How Agentic AI is forcing P&C insurers to rethink BPO partnerships
The spreadsheet on a carrier COO's desk tells a new story. A BPO is billing $525K a month for claims support. A pilot agentic AI did the same work for $120K, 60% faster, with higher accuracy. The math is obvious. The hard part is deciding what to do with a 20-year relationship when the unit economics flip overnight.
Your strategic options
- Do nothing. Not viable in a competitive market.
- Give BPOs time to embed AI and offer hybrid pricing.
- Adopt AI from your core systems vendors to reduce BPO dependency.
- Partner with AI-native agentic firms for targeted processes.
- Build home-grown AI-human operations.
The right choice depends on economics, lock-in risk, operational impact, and talent. Entry- and mid-level roles will shift. Your talent plan has to shift with them.
Selective disruption: What moves to AI, what stays human
- Moving fast to AI: FNOL intake, simple endorsements, routine underwriting checks, document classification, data enrichment.
- Staying human (for now): Complex negotiations, multi-jurisdiction compliance, high-severity or litigation-prone claims, relationship-sensitive accounts.
Leaders aren't asking if AI will fit. They're deciding where AI is the primary operator and where people handle exceptions, nuance, and trust.
What executives want from BPOs
"I don't need cheaper labor for routine work," says one operations leader. "I need partners who design AI-human workflows and take responsibility for outcomes."
The FTE conversation is fading. The question is, "What's your AI strategy?" Too many proposals still pitch the labor model from five years ago.
Resistance and reality
- Contracts: Multi-year SLAs tied to legacy metrics and volume floors.
- Footprint: Offshore centers optimized for headcount, not orchestration.
- Talent: Thin bench of AI engineers, solution architects, and model risk expertise.
- Substance: Chatbots sold as AI transformation instead of true workflow reinvention with agentic execution.
The result: uneven progress. Big carriers push ahead with budgets and oversight. Smaller players struggle to resource both AI investment and vendor governance.
Strategic move: Bet on AI-native players (directly or via your BPO)
- Early access: Integrate agentic workflows before they're commoditized.
- Value shift: Compete on outcomes, not hourly rates.
- Clear signal: Shows a BPO is reinventing, not defending headcount.
If your partner is investing in AI-native capabilities, that's a strong indicator they'll still be relevant in three years.
The new value proposition for BPOs
- Exception handling: High-judgment edge cases and escalations.
- Quality assurance: Audit, drift detection, and regulatory controls.
- Strategic oversight: Trend analysis, playbook updates, and continuous improvement.
Keep the BPO. Redefine what you're buying.
A practical operating blueprint (90-180 days)
- Pick 2-3 processes with high volume and clear rules (e.g., FNOL triage, endorsements).
- Run a dual-track pilot: BPO-led hybrid vs. AI-native partner. Same data, same SLAs.
- Price on outcomes: Per-transaction fees with accuracy and cycle-time bonuses; penalties for rework.
- Stand up governance: Model risk review, prompt/version control, human-in-the-loop thresholds.
- Re-map talent: Upskill processors into exception specialists and QA analysts.
KPIs that actually matter
- Unit economics: Cost per transaction, rework cost per transaction.
- Speed: Cycle time, queue time, time-to-first-touch.
- Quality: Accuracy, exception rate, leakage, complaint rate.
- Control: Audit pass rate, model drift alerts, PII exposure incidents.
Vendor assessment checklist
- Architecture: Agent orchestration, tool-use, retrieval, and auditability.
- Data controls: PII redaction, data residency, SOC 2/ISO 27001, model access logs.
- Workflow design: Clear human-in-the-loop triggers and rollback paths.
- Retraining/programming: Turnaround time for playbook changes; cost to update prompts/policies.
- Pricing: Outcome-based with quality gates; no lock-in to opaque "AI fees."
- Exit options: Data portability, prompt/library IP rights, switch costs defined upfront.
Compliance and model risk
- Align controls with industry guidance such as the NIST AI Risk Management Framework.
- Map workflows to insurance regulatory expectations on fairness, transparency, and consumer protection. See the NAIC overview on AI.
- Document decision logic, guardrails, and audit trails at the task level, not just the model level.
Pricing model shift
- From: FTEs, volume bands, and seat-time.
- To: Outcome SLAs (accuracy, cycle time), exception thresholds, and shared savings for ongoing improvement.
Lock-in risk is now about data, prompts, and workflow IP-not cubicles. Write contracts accordingly.
Talent: Build the bench you'll need
- Upskill adjusters and processors into exception specialists, AI QA, and prompt/playbook owners.
- Create an internal academy for agentic workflow design and governance.
- For structured paths and certifications, see courses by role at Complete AI Training.
What good looks like in 5 years
- BPOs operate as AI-first partners with specialized human expertise.
- Insurers own playbooks, data, and oversight; partners supply orchestration and scale.
- Cost curves bend down while quality and control improve. Not everyone makes the transition.
The takeaway is simple. Efficiency is no longer cheaper labor. It's better design. Choose partners who can prove it-on your data, in your workflows, with your risk standards-and let the spreadsheet decide.
Your membership also unlocks:
