Do PEO Health Insurance Renewal Rates Stay Cheap or Revert to the Mean?

Do you want details on the method? Those will be in another paper called the PEO Renewal Transparency Index. That paper will explain what Form 5500 can and cannot show and how Benefitra proposes to rate PEO renewal disclosures.

What public data shows about PEO renewal rates

Why this question matters

Year one PEO health insurance quotes are easy to shop. Year two and year three renewals matter at least as much, and they are harder to compare. Brokers hear two competing stories.

Vendor story. "Our pool and buying power keep renewals below market."

Street story. "Cheap year one, painful true up later. Everyone meets in the middle."

Say the street story is mostly right. Then shopping PEOs on a single renewal headline is a trap. But if some PEOs really do stay cheap, you'd want to know which ones. Right now the market gives you no clean way to tell which it is. (Our guide to PEO honeymoon rates covers the year one side of this, and how to evaluate a PEO covers the rest of the buying decision.)

Terms. PEO = professional employer organization. FI = fully insured (the insurance carrier, not the employer, holds most claims risk). Master policy = one large policy the PEO sponsors that many small client companies join. Mean reversion, as used here = a PEO whose renewals run below market trend in early years later runs above trend, so its multi-year average drifts back toward market trend.

Public PEO renewal data that exists today

Hardly anyone publishes multiyear renewal numbers you can compare.

Direct (absence in sources searched). We couldn't find any multi PEO, multiyear, apples to apples health insurance renewal panel in the public domain. In September 2026 we looked into PEO websites and transparency reports, SEC 10-K filings and earnings calls for TriNet, Insperity and Paychex, as well as publications by NAPEO, material from state insurance departments, academic databases (PubMed) and DOL Form 5500 data.

Rippling's 6.5% versus a single carrier master figure

Different things measured, different numbers. So don't set Rippling's 6.5% beside a single carrier master figure (a Justworks Aetna master renewal of +12.9%*, say) as if the two were the same statistic.

* Where does the +12.9% come from? It is the Justworks Aetna master plan renewal, which a Justworks sales rep gave Benefitra verbally during the client's 2026 to 2027 renewal (late August to early September 2026). Justworks hasn't published that number, and it has not been independently verified. Brokers typically get renewal figures verbally. We use it only to show what a single carrier master number looks like, not as a measure of Justworks' book.

Inference. Truncating to the at or below base bucket softens the headline. By how much? Use only Rippling's own buckets (84% averaging 6.5%, another 5% "under 17%", 11% above 17%) and the full offered average can't be lower than about 7.8% (0.84 × 6.5 + 0.05 × 9 + 0.11 × 17). That's at least 1.3 percentage points higher than the headline, and it makes no assumptions at all. Plug in plausible bucket averages and the gap is 1.7 to 3.6 points (the survivorship section below shows this). Carrier blending is something else. We haven't measured how much that moves a headline.

Interpretation. It's important to give credit where credit is due, and Rippling does publish its definitions and footnotes. That gives more disclosure than most PEOs offer, and it's the reason this decoder is possible at all.

Company evidence from TriNet, Insperity and others

Two episodes at real companies suggest that cheap pricing can unwind, but they do not prove it.

Benefitra interpretation. So "cheap gets expensive" has real examples. There are two of them, from two companies, and they're different kinds of episode. What about the reverse path, where a pricey PEO gets cheaper over time? That's barely documented.

Why Form 5500 filings cannot settle mean reversion

Public data can't show mean reversion in PEO renewals, and it can't rule it out. Here's why, in plain words.

The biggest public dataset on PEO health insurance costs is the DOL Form 5500. For each PEO plan it gives total premium and the number of participants, and if you divide one by the other you get premium per participant. That is not a renewal rate. A lot of things besides renewal pricing influence it.

Who left (churn and survivorship). If clients with big increases go, the ones that stick around bring the average down.

Plan buy downs. Clients who move to a higher deductible pay less for less health coverage. The premium falls even when the rate rose.

