What 401(k) Plans Actually Pay: Fee Benchmarks From 631,893 Plans

The trouble with almost every 401(k) fee benchmark in circulation is that it comes from a survey. A consultant polls a few hundred plan sponsors, asks what they pay, and publishes the median. The sponsors who respond are the ones who have the number to hand, which tends to mean the ones who have recently benchmarked, which tends to mean the ones already paying less than average. The figure that results gets quoted for years afterward as though it described the market, when what it really described was the subset of the market that was paying attention.

There is a better source, and it has been sitting in public the whole time. Every plan covered by ERISA files an income and expense statement with the Department of Labor once a year, and those filings can be read in bulk rather than one at a time. We did that for the 2024 plan year — every defined-contribution plan with reported financials and at least one active participant, 631,893 of them — and what follows is what those plans reported paying out of their own assets. Not what a vendor quoted, and not what a sponsor remembered.

The headline finding is the least surprising one, which is reassuring rather than disappointing: scale works, and it works steeply. A plan with between 25 and 99 active participants reported a median cost of 42.4 basis points of assets. By 250 to 999 participants that has fallen to 34.1; by 1,000 to 4,999 it is 20.8; and above 5,000 participants the median plan pays 10.5. The curve is smooth and monotonic across every band from 25 participants upward — roughly a three-quarters reduction in cost for what is, in administrative terms, largely the same job performed at different scale.

The one band that breaks the pattern is the smallest, and it breaks it in a way that says more about disclosure than about pricing. Plans with fewer than 25 active participants report a median of 27.6 basis points and a twenty-fifth percentile of 1.5 — figures that would be remarkable if they were real. They are not. In the smallest plans the cost mostly does not run through the plan at all: the employer pays the administration out of corporate cash, or the entire fee load sits inside fund expense ratios where the Form 5500 has no visibility. That row measures how little gets disclosed, not how little gets paid, and it deserves to be read that way every time it turns up in a benchmarking conversation.

Basis points are the industry’s preferred unit, and they flatter the large end of the market for a reason that has nothing to do with pricing. Converted to dollars per active participant, the same data tells a much milder story: the largest plans pay $79 a head against $160 in the 100-to-249 band. That is a factor of two, where the basis-point comparison implied a factor of four. The difference is account balance. A large plan looks cheap partly because it is genuinely better priced and partly because its participants have more money in it, so an identical dollar fee divides into a bigger number. Any conversation with a sponsor conducted entirely in one unit is winnable by whoever chose the unit, which is worth knowing whichever side of the table you happen to be sitting on.

More useful than either median, though, is how little the medians constrain anything. Within the 250-to-999 participant band, the middle half of plans spans 13.4 to 65.1 basis points — a near five-fold range among plans of the same size, in the same year, doing substantially the same thing. The spread inside a single band is wider than the gap between the bands. That makes “a plan your size pays about 34 basis points” a statement with very little information in it, because a quarter of plans that size pay under 13 and a quarter pay over 65. Size sets a floor on what is achievable; it does not predict what any particular sponsor is actually paying, and nothing short of benchmarking that specific plan will.

Industry moves the number more than size does

Restricting the analysis to plans with 100 or more active participants, where disclosure is far more complete, and grouping by the sponsor’s industry produces a spread of 4.6 times from top to bottom — wider than the spread across the entire size distribution. Accommodation and Food Services sits at 66.0 basis points, Administrative and Support at 56.1, Retail Trade at 48.6. At the other end, Finance and Insurance pays 19.3 and Educational Services 14.4.

Read the two columns together, however, and the ranking inverts almost completely. Accommodation and Food Services has the highest cost in basis points of any sector in the study and the lowestcost per participant in it — $37 against Wholesale Trade’s $209, even though Wholesale looks cheap at 33.7 basis points. The reason is that sectors with high turnover and low account balances will look expensive on any asset-based measure no matter how well their plan is priced, simply because there are not many assets to spread a largely fixed administrative cost across. A restaurant group and a wholesale distributor of the same headcount can be paying their recordkeeper almost identical dollars and land two hundred percent apart on the industry table. Educational Services at the bottom is a genuine outlier for different reasons, mostly compositional: 403(b)-adjacent plans, heavy TIAA presence, large asset bases and long-tenured participants who have had decades to accumulate.

What the measure cannot see

Most fee studies bury their limitations, which is a mistake, because in this case the limitations all run in the same direction and that direction is worth stating plainly. The Form 5500 income and expense statement captures money that leaves the plan. It does not capture fees netted inside fund expense ratios, so if a participant holds a fund charging 60 basis points and 25 of those are rebated to the recordkeeper, the plan’s expense line may show nothing at all — the largest blind spot in the data, and the one that falls hardest on small plans. It does not capture costs the employer pays directly from corporate cash, which is a real cost of running a plan and simply invisible here. And it does not separate recordkeeping from advisory from audit; the measure is total non-benefit expense, undifferentiated.

Every one of those omissions pushes the reported figure down, and every one of them pushes it down furthest at the small end. The true cost curve by plan size is therefore steeper than the table above, not shallower. Small plans are worse off relative to large ones than these numbers show.

Larger plans file a Schedule C, which itemises what each individual service provider was paid and includes indirect compensation, and that is the honest way to see a whole fee load rather than the portion that happens to run through the plan’s own books. It is also the basis of our analysis of what actually predicts a provider change — which, as it turns out, is not fee level at all.