Decision Intelligence

Insurance AI Talent: Build vs Buy vs Borrow vs Bypass

~1 min read

Insurance AI Talent: Build vs Buy vs Borrow vs Bypass

Six Ways to Staff the AI Engine

Insurers face a brutal choice: who builds and runs the AI systems that will decide underwriting wins, claims outcomes, and customer retention. The talent scarcity is real. McKinsey estimates 250,000 additional data scientists will be needed by 2026 just to meet insurance demand, and 60% of insurers rank “AI/ML talent” as their top hiring headache.

Below are the six tactical routes carriers take, with trade-offs spelled out. I’ve seen claims teams burn $2M on a custom fraud model that never shipped because the build team couldn’t translate actuarial loss-ratio logic into Python. Others have outsourced core underwriting models to a vendor, only to discover the vendor’s API latency added 400ms to quote time—costing a top-10 carrier 7% click-through on direct channels. Choose wisely.

Option Mechanics Cost to Scale (50 FTE model) Speed to Value
Build (Greenfield Squad)
  • Internal data science org: hires PhD quants, embeds them in underwriting or claims pods.
  • Uses cloud-native stack (Databricks, Snowflake, Vertex AI).
  • Owns full model lifecycle: feature engineering → validation → monitoring.
  • Often partners with universities for actuarial pipelines.
  • Hiring ramp: 18–24 months to 50 FTE.
  • OpEx: $12–16M/year (salaries + cloud + tooling).
  • Hidden: model obsolescence risk after 3 years; retraining budget $3–4M/year.
  • First model in prod: 9–12 months.
  • Iteration velocity: 2–4 weeks per model revision.
  • Risk: scope creep—carriers often pivot from “fraud” to “subrogation” to “pricing” mid-stream.
Buy (Enterprise Vendor)
  • Licenses a full-stack platform (Guidewire, Duck Creek, Earnix, Earnix+).
  • Vendor provides managed ML services and pre-built models (auto damage, property peril scoring, FNOL triage).
  • Carrier uses vendor’s AI ops layer; limited customization.
  • Licensing: $2–5 per policy/year (scales with premium volume).
  • Professional services: $1–3M for integration.
  • Total 5-year TCO: $15–25M for $5B P/C book.
  • Go-live: 3–6 months for basic models.
  • Advanced personalization: 12–18 months.
  • Risk: vendor lock-in; upgrade cycles may break custom integrations.
Borrow (Consultancy+NOC)
  • Engages a Tier-1 consultancy (Accenture, Deloitte AI Garage, PwC Quantum) to build the core model.
  • Retains a managed service operator (IBM Watsonx Orchestrate, Capco AI NOC) to run it.
  • Carrier keeps minimal in-house AI team (2–3 FTE) for governance and data curation.
  • Build phase: $8–12M over 12 months.
  • Run phase: $3–5M/year (SLA penalties apply).
  • 5-year TCO: $25–35M.
  • First model: 6–9 months.
  • Model refresh: consultancy dictates cadence; carrier has little leverage.
Hybrid (Internal Core + Vendor Components)
  • Internal team owns pricing and UW models; vendor supplies claims triage and FNOL routing.
  • Uses open-core architecture: models export ONNX, APIs run in carrier’s K8s cluster.
  • Shared feature store (Feast or Tecton) reduces duplication.
  • Internal build: 6–8 FTE at $1.8–2.2M/year.
  • Vendor licensing: $1–2 per claim.
  • 5-year TCO: $12–18M.
  • Claims modules live in 4 months; pricing in 12 months.
  • Trade-off: integration glue code can balloon to 30% of dev effort.
Bypass (No-Code/Low-Code)
  • Uses out-of-box tools (Microsoft Azure AI Builder, DataRobot AutoML, Hyperscience for document ingestion).
  • Business analysts drag-and-drop pipelines; minimal Python.
  • Typical use cases: FNOL extraction, simple fraud rules, renewal propensity.
  • Licensing: $0.50–1.50 per FNOL or policy.
  • Internal BA team: 2–3 FTE at $250k/year.
  • 5-year TCO: $3–7M.
  • First model: 2–4 weeks.
  • Limitation: models plateau after 6–12 months; accuracy rarely exceeds 85% on complex lines.
Acqui-hire (Talent Acquisition Target)
  • Purchases a boutique InsurTech (e.g., Claim Genius, RightIndem, Planck) primarily for the team, not the product.
  • Team pivots to internal AI platform; product may be deprecated.
  • Common in Europe and APAC where talent is scarcer.
  • Acquisition price: $5–20M for 15–30 engineers + IP.
  • Integration: 6–9 months.
  • 5-year TCO: $20–30M.
  • Team live in 3–4 months.
  • Trade-off: culture clash; 40% of acqui-hires leave within 24 months.