New or younger groups joining. A wave of new, healthier clients lowers cost per participant without any renewal changing.

Mix changes. More dependents per employee, a different industry or region mix, or a different carrier mix all move the average.

Plan design changes across the book, like new networks or pharmacy terms, move it as well.

Benefitra's Form 5500 findings, 1999 to 2025

Benefitra analysis. We conducted our own long run analysis of Form 5500 filings from 1999 through 2025 [14]. In the pooled data, mean reversion isn't detectable. The pooled year to year pattern (lag 1 autocorrelation of −0.23 across 10 series and 76 yearly changes) is about what short samples throw up by chance (about −0.15. Permutation p = 0.35). Two single PEO tests did show reversion type slopes. One was AdvanStaff HR, which rests on 5 observations, and the other was ADP TotalSource, driven by a 2012 headcount jump. At 7 to 10 years the spread across PEOs also narrowed a little faster than averaging alone predicts, but only about two PEOs reach those windows. None of it is strong enough to generalize from. ADP TotalSource's premium per participant grew about 2.3 percentage points a year slower than the national benchmark over plan years 2005 to 2024. And about 1.3 to 1.5 points a year of that gap isn't explained by anything in the filings. Insperity? Its trend ran close to the national rate from 2006 to 2024. Over the 10 and 15 years ending in 2024, though, it ran below it. We will lay out the method behind these figures in the companion PEO Renewal Transparency Index paper.

Inference. That study also can't settle the question, for the same reasons mentioned before. Not detecting reversion in premium per participant doesn't mean that renewals don't revert. A lower long term trend does not show that a PEO renews at a low price. There could be other factors involved, such as plan buy downs. Or it could be a younger client base, or churn driven selection.

Why ADP TotalSource's low Form 5500 trend is probably not pricing

Benefitra analysis / inference. ADP TotalSource is the one large PEO with a Form 5500 trend that runs well below the market. What do the public numbers tell us here, and what don't they reveal?

What the gap would mean if it were price. Let's say ADP really had beaten the market by about 2.3 points per year for 19 years. A gap like that compounds. (1.021 ÷ 1.044)¹⁹ ≈ 0.65. So for the same coverage ADP would be about 35% cheaper than the market today, or if you flip it around, the market would cost about 53% more than ADP. Which start year you pick matters (the gap runs about 1.6 to 3.4 points a year depending on it). Still, on any of those windows it adds up to a large edge, roughly 24% to 39% cheaper (our arithmetic). Now, this reading assumes ADP's premium per participant was at market level in 2005. What if it started out above the market? In that case the slower trend could just be ADP converging toward the market and not a lasting price edge, and that would itself be a form of mean reversion.

Would we see it? A price edge that large, held that long, would be hard to miss. ADP would be winning and keeping small group clients far faster than the rest of the market. The public record doesn't show that. Average worksite employees grew 3% in fiscal 2025 and 2% in fiscal 2026. ADP puts PEO client retention at “approximately 6 years”, down from about 7 years in its fiscal 2016 and 2021 filings. That's roughly one client in six leaving each year (our arithmetic, since ADP publishes no annual PEO retention rate). Insperity reported attrition of about 17% in 2025 [6], so that's roughly similar. And ADP is the largest PEO, with about 14% of PEO worksite employees (our arithmetic). In fiscal 2026 its benefits pass throughs rose 7% while average worksite employees rose 2% [19]. Per employee, that's an increase of about 5% (our arithmetic, and the line covers all benefits, not only medical).

The plan structure. ADP TotalSource's health and welfare plan is almost entirely insured. Directly paid benefits were about 1.3% of the plan's benefits [14]. Its Form 5500 for plan year 2024 to 25 lists 39 insurance contracts. By persons covered, the biggest medical contracts are UnitedHealthcare (about 257,000) and Aetna (about 230,000), plus Blue Cross, Anthem and Kaiser plans in some states [22]. Those carriers price insured business from claims history and risk. So how would their ADP books stay about 35% below comparable business, for the same risk, year after year? It's hard to see. A lasting per participant gap more likely comes from plan buy downs (higher deductibles, cheaper tiers), a younger or healthier client mix, fewer family enrollments, how participants get counted (about 1 point of the gap by itself) and churn and selection.

