Pull up your most recent high-cost claimant report. You'll see a claimant ID, a diagnosis category, paid-to-date, probably a projection for the balance of the plan year. What you won't see is a column telling you whether that person still works for you.
That missing column costs money.
Self-funded plan reporting is built to answer questions about care. Which conditions drive spend. Which providers are out of network. Which specialty drugs are moving the pharmacy trend. It isn't built to answer questions about employment status. So the population on your plan with the most predictable claims profile gets spread across every category in the report, and the cost never gets traced back to the decision that created it.
The five levers every cost-containment list covers
Search for how to reduce self-funded health plan costs and you'll get a consistent answer.
Network strategy: direct contracting, reference-based pricing, narrower networks.
Pharmacy: carve out the PBM, audit the rebates, get specialty spend under management.
Stop-loss: shop the market, move the specific deductible, look at a captive.
Care management: high-cost claimant intervention, centers of excellence, second-opinion programs.
Plan design: raise deductibles, incentivize HDHP enrollment, modify the contribution strategy.
Those are real levers, and employers who work them well save money.
Notice what the list leaves out. Every item on it addresses care delivered to people who work for you. None of them touch the population generating claims on your plan that doesn't work for you anymore.
Why COBRA never makes the list
Four reasons, and none of them are that COBRA is cheap.
The first is the reporting problem. Your TPA can almost certainly segment paid claims by COBRA versus active status. Most employers have never asked, so the quarterly report doesn't include it, so the question never comes up in the renewal meeting. The data exists. It just isn't on the page anyone reads.
The second is that COBRA sits in the wrong file. It arrived in 1985 as a compliance requirement and 40 years later it's still owned by whoever owns compliance. Notices go out on time. Elections get processed. Premiums get collected and remitted. Every one of those is a yes-or-no audit question, and when the answer is yes, the line item passes review and nobody asks what it cost.
The third is that the standard levers are shaped by who sells them. Network strategy has vendors. PBM carve-outs have vendors. Stop-loss has carriers, MGUs, and captive managers, all of whom publish material on how to think about stop-loss. COBRA has vendors too, but they sell administration, and administration is priced per participant per month. Nobody in that supply chain has a commercial reason to publish an article about reducing COBRA enrollment, so the article doesn't get written and the lever doesn't get named.
The fourth is scale. COBRA participants are a small share of covered lives at most employers, and nothing that small earns its own slide in a renewal deck. But the share of enrollment is the wrong way to size this. Sun Life reports that 88 percent of self-funded employers hit at least one stop-loss claim in a benefit year, and that million-dollar claims rose 46 percent in frequency between 2022 and 2026. When a single claimant can move your renewal, the size of the population matters far less than who's in it.
What hidden costs exist in self-funded plans?
The expensive ones usually aren't hidden. They're categorized in a way that separates the cost from the decision that caused it.
Start with the most common misunderstanding, which comes up constantly in conversations with benefits teams: leadership often believes COBRA is financially separate from the plan. It isn't. You collect the COBRA premium, and you still pay the claims. Someone paying $1,000 a month can generate many multiples of that in claims, and every dollar of it lands in your loss fund in real time.
Then look at who elects. KFF found annual spending for COBRA enrollees runs about twice that of other large-group enrollees, $11,695 against $6,144. The reason isn't mysterious. The same research shows COBRA enrollees skew older, with 33 percent aged 55 or over against 21 percent of other enrollees, report more chronic conditions, and are more than twice as likely to describe their health as poor.
People who expect to use coverage keep it. Someone mid-treatment, on a specialty prescription, or facing a scheduled procedure looks at switching plans and sees a reset deductible, so they stay on the plan they know at full premium. Healthier people shop, join a spouse's plan, or take a marketplace subsidy.
That sorting is adverse selection, and it's the most reliable thing about COBRA. The claims aren't bad luck. They're the predictable output of a default that rewards your highest utilizers for staying.
It also isn't only about layoffs. A dependent aging off at 26, a divorce, an employee becoming Medicare-entitled, a reduction in hours that drops someone below benefits eligibility: each triggers its own election, some for 36 months rather than 18. That last one barely registers anywhere, because the person is still active in your HRIS. Nobody counts them as an exit.
The cost arrives a plan year late
This is what makes COBRA hard to see even once you know to look for it.
Someone terminates in March. They elect in April, retroactive to their termination date. Claims land across the following 12 to 18 months, and those claims feed the experience period your stop-loss carrier uses to price the next renewal.
By the time the number shows up, it reads as a renewal increase driven by a bad claims year. The reduction that produced it happened three quarters earlier, approved by a different set of people, working from a severance model that treated the COBRA subsidy as a fixed per-person cost and never modeled the claims tail behind it.
Run it against a 100-person reduction. The severance model has a single COBRA line: three months of subsidy for anyone who elects. Thirty people elect, the weighted premium is about $1,366, and the line comes to roughly $123,000. Budgeted, approved, closed.
Those thirty stay on the plan an average of six months. At KFF's spending rate for COBRA enrollees, that's about $175,000 in claims across the window. More than the subsidy, and it appeared nowhere in the model, because the model was built to price a benefit rather than a risk.
That's the quiet version, where nobody gets seriously ill. Let one of the thirty turn into a $200,000 claimant and the average stops being the relevant number.
Two budgets, one decision, and nothing connecting them.
Four questions to ask before your next renewal
What share of our paid claims came from COBRA participants last plan year?
Ask your TPA to run it. If they say they can't, ask what's actually blocking it.
How many of our current high-cost claimants are on COBRA?
One name on that list changes the math for the entire conversation.
How many people did we terminate last year, and what was our election rate?
Election rate is the leading indicator. Claims are the lagging one. You can act on the first.
What did we assume about COBRA in our last stop-loss renewal?
If the answer is nothing, your carrier priced that risk and you didn't.
None of these require a new vendor, a new data feed, or a project plan. They require asking for a segment that already exists in someone's system.
COBRA is a cost-containment lever, not a compliance task
Every other lever on the standard list works on claims that are already in motion. Network strategy reprices care that's being delivered. Care management intervenes on someone who's already sick. Stop-loss transfers risk you've already accepted.
The COBRA lever works earlier than any of them. The 60-day election window is a real decision point, and most departing employees make that decision with a packet of paperwork and no help. The default runs toward the most expensive option on the table for both sides, because it's the only one anybody explains.
That's the part an employer can change. When someone gets an actual comparison of COBRA against marketplace and Medicare options priced for their own situation, and a licensed, non-commissioned agent to walk them through it, a meaningful share land somewhere that fits them better and costs your plan nothing. The people who genuinely should stay on COBRA still stay, and they should. When's agents are salaried rather than commissioned, so that recommendation carries no incentive either way.
The rest is a population you were carrying because nobody asked them a question at the right moment.
None of that shows up on the plan report. Ask for the column.
