High Fees Don’t Predict a 401(k) Switch. Here’s What Does.

The standard prospecting theory for retirement plans is that overpriced plans go to market. Find the sponsors paying well above the benchmark, show them the gap, win the plan. It is a clean story, it flatters everyone involved, and it is what most plan-intelligence products are built to surface — fee outliers, ranked, with a percentile beside each one.

We tested it directly. Across 152,704 defined-contribution plan-years spanning 2021 through 2023, we scored every plan on its cost percentile against genuine peers — same size band, same sector, same year — and then checked whether it changed its lead provider the following year. The pooled base rate across the whole population is 16.1 percent. If the theory holds, the expensive tail should switch noticeably more often than the cheap one.

It does not. The most expensive tenth of plans changed providers at 16.8 percent. The cheapest quarter changed at 16.5. In between, the second quartile sat at 15.9 and the third at 15.7, meaning the cheapest plans in the study switched marginally more often than the median ones. Against a 16.1 percent base rate the entire spread from cheapest to most expensive is about one percentage point, and it is not even monotonic.

With samples this large those differences are measurable, but measurable and useful are not the same thing. Knowing that a plan sits in the ninety-fifth percentile on cost moves your prior on whether it will change provider next year by well under a percentage point. As a variable for deciding who to call, it is worth close to nothing.

This is not an argument against fee analysis, and it would be easy to overread it as one. A plan paying eighty basis points where its peers pay thirty is a genuinely good conversation to walk into, and the spread within a single size band is enormous — wide enough that the outliers are real outliers rather than measurement noise. What the data argues against is fee level as a timing signal. Sponsors who are overpaying mostly carry on overpaying, year after year, and the ones who do move were not necessarily overpaying to begin with. Cost tells you what to say once you have a reason to call. It does not tell you when to call.

The obvious backup candidates fare no better. Plans whose assets fell more than ten percent changed provider at 15.7 percent; plans whose assets grew more than ten percent changed at 16.0. That is a null result to two significant figures. Headcount change produces the one weak-but-real effect in this group, and it is U-shaped rather than directional: plans shrinking by more than twenty percent switched at 17.2 percent and plans growing by more than twenty percent at 17.8, against 15.1 for plans that stayed roughly flat. Disruption in either direction moves plans, which is intuitive enough. But the whole effect is 2.7 percentage points wide, and the two ends of it call for completely opposite conversations — one is a company in trouble, the other is a company outgrowing its arrangements.

What does work is how new the relationship is

The cleanest signal in the data is also the one nobody talks about. Sort plans by how many consecutive years the current lead provider has held the relationship, and the switching rate falls sharply with age: 19.6 percent in the first year, 13.0 percent in the second, 11.9 percent by the third. A relationship in its first year is roughly 1.6 times more likely to end than one in its third.

Two mechanisms are probably at work and they point the same way. Some first-year relationships are failing conversions — the plan moved, the transition went badly, and it moves again within a year or two. Others reflect plans in an unsettled period generally, where a new provider is one of several things changing at once: a new CFO, an acquisition, a committee that has decided to review everything. Either way the practical reading is counterintuitive and worth sitting with, because it inverts how most territories get worked. The plan that just left someone else is the plan most likely to move again. The book worth your attention is not the fifteen-year incumbent relationship you have been patiently chipping at; it is the plan that changed hands last year.

That tenure figure is also a floor rather than a ceiling. Our filing history begins in 2021, so a provider that has served a plan continuously since 2010 registers a maximum measurable tenure of three years within this window. The third-year bucket therefore mixes genuinely three-year-old relationships with much older and far stickier ones, which almost certainly drags its 11.9 percent upward toward the average. The true gap between a new relationship and a long-settled one is wider than the table shows.

The steepest gradient is also the least trustworthy

One variable outruns everything else in the study, and it deserves to be presented alongside the reason not to believe it. Sorted by the number of service providers disclosed on a plan’s Schedule C, the switching rate climbs from 9.2 percent for single-vendor plans to 15.8 at two, 19.1 at three, 22.8 at four, 25.8 at five, and 28.3 percent for plans disclosing six or more. A complex plan appears to change lead provider three times as often as a simple one.

The problem is that the outcome being measured is “the highest-paid firm changed,” and that outcome is mechanically easier to trigger when there are more firms to rank. With a single vendor it can only happen if the vendor is genuinely replaced. With six, it also happens whenever the ranking reshuffles while every firm keeps its seat — a compliance project, an actuarial year, a one-off consulting engagement. A meaningful portion of that 28.3 percent is arithmetic rather than client loss. It is the same effect that makes defined-benefit plans look like they churn constantly when they barely churn at all.

Some residual signal is probably real. Complex plans have more decision-makers, more consultants with opinions, and more renewal events per year, all of which plausibly raise the odds of something actually changing. But the slope should be treated as an upper bound, and any prospecting model that leans on it will spend its time pointing at large, complicated, heavily advised plans that are not going anywhere.

Put the findings together and the honest summary is unglamorous. Cost percentile is a conversation, not a trigger — build the pitch with it once you have a reason to call, but do not use it to choose who to call. Relationship age is the closest thing to a timing variable in the data, and it is measurable directly from filing history without any modelling at all. Sharp headcount movement in either direction earns a look, at 17 to 18 percent against a 16.1 base, though it is small enough that it only really works combined with something else. And any signal that rewards roster complexity should be regarded with suspicion. None of this is a strong single predictor, which is the real conclusion: a 16.1 percent base rate with the best available signal moving it to around 20 percent is a useful edge, and it is not a crystal ball.