Benefitra client case (published). One side by side quote. In a published Benefitra case study, an 11 employee Boston area law firm was days away from signing with ADP TotalSource. ADP's comparable quote was $155,256 for year one, health only, with a Harvard Pilgrim (HPHC) HMO. A second PEO quote, from Prestige PEO with a Cigna $3,500 deductible PPO, came in at $109,133. That's $46,123, or 29.7%, lower than the ADP quote (our arithmetic from the case study's figures). The plans weren't identical. Prestige's option was a national PPO with direct specialist access, while ADP's was an HMO. For the same group a broader PPO network would normally cost more, so the plan difference makes the gap more striking, not less. Still, this is one group and one quoting cycle, priced on that group's own census, and it isn't proof of anything about ADP's book. Past results do not predict future outcomes. And Benefitra was the broker on this case, so we have a commercial interest in it.

Benefitra first hand experience (broker quoting). When Benefitra has compared PEO quotes side by side for client groups, ADP TotalSource's medical rates haven't been materially lower than other PEOs' rates. That's hard to reconcile with a 19 year cost advantage in the Form 5500 trend. No one else can check these quotes (they are just the ones Benefitra has seen), so you should treat them as one broker's experience.

Inference. Put it all together and you have circumstantial evidence that ADP's low Form 5500 trend reflects mix, buy down and churn effects, and not a real price advantage. It is not proof, because Form 5500 can't separate these factors. And to be clear, nothing here suggests ADP has misreported anything. The filings just measure something other than renewal pricing.

Data that would settle the question

A per client renewal panel, across several PEOs and several years. That would settle it. For each client you'd see the renewal offered and the renewal accepted, with plan design held constant (or the change measured). Clients who left would be in it too, along with what they were offered. Results would be broken out by tenure (year 1, 2, 3+), carrier and group size, then compared with a market trend benchmark for the same years.

Nobody publishes this today, and that is exactly what the Persistence Tracker is designed to collect.

How survivorship bias lowers published renewal averages

Here's the idea in one line. If the clients who get the worst renewals leave, then any average built from the clients who are still there leaves out the worst renewals. The published number looks better than what the book was offered.

Which way survivorship bias pushes averages

Down. Here's the evidence.

Direct, TriNet. Retention fell to roughly 80% "with pricing cited most often," and where increases were outsized, attrition was higher. [3] Does the ~80% count clients or worksite employees? TriNet doesn't say.

Direct, Insperity. Client attrition was "approximately 17% in 2025." First on the 10-K's list of causes is "client non-renewal due to price or service factors." Another one is "our termination of the CSA resulting from the client's non-compliance or inability to make timely payments." [6] And on the call, management said "the profitability of the clients that terminated in our year-end transition was significantly lower than the clients that remain active." [5] That fits with less profitable groups leaving the book, no matter who started the exit.

Benefitra analysis, Insperity over 22 years. After years with a bigger cost increase, client retention was lower the next year (r = −0.42, p = 0.05. Spearman −0.54, p = 0.009. N = 22). [14] Same direction. How big the effect is, though, can't be measured from filings.

Direct, academic (Wickizer, Travis & Feldstein, 1998. 1985 to 1991 data). The study covered 95 insured groups. Groups that switched carriers had benefit payment growth of 18% a year, against 11% for the ones that stayed. But their observed premium growth was not significantly higher, "suggesting that employers decided to switch insurers before absorbing an increase in premiums." [11] It also notes that some switchers had below average increases. Keep in mind the data are old, and they don't come from PEOs.

Inference. When you average renewals accepted or paid by clients who stick around, that number is biased down compared with the renewals offered to everyone on the book.

