Every year, a benefits team somewhere sits around a table, opens a thick actuarial report, and tries to divine the future from claims that are already 18 months old. Someone points to a spike in oncology spend, someone else worries about a new hire with a rare condition, and the stop-loss carrier's quote feels like a bet against your own population. The decision to self-fund, or not, hinges on a snapshot that was stale the moment it was printed. We've been running the same playbook since the 1990s, and it's starting to look less like prudence and more like negligence.
The technology to flip this script exists right now. It's just buried under procurement hurdles, outdated carrier data agreements, and the comfortable inertia of 'this is how we've always done it.' I've seen a handful of employers quietly build a continuous monitoring capability, and what they're doing should reset expectations for every plan sponsor with a fiduciary pulse.
The Stale-Snapshot Problem
Traditional self-funding feasibility studies grab 12-24 months of medical and pharmacy claims, scrub for high-cost outliers, layer on trend assumptions, and project a range of possible costs under self-funding versus a fully insured premium. The output is a point-in-time guess. It can't see a sudden spike in GLP-1 prescriptions, which pushed total GLP-1 spend up roughly 50% in 2025 and now accounts for about a fifth of all prescription drug costs. It can't anticipate a local hospital merger that shifts facility fees overnight. It doesn't know your mid-year plan design changes will funnel more members toward specialty drugs.
Worse, the data itself is mushy. Because claims run out slowly, the most recent three to six months are educated estimates rather than hard counts. So you're making a multi-million-dollar stop-loss attachment decision on incomplete, rearview information. That is a fragile position for an ERISA fiduciary, and a hard one to defend under a DOL audit or in court. The Supreme Court has read the statutory duty of prudence as continuing over time: in Tibble v. Edison and again in Hughes v. Northwestern, the Court held that a fiduciary must keep monitoring what it has selected and remove what becomes imprudent.
What a Living Feasibility Engine Actually Looks Like
The fix is a shift from an annual event to a constant capability, powered by the same APIs and data infrastructure that already run modern payroll and HR platforms.
Stop-Loss Decisions on a Dial
When your TPA or carrier streams adjudicated claims data into a secure environment (ideally with a 30-60 day lag at most), you can start stress-testing stop-loss attachment levels every month instead of once a year at renewal. Imagine a dashboard flagging that, based on current run rate, a $200,000 specific deductible would save you more than $250,000 given the emerging claim clusters. You can have that conversation with your stop-loss carrier mid-year, or at least arm yourself with data that prevents them from lowballing you at renewal. Stop-loss carriers increasingly laser high-cost claimants at renewal, assigning one person a higher deductible or exclusion. When your own monthly trends show the underwriter's model is stale, you have leverage to push for a no-new-laser provision instead of absorbing the surprise.
Catching Risk Before It Catches You
This isn't just about stop-loss. When you combine claims, biometric screenings, wellness engagement, and even workers' comp data into a privacy-compliant lake, machine learning models can spot trouble early. A surge in diagnostic codes tied to musculoskeletal issues might reveal an ergonomic problem in a new distribution center. A sudden cluster of new enrollees on high-cost specialty meds signals you need a targeted case management push before costs spiral. The feasibility of self-funding then stops being a yes-or-no question and becomes a conditional yes, depending on whether you act on these signals. The study becomes a management tool instead of a one-time gate.
Modeling Real-Time 'What Ifs'
Whenever you think about layering on a point solution, switching PBMs, or steering to a narrow network, traditional feasibility offers zero help. It's backward-looking. With a continuous data stream, you can overlay the proposed change onto the current population's claims and see the impact immediately. What would reference-based pricing have saved last month? How would that diabetes management app change trend if you launched it now? You become an agile operator, testing ideas against live data rather than waiting a year to see what happened.
The Tech Stack and Why It Usually Stumbles
Under the hood, this needs more than a vendor promising a miracle. It's about three interconnected capabilities:
- API-Driven Data Feeds: CMS now requires many payers to offer FHIR-based Patient Access and Payer-to-Payer APIs, with a January 1, 2027 compliance date for the payer-to-payer exchange. Employer access to raw, plan-level data still varies by carrier and TPA, so treat it as a contract term. Dashboards won't cut it; you need the source to do your own modeling.
- Data Lakehouse Architecture: Think Snowflake, Databricks, or even a well-architected cloud warehouse that can marry claims, enrollment, payroll, and wellness data without creating a privacy nightmare. De-identification and role-based access are non-negotiable; only limited data sets under a BAA should see the light of detailed modeling.
- Automated Quality Checks: Real-time data is messy. Duplicates, coding errors, and orphaned records creep in. You need anomaly detection, unsupervised machine learning that flags when something looks off, so you don't base fiduciary decisions on junk data. Batch studies miss coding drift because they see a single frozen snapshot; a continuous pipeline compares each new feed against what came before and flags the divergence.
Federal Rules Already Mandate Some of This Data
Some of this data infrastructure is no longer optional. The Consolidated Appropriations Act of 2021 requires plans and issuers to report prescription drug and health care spending data to CMS each year through the RxDC reporting process, due June 1 for the prior calendar year. A self-insured plan sponsor can delegate the filing to a TPA or carrier, but the sponsor stays responsible for it. That annual submission establishes a baseline, and it proves your vendors already assemble the prescription and spending data you need. Use that as a concrete reason to ask for the same feed monthly instead of once a year.
The same statute runs in the opposite direction as well. Its gag-clause attestation requires plans to confirm each year that their service agreements don't block access to cost and quality data, and the No Surprises Act's transparency rules treat claims and pricing information as something the sponsor is entitled to see rather than a courtesy the carrier extends. When a TPA says a monthly feed is impossible, ask whether they are already pulling that data for their own RxDC filing. If the answer is yes, the monthly feed is a delivery question, and delivery questions get resolved in contract negotiations.
Fiduciary Protection Is the Point
The ERISA obligation to act as a prudent expert isn't satisfied by a one-time check-the-box study. Courts increasingly expect sponsors to monitor and manage the plan's financial health proactively. A documented, continuous process of reviewing claims trends, stop-loss adequacy, and risk forecasts creates a fiduciary record that's hard to challenge. When your benefits committee minutes show that every quarter you reviewed real-time data, analyzed emerging risks, and made course corrections, you're no longer an easy target. This capability improves the math and strengthens the legal posture.
How to Start Without Boiling the Ocean
You don't need to rip out your entire infrastructure. Begin by asking your TPA what data they can deliver monthly and in what format. Most will push back, but your contract renewal conversation is leverage. Even a limited monthly feed that you manually load into a basic analytics tool is a huge leap beyond the annual actuarial report. Pick a few metrics, say top 10 claimant cost trends, drug spend by therapeutic class, and specific stop-loss burner analysis, and track them month over month. Present those trends to your benefits committee in a simple dashboard. Then layer on more data sources as you build confidence. The goal is to create a rhythm where the plan's financial reality is never more than a few weeks out of view.
The self-funding study of yesteryear is a relic. What matters now is whether you're wired to see trouble before it arrives, and to act before the renewal letter lands. That's the feasibility demonstration that actually protects your people and your bottom line.
This article is for general information only and is not legal, tax, or medical advice. Employers should consult their own advisors.