Operational Playbook: AI must move from automating cost controls to improving MLR and protecting margins

This HFS Operational Playbook is for CIOs of US health plans using AI to manage population risk, improve medical loss ratio, and protect margins.

This playbook is for CIOs of US health plans who must enable the technologies, data infrastructure, and AI capabilities required to improve medical cost performance while supporting provider and member experience. It helps them embrace the key tenets of Services-as-Software™ and leverage AI to reduce costs and operational risks.

When executed with discipline, this approach enables health plans to materially reduce avoidable hospitalizations, improve chronic condition control, and bend the medical cost curve while preserving compliance and risk posture.

HFS Operational Playbooks are practical guides to solving key enterprise challenges that consume significant costs, time, and resources. The playbook provides enterprise leaders a realistic roadmap with specific “to-dos” to address their everyday challenges so they can clear mental and financial space to deliver next-level value.

The problem: Health plans have risk signals but no infrastructure to act on them before the cost is incurred

For health plan CIOs, cost pressure now centers on a single metric: medical loss ratio (MLR). As medical and pharmacy costs outpace premium growth, every unmanaged dollar flows straight into it. The pressure is already visible (see Exhibit 1). GLP-1s drove nearly half of the 2024 increase in commercial drug spend, and most large insurers reported higher MLRs through 2025, with Centene reaching 91.9%.

Exhibit 1: The cost curve is a risk curve, with drug-cost pressure concentrating in MLR

Sources: IFEBP/SHRM, 2025; Evernorth, 2025; KFF, 2026; Becker’s, 2026. Analysis: HFS Research, 2026

Health plans have responded with the levers they know best: utilization management, prior authorization, and care management. These decide what gets approved and paid, but they never reach the underlying risk: the chronic disease and specialty-therapy exposure that drives cost in the first place. The result is more administrative friction without a durable bend in trend, because the insight that could prevent the rising cost remains trapped in analytics dashboards and quarterly reviews, disconnected from the claims, care management, and member-outreach systems where intervention actually happens. The gap is not analytic capability; it is the absence of an operating layer that turns a risk signal into a triggered, owned, and measured action inside existing workflows.

Plans already hold the signals that flag a member heading toward a high-cost event. What they lack is the infrastructure to act on those signals, not after it is locked in.

This playbook introduces a Services-as-Software approach to turn those signals into timely action using existing systems and workflows without a large upfront transformation program. It enables three shifts:

• From reactive, visit-based intervention to continuous risk detection that reduces
  late-stage cost exposure
• From effort-based measurement to
  telemetry-driven outcomes linked to per member per month (PMPM) and MLR impact
• From ad hoc care management to codified services that improve consistency and cost control

Playbook benefits: Measurable impact on MLR, cost, and operational efficiency

Implementing population risk management lowers medical costs that affect MLR, reduces the operational burden of managing high-friction cost controls, and highlights the impact of each intervention, enabling health plans to invest in effective solutions and discontinue those that do not work.

Financial impact

• Reduced PMPM cost through earlier identification of rising risk members
• Improved MLR stability through smoother year-over-year cost trends and reduced tail-risk
• Lower catastrophic claims exposure driven by unmanaged chronic and metabolic disease

Operational impact

• Reduced reliance on high-friction transactional cost controls as primary cost levers
• Lower provider abrasion and administrative burden
• Improved efficiency of care management and utilization operations

Outcomes visibility

• Improved visibility over interventions delivering measurable cost and utilization impact
• Better management of high-cost chronic conditions, including diabetes, cardiovascular disease, and renal disease
• Measurable improvements in utilization and adherence metrics linked to cost reduction

The opportunity cost of continuing to rely on reactive, transactional cost controls is increasingly unsustainable in a high-trend, margin-constrained environment.