Direct. There's one current running the other way (Urban Institute, 2017). Level funded products "are targeted to small employers that have relatively young and healthy workers." [13] Interpretation. When healthy groups leave, the risk pool they leave behind gets costlier, and future renewals go up. That doesn't cancel the point in time bias above.

Offered, accepted and paid renewals compared

First of all, know which one you are looking at.

Here's how the numbers cited in this article match up.

Rippling 6.5%. This one is mostly a truncation problem, not survivorship. The quotes are for the upcoming plan year, which means this year's would be leavers still get counted. The cohort filter is the catch (joined on or after 12/31/2024 and still a customer). It drops anyone who left at an earlier renewal.

Rippling "~99%" retention. According to its blog, Rippling "retained more than 99.5% of its PEO customers during the fall 2024 renewal season." [2] Clients or employees? All causes of exit? It doesn't say. If the figure is annual and all cause, then Rippling's survivorship bias is small (at most ~1% of clients × a 10 to 20 point gap ≈ 0.1 to 0.2 percentage points, inference). But some PEOs report retention monthly. Insperity, for one, reported client retention "averaging 99% per month" in Q4 2025 alongside about 83% annual retention. [5][6] So unless its basis is stated, a "99%" figure can't be compared with annual figures.

TriNet "average health fee increases of over 9%". That is revenue per coemployed worksite employee, which is a paid measure among stayers and new clients. It is exposed, especially in 2025, when retention dropped roughly 5 points.

Insperity "teens". It's pricing for 2026 renewals. That puts it closer to offered.

"X years in a row below market" claims. These are the most exposed. Every year, those who leave drop out before the next year gets measured.

Size of the bias, with the formula and inputs

Formula (mixture model). Let S be the average increase for clients who stayed, let a be the share of renewing clients who left because of the renewal, and let g be how many percentage points higher the increases of those who left were compared to those who stayed.

True average if leavers had stayed = (1 − a) × S + a × (S + g) = S + a × g

Bias = a × g (percentage points by which the survivor average understates)

Inputs

Results (mixture grid). Bias expressed in percentage points (a × g), with the true average if S = 9%.

The mixture grid goes from 0.2 to 2.4 percentage points and has 0.7 in the central cell.

Cross check using Rippling's own distribution. Start with Rippling's disclosed buckets. There's 84% at or below 9%, averaging 6.5%. Another 5% sits "under 17%" (we read that as 9 to 17%, which is an interpretation). And 11% is above 17%. On top of that we add assumed bucket averages, plus assumed exit rates that climb as the increase gets bigger. Then we compare the full offered average with a stayers only average.

The range for the bucket cross check is from 0.2 to 1.5 percentage points, and the central value is 0.5 (Pp = percentage points). The 50% exit rate in the High scenario is an assumption, not a measured figure.

Over several renewals (illustrative. Parameters to be published). We model a 3 renewal cohort. Every group gets a lasting "runs hot or cold" factor. We assume year one underpricing gets caught up at half the remaining gap per renewal. The bigger the increase, the higher the chance of leaving (a logistic curve). Market trend is set at 9%. We haven't published the simulation's parameters yet (spread of the hot/cold factor, logistic coefficients, cohort size, random seed), so treat this table as illustrative only. That's not included in the data files below either.

Look at the high case. By renewal 3 the stayers' average (8.5%) sits below the 9% market trend, while the full cohort still runs above it (9.7%). That's how survivorship could make a book look like it "stays cheap." The simulation builds catch up in as an assumption, so what it shows is the size of survivorship bias. It doesn't count as evidence either way for mean reversion.

Plain words finding.

Direction. Down. When you look at the average of renewals that clients who stayed accepted or paid, you end up understating what the book was offered. Offered measures, like Rippling's, have no exposure for the current cycle.

Size. Over one renewal cycle the bias comes out at roughly 0.2 to 1.5 percentage points in the bucket cross check, with a central value of 0.5. The mixture grid goes from 0.2 to 2.4, with 0.7 in the middle. That top end of the grid fits a catch up year like TriNet's 2025 (retention near 80%, with higher attrition where increases were outsized).