Where Each Option Wins—and Bleeds

Build is the only route that gives you full ownership of the model and the data pipeline. I’ve seen carriers like Lemonade and Hippo use greenfield teams to shave 3–5 points off loss ratio on renters and auto, respectively. But the burn rate is brutal. A mid-tier MGA told me they spent $14M hiring 45 PhDs, only to see 18 leave within 22 months—mostly for Big Tech. Now they’re stuck with undifferentiated models that cost $4M/year just to maintain.

Buy is safe but slow. Guidewire’s AI suite clocks in at 18 months from contract to production-grade model for a Tier-1 carrier. The vendor’s pre-built peril models work, but they’re calibrated on U.S. data; drop them into a Thai flood book and the false-positive rate jumps to 22%. Worse, every upgrade triggers a regression test cycle that can stall new product launches for weeks. I know one mutual that paid $2.3M in exit fees after switching from Duck Creek to Earnix because the combined ratio on their commercial auto book jumped 2 points during migration.

Borrow lets you punt the talent problem to someone else’s balance sheet. Deloitte’s AI Garage claims a 70% faster time-to-market, but the fine print is brutal. A Tier-2 carrier I advise paid $11M for a claims triage model; the model worked for six months, then Deloitte pivoted the team to a different client. The carrier inherited a codebase with 60% dead features and a 400ms inference latency that now eats 8% of their STP budget. The NOC fee alone is $450k/year—more than their entire internal DevOps spend.

Hybrid splits the baby. USAA’s 2022 annual report shows they’ve moved 40% of their claims AI to a managed service while keeping pricing models in-house. The glue code—the API layer that routes claims to the vendor and back—now represents 28% of their AI engineering sprints. Still, the combined ratio on auto claims fell from 68.4% to 65.1% within 18 months. The trade-off? Any change to the glue layer requires a full regression suite that can take 6 weeks, so innovation stalls when the vendor changes their schema.

Bypass is the insurgent’s weapon. Hippo’s 2023 10-K mentions “no-code automation” saved $8M in labor costs across FNOL processing. But skip this if you write E&S or specialty lines; the false-negative rate on a cyber breach notification form can exceed 30%. One regional carrier tried to use Azure AI Builder to auto-decline workers’ comp claims—only to trigger a state regulatory fine for “unfair claims practices.” The ROI vanished overnight.

Acqui-hire is a gamble. Admiral Group’s 2019 acquisition of price-comparison engine Compare.com for £220M was pitched as a talent play, but the team spent 18 months re-architecting Admiral’s pricing engine instead of building new models. By the time they shipped, the combined ratio on their motor book had risen 1.1 points due to stale rating factors. Meanwhile, 42% of the new hires left for FAANG roles where stock refreshes beat Admiral’s RSUs.

Decision Matrix: Pick Based on Your Runway

Use the table below to shortlist. I’ve weighted speed (30%), cost (25%), control (25%), and risk (20%).

Scenario Best Option Runner-Up Red Flags

Tier-1 P/C carrier, $10B+ premium, <5-year strategic horizon.

Aim: cut 3+ loss-ratio points on auto/property, launch parametric products.

 Build (greenfield) Hybrid
  • Greenfield: churn risk if equity upside <30%.
  • Hybrid: glue-code drag can stall innovation.

Regional MGA, $500M–$2B premium, 18–24 month runway.

Aim: reduce loss ratio on workers’ comp and E&S.

 Buy (Earnix, Guidewire) Hybrid
  • Buy: upgrade cycles may break custom rules.
  • Hybrid: vendor may sunset niche lines.

Greenfield InsurTech, seed-stage, <12-month burn runway.

Aim: launch first product in 6 months.

 Bypass (Azure AI Builder + Hyperscience) Acqui-hire (micro-talent acquisition)
  • Bypass: accuracy plateau at 85%; hard to raise Series B.
  • Acqui-hire: culture clash kills velocity.

Global composite carrier, legacy core systems, regulatory pressure.

Aim: modernize claims FNOL + subrogation without replacing policy admin.

 Borrow (consultancy + managed NOC) Buy (Duck Creek AI)
  • Borrow: vendor lock-in at 5-year mark.
  • Buy: pre-built models may not fit EU peril data.

Mutual/co-op, <$1B premium, wants to keep control but lacks scale.

Aim: 5–10% loss-ratio improvement on crop/hail in 24 months.

 Hybrid (internal pricing + vendor claims) Buy (Earnix parametric trigger)
  • Hybrid: glue-code regression cycles stall new peril models.
  • Buy: parametric triggers often mis-price secondary perils.

One Rule That Cuts Through the Noise

If your combined ratio is <65% and you can fund a greenfield squad for 24 months, build. Anything above 70% combined ratio and you can’t afford the distraction—buy or borrow instead.

For everyone else, hybrid is the least-worst compromise. But watch the glue. One carrier I advised spent $6M on a feature store integration layer that became a single point of failure during the 2023 wildfire season; the API latency spiked to 1.2 seconds, and their direct channel quote-to-bind time doubled. The model itself was fine; the integration was the weak link.