How Often Do Employers Actually Switch 401(k) Providers?

Ask how often employers change retirement plan providers and you will get answers anywhere between five and twenty percent a year, every one of them sourced to something real. The range is not sloppiness, and it is not a disagreement about data. It is that the word “switched” has at least two entirely defensible definitions, and applied to the same filings in the same years they produce answers that differ by more than a factor of two.

The filings themselves make the ambiguity easy to see. Plans above roughly a hundred participants file a Schedule C, which lists every service provider paid $5,000 or more during the year and exactly what each one received. That gives you a ranked roster for every plan, every year, and two ways to ask whether the plan changed hands. The first is to rank by compensation, take the firm at the top, and ask whether it differs from last year’s top firm. This is what most published switching rates approximate, because it is the measure you can compute without much effort. The second is stricter: take last year’s top firm and ask whether it appears anywhereon this year’s roster at all. If it has vanished entirely, the plan genuinely ended that relationship.

Run both across three consecutive year-pairs and the gap is consistent and large.

Roughly sixteen percent of plans changed which firm sat at the top of their provider roster. Only seven percent actually stopped working with the incumbent. The nine points in between are reshuffles — the same firms in a different order, usually because a project year moved the money around. An actuary who ran a plan-design study, a consultant engaged for a one-off compliance cleanup, an auditor whose fee happened to exceed the recordkeeper’s in a quiet year: all of these produce a change at the top of the roster while every existing relationship stays exactly where it was.

Both numbers are correct, and the distinction between them is not pedantic. They answer different questions and they support different decisions. If you want to know how many plans had something change at the top of the relationship this year — a reasonable definition of a plan worth a phone call — sixteen percent is the right figure. If you want to know how many incumbents actually lost a client, it is seven. The commercial consequence of conflating them is specific: a reshuffle is not an opening. A plan whose actuary out-earned its recordkeeper because of a de-risking project has not gone to market, is not unhappy, and will not appreciate being contacted as though it were.

What is striking across the three pairs is how little either number moves. The strict rate sits between 6.7 and 7.0 percent, the loose rate between 15.7 and 16.5, across years that contained very different market conditions. Whatever drives provider turnover in this market, it is not the cycle. Retirement plan relationships appear to turn over at something close to a constant rate, which is what you would expect from a process governed by contract renewal calendars and committee meeting schedules rather than by news, performance, or anything else that varies year to year.

Defined-benefit plans reshuffle constantly and change almost never

Splitting the most recent pair by plan type sharpens the point considerably. Welfare plans show the highest genuine turnover at 7.6 percent, which fits how that market actually works: health and benefits programmes get re-marketed on a far shorter cycle than retirement plans, and moving carriers is a routine annual exercise rather than a multi-quarter conversion project. Defined-contribution plans sit at 7.0 percent against a loose rate of 16.5.

Defined-benefit plans produce the widest gap in the entire dataset, and they produce it in the direction that should worry anyone building a prospecting list. Nearly a quarter of DB plans — 23.4 percent — changed their top-paid provider between 2023 and 2024, the highest rate of any plan type. Only 5.4 percent dropped the incumbent, the lowest rate of any plan type. The most churn-looking segment in the data is in fact the least churning one.

The explanation is roster size. A defined-benefit plan pays an actuary, a trustee, a custodian, an investment consultant and frequently an ERISA counsel, and which of those is largest in any given year depends entirely on whether there was a valuation cycle, a de-risking study, or a lump-sum window. The roster is stable; the ranking is volatile. Any switching signal built on “the largest provider changed” will fire almost continuously on DB plans and be wrong the substantial majority of the time — which is a good reason to be sceptical of any product that surfaces those plans as opportunities without saying which measure it used.

Plans do not disappear, they change hands

The other way to lose a plan is for it to stop existing, and at the large end of the market that essentially never happens. Among 2024 defined-contribution filings, 3.25 percent of plans with fewer than 25 participants were marked as a final return, falling to 2.06 percent between 25 and 99, 0.70 percent between 100 and 999, and 0.07 percent above a thousand. Seven plans in every ten thousand above the thousand-participant mark wound up, merged, or ceased operations.

At that end of the market the book is effectively closed. Every plan worth having is already someone else’s, no new ones are being created in any meaningful number, and the only route in is the seven percent event. The picture below a hundred participants is genuinely different — plans terminate and get created constantly down there, and a meaningful share of the apparently new small plans in any given year are the same employer reappearing under a new arrangement — but that segment is largely invisible to Schedule C in the first place.

A seven percent annual rate means a territory of four hundred plans produces about twenty-eight genuine provider changes a year, and nothing in the filings tells you in advance which twenty-eight. That is an argument for working the base rate rather than the anecdote, and for knowing which plan characteristics move it, since some segments run at nearly double the average. We tested the signals most commonly used for exactly that purpose in a companion piece on what actually predicts a switch, and the results eliminated most of them.