Rippling's headline. The truncation of Rippling's 6.5% to the at or below base bucket (at least 1.3 points, and 1.7 to 3.6 points in our scenarios) is larger than the survivorship effect in our central scenarios.

Survivorship and the reversion question

Attrition can hide catch up. Catch up renewals push the worst hit clients out. So the average of the ones remaining looks calmer than the catch up really was (inference, and TriNet 2025 is the live example).

Recent joiner cohorts are the least tested. Take a cohort that joined on or after 12/31/2024. It has had one or two renewals at most, so its expensive tail hasn't had time to leave. Research on employer insurance finds switchers get "a one-time discount of 9 percent in the year they switch" (Craig 2021, Massachusetts large group market, 2001 to 2018, not PEOs). [12] As evidence of lasting cheapness, year one to two numbers are the weakest there is.

Truncated tranches cut the tail on purpose. An average of only the at or below base bucket removes the same clients that attrition would later remove.

PEO initiated nonrenewals and carve outs. Say a PEO decides not to renew costly groups, or moves them to client sponsored or open market plans. The master plan's average improves, but nothing about the underlying cost has changed.

Healthy group exits cut the other way over time. When good risks leave the risk pool for level funding, the pool's future costs go up.

Questions and checks for your next PEO renewal

Why fully insured PEO plans can still differ

Direct. NAPEO told the Maryland Insurance Administration that "with respect to PEO-sponsored health coverage, the most typical model involves the use of a commercial large group insurance policy," and that "PEO-sponsored, fully insured group health plans are generally treated akin to single employer large group health plans." [9] So a PEO can only "contain cost" so far, compared with what a self funded plan with strong clinical management might do.

PEOs can still differ for a few years, though. How? It comes down to underwriting who gets into the risk pool, and to the industry and geography mix. Then there are carrier contract terms and pooling levels, plan design and contribution steering. A PEO can also move tougher groups to client sponsored or open market plan options. Or it can grow new, healthier cohorts that dilute the book.

Hypothesis (Benefitra). Without lasting selection or contract edges, a "cheap" book should drift back toward industry trend as its cohorts mature. That's a hypothesis to test. It isn't a conclusion, and we can't measure it for any PEO yet. Weighing a PEO against other ways to fund a plan? Our health plan funding simulator compares fully insured, PEO, Taft-Hartley, level funded, self funded and captive options side by side.

Can a PEO contain costs better than a fully insured plan?

Benefitra analysis. Hypothesis. Short answer? Usually not by much.

Same carriers, same networks. The most typical PEO health model is a fully insured commercial large group policy [9]. So who owns the network discounts and the claims management? Mostly the insurance carrier, not the PEO.

The edge is in rating, not in medical cost. A PEO's pricing advantage comes mostly from how its master plan is rated. That means large group pricing, with health based underwriting at the point a group joins. Which is selection, really. It isn't cost containment.

Across a whole book, costs should track the market trend plus the PEO's margin. The public data only partly fits this. Our Form 5500 study found Insperity close to the national trend over 2006 to 2024 (and below it over the 10 and 15 years ending in 2024). ADP TotalSource is a different story. It ran about 2.3 percentage points a year below the national benchmark over about 19 years [14]. The trouble is that premium per participant can't separate pricing from churn, plan buy downs and mix. So this doesn't prove a real edge, and it doesn't rule one out either.

For one group, a PEO renewal can swing more. Affordable Care Act (ACA) small group plans use adjusted community rating, where rates can vary only by family size, area, age and tobacco use. The group's health doesn't count [17]. Instead they move with the carrier's filed rates. In most states small group means 1 to 50 employees, and in California, New York and Vermont it's 1 to 100 [18]. Colorado moved back to 1 to 50 on January 1, 2026 (Colorado General Assembly, SB24-073). A PEO's large group master plan works differently. It can rerate a group based on its own claims history or demographics, which is why small businesses (especially under 50) can see bigger swings.