The solution: Manage population risk to improve MLR, not just transactions

Rather than launching another transformation program or buying more standalone tools, health plans should operationalize population risk management as a service. This involves using AI to detect rising risk, trigger intervention, and measure the resulting financial outcomes in PMPM and MLR, all within the systems they already run (see Exhibit 2):

Exhibit 2: From reactive transaction control to continuous population-risk management

Source: HFS Research, 2026

This is where Services-as-Software can earn its place. Rather than relying on large transformation programs, it helps health plans operationalize AI for earlier risk detection, consistent intervention, and measurable financial outcomes. And because population risk management spans clinical, operational, and financial functions, it serves as a common operating model that aligns stronger alignment among business leaders, typically market presidents, line-of-business heads, and clinical, operations, and financial stakeholders. More importantly, the CIO’s job shifts from leading technology transformation to supporting business-led cost control strategies by enabling AI-driven risk detection, scalable intervention infrastructure, and telemetry-based outcome measurement.

But technology is only the starting point. Models, dashboards, and risk scores alone do not deliver value. Real value is only fully realized through measurable financial outcomes, including PMPM cost reduction and improved MLR performance. The real shift is moving away from investing in standalone capabilities to operationalizing measurable financial and utilization impact. Because population risk management directly impacts cost, MLR, and regulatory compliance, the execution must be sequenced, practical, and aligned to business priorities (see Exhibit 3).

Exhibit 3: Five sequenced steps turn population risk management into measurable PMPM and MLR impact

Source: HFS Research, 2026

The playbook: Operationalize AI to improve MLR and protect margins

When a new operating paradigm is being adopted, execution discipline matters more than ambition. The following sections underline the key tenets of each step in the playbook to ensure population risk management delivers measurable outcomes rather than remaining an analytical exercise.

Step 1: Enable data foundation for cost and risk visibility

Step 2: Detect early cost and utilization risk using AI-enabled signals

Step 3: Prioritize high-cost and rising-risk cohorts

Step 4: Enable closed-loop intervention workflows

Step 5: Govern on MLR and financial outcomes

Execution timeline: Target twelve months for a visible MLR impact or reassess your execution

Health plan leaders must avoid analysis paralysis by adopting a least-resistant execution path focused on reducing cost and improving MLR outcomes. Initial data readiness and AI-driven risk detection must be completed within the first 8–12 weeks, immediately followed by rapid cohort prioritization and workflow integration (see Exhibit 4). Measurable improvements in PMPM cost, utilization, and MLR performance should be visible within one or two quarters. If material progress is not achieved within a year of execution, it must be reassessed.

Exhibit 4: The full five-step sequence fits one 12-month cycle, ending in MLR governance and scale

Source: HFS Research, 2026

The risk: Getting comfortable with the promise of AI

As AI becomes embedded in population risk management, the biggest threats to sustained MLR improvement are behavioral, not technical. Once the platform works, three patterns quietly erode the gains:

• Entrenched thinking and execution: Treating AI as a point solution rather than a new operating model will limit impact
• Outcome complacency: Outcome-based approaches can breed stagnation if KPIs are not continuously raised and validated.
• Governance gaps: Risk management requires explicit ownership of data rights, auditability, bias monitoring, and remediation mechanisms tied to outcomes.

None of these risks is about technology failing, but about leaders becoming comfortable. AI can flag the risk, but only disciplined execution and clear ownership convert detection into sustained MLR improvement.

The Bottom Line: For CIOs, the MLR win comes from stepping in before the cost is ever incurred.

Health plans have spent a decade automating the transactional layer: prior authorization, claims, documentation, and call centers. Those gains are real and now fully priced into the market. The next basis points of MLR performance will not come from automating more transactions, but from intervening before cost is incurred.

CIOs who shift AI investment from transaction automation to population-risk infrastructure, covering continuous detection, closed-loop intervention, and PMPM-linked governance, will convert AI into an MLR-defense asset. Those who keep scaling administrative automation will stabilize the cost of operations while medical costs continue to erode margins.

Check out our other perspectives

• Health plans must anchor value in health to survive and thrive in the BBB era
• Payers need a bold strategy to address declining margins and systemic pressures
• Stop funding sickness and invest in prevention that pays off