The Barbell: Why Consumer Finance Has Giants and Minnows and Little Else
The Barbell: Why Consumer Finance Has Giants and Minnows and Little Else
Consumer financial services has very large institutions and very small ones, and the space between them keeps thinning. The usual explanation is that scale wins because bigger portfolios are safer — and that explanation is wrong, though the conclusion it supports is right. Our correlation analysis showed loss volatility stops falling within a few thousand accounts, so a lender with five million customers has essentially the same loss stability as one with fifty thousand. Scale in consumer lending buys something else entirely: the ability to spread fixed costs and to accumulate data. Both are real and decisive. Neither is diversification, and the difference determines what survives at each end.
In this report
The wrong reason and the right one
The standard case for scale in lending runs through diversification: more borrowers, less risk. It's intuitive, it's how the industry talks about consolidation, and for consumer credit it's close to false.
Our correlation analysis worked the arithmetic. At a pairwise correlation of just 0.03, loss volatility falls from 3.09% at 100 accounts to 1.52% at 5,000, then to 1.49% at 50,000. Going from five thousand accounts to fifty thousand — a tenfold increase — reduced loss volatility by about 2%. Beyond that, nothing.
So what does scale actually buy?
| Claimed advantage | Real? | Why |
|---|---|---|
| Risk diversification | Largely not | Exhausted within a few thousand accounts |
| Fixed cost absorption | Yes, decisively | Compliance, technology, and permissions cost the same at any size |
| Data for modelling | Yes | Defaults are what models learn from, and they accumulate with volume |
| Funding cost | Partly | Access and rating effects, though not universal |
| Negotiating position | Yes | With vendors, networks, and counterparties |
| Behavioural prediction | Yes | The mechanism in our asymmetry analysis requires large observed populations |
Two things follow from getting the reason right. The advantages that are real don't diminish — fixed-cost absorption improves monotonically with size, unlike diversification which plateaus. And they're all about cost and information rather than about lending better, which means a small institution isn't disadvantaged at credit judgment. It's disadvantaged at paying for the apparatus around it.
Scale doesn't make a lender better at lending. It makes them able to afford the apparatus that lending now requires.
The fixed-cost stack
What a consumer lender has to have, roughly independent of how many customers it serves:
- A compliance program — policies, training, monitoring, testing, and the people to run them. The requirements in our governance framework apply to a lender with ten thousand accounts and one with ten million.
- Model risk management — validation, documentation, monitoring, and the expertise to do them.
- Technology — origination, servicing, payments, reporting, and security.
- Regulatory permissions, which take years and money to obtain per state.
- Fair lending analysis, requiring data infrastructure and specialist capability.
- Furnishing infrastructure and dispute handling, per our furnisher guide.
- Capital and liquidity arrangements.
- Fraud detection capability.
- Examination readiness — the ongoing cost of being supervisable.
None of these scales with volume in any meaningful way. A fair lending analysis costs roughly what it costs; a model validation costs roughly what it costs. Which produces the arithmetic in the next section, and which is the same fixed-cost-against-size structure our selection analysis identified at the level of an individual loan — operating at the level of an entire institution.
That parallel is worth stating explicitly. The selection analysis showed fixed origination costs make small loans unviable. This report shows fixed institutional costs make small