Real cost containment levers need a different structure. Reference based pricing, pharmacy benefit (PBM) carve outs, direct contracts with providers and care navigation. These generally need a self funded or level funded health plan. You don't usually find them inside fully insured master plans, and not every group can use them. New York, for example, requires 101 or more full time equivalent employees for self funded, level funded or self funded captive arrangements. Groups of 100 or fewer count as small group there, and their plan options are fully insured, PEO, Taft-Hartley and MEWA.

Checklist for your next PEO renewal meeting

Deadlines and paperwork are covered in how to evaluate a PEO proposal before your renewal deadline and when to switch PEO providers. And for how carriers set renewals in the wider market, there's our renewal prediction research.

Frequently asked questions

Methods, limitations and disclosures

Methods and data sources

Source search. We did the search in September 2026. It covered PEO transparency reports and blogs, SEC filings (10-K, 8-K) and earnings call transcripts for TriNet, Insperity and Paychex, NAPEO publications and state regulator material, PubMed, and DOL Form 5500 bulk data. A source made it in if it reported renewal pricing, retention or PEO health plan structure. Unless noted, every reference below was retrieved or rechecked on 2026-09-27 (ET).

Evidence labels. Direct information is stated in the cited source. Interpretation is Benefitra's reading of a source. Inference is something Benefitra derived from sourced facts. A hypothesis is a claim we want to test. An assumption is an input we chose, not a published number. Anecdotal means unverified reports.

Survivorship calculation. You will find the mixture formula (bias = a × g) and all the inputs in the survivorship section above. Each number in the grid and in the cross check can be worked out by hand with the help of the tables. Data files are available on request [16] (see Data availability). The 3 renewal simulation is illustrative, and its parameters are to be published.

Form 5500 study. This is Benefitra's analysis of DOL Form 5500 bulk data (1999 to 2025 filing years) and SEC 10-K filings for PEO sponsored health plans. It uses clean run rules and permutation tests [14].

Limitations

Public sources only. The company figures come from vendor reports and earnings calls. None of that is audited renewal data.

The evidence mostly runs one way. We did find some cases of cheap pricing unwinding. Few sources, though, look for the reverse path.

Client retention is not worksite employee retention, and neither is health plan retention. Some clients move medical off the master policy but stay on for payroll and HR. Client retention numbers don't reflect this.

Not every exit is about price. The price driven share of exits (our 25 to 60%) is an assumption.

Leavers' offered increases are not observed. We anchored the g values to Wickizer (1998) and TriNet's remarks. These values are not measured.

Vendor retention claims are undefined. There's no published method behind Rippling's "99.5%".

The +12.9% Justworks figure is a verbal, unverified report, and we use it only as an illustration.

The academic studies are dated or not PEO specific (Wickizer. 1985 to 1991. Craig. Massachusetts large group).

Web searches were limited. Reddit blocked automated access. So did the Justworks Help Center, for automated rechecks on 2026-09-27.

How to cite

Newland, S. (2026). Do PEO Health Insurance Renewal Rates Stay Cheap or Revert to the Mean? (Version 1.0.6). Benefitra. benefitra.com/peo-renewal-mean-reversion/. Accessed .

License. © 2026 Benefitra. Text and Benefitra derived tables are licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may share and adapt with credit to Benefitra and a link to this page. Third party data (CMS, AHRQ, KFF, vendor figures) remains under its owners’ terms. The Funding Simulator software is proprietary and not licensed for reuse.

Data availability

You can check the survivorship numbers yourself. Every value rederives from the inputs and formulas in the survivorship section. Do you need the input table, the bias grid or the Rippling cross check as data files? Ask the corrections contact listed below.

We didn't include the 3 renewal simulation, since its parameters haven't been published yet. If you want the Form 5500 extracts behind the Benefitra study, they're available on request through the same corrections contact.

Changelog and errata

Corrections contact. info@benefitra.com. Every correction gets logged right here, along with its date, what was wrong, what changed and which numbers moved.

References