence, third-party AI due diligence. 25 states + D.C. expect this today¹³; examiners in 12 pilot states are testing the NAIC AI Systems Evaluation Tool (inventory, governance, high-risk-system detail, data lineage) through September 2026¹⁴; expect national adoption after the November 2026 NAIC meeting.
• Before AI touches selection or pricing (Phase 2 gate): documented outcomes-testing methodology, actuarial validation (ASOP 56), human-override authority. New York DFS Circular Letter 7 (2024)¹⁵ requires annual testing and actuarial validity; Colorado's regime (expanded to auto/health October 2025)¹⁶ is the strictest: build to Colorado and other states are satisfied.
• Before automated or consumer-facing decisions (Phase 3 gate): consumer disclosure templates, appeal process with human review, drift monitoring. Several states are moving to require human review of adverse claim decisions.
• E&S note: surplus-lines status exempts carriers from rate/form filing, not from AI governance: the bulletins apply to all licensed insurers.¹² Conversely, E&S filing freedom is exactly why specialty can deploy pricing-adjacent AI faster than admitted carriers; that is a strategic asset.
• Existing law already applies: unfair discrimination, unfair trade practices, and claims practices acts apply to AI-assisted decisions exactly as to human ones. "The model did it" is not a defense.

7 · What is asked of the CEO and board

1. Name the owner. One accountable executive (data/analytics or business-side, with CIO partnership). Joint data + business P&L ownership is the pattern that scales (BCG³⁸). AI owned by no one stays in purgatory.
2. Charter governance. A small cross-functional group (data science, legal/compliance, IT security, business sponsor) with a board reporting cadence. Deliverable #1: the written AIS Program.
3. Approve Phase 1: the budget envelope (~2–3% of IT spend⁴⁴³⁹), enterprise AI licensing, the data readiness assessment, and 2–3 assistive pilots with named business owners and 90-day metrics.
4. Set the posture: AI drafts, humans decide, until a use case passes its governance gate. This is both the regulatory expectation and the fastest credible path to scale.
5. Ask for the metrics: time-to-quote, hit ratio, underwriter capacity, claims cycle time, adoption, reviewed quarterly. 42% of insurers track nothing⁶; that is the difference between a program and a collection of pilots.

8 · The first 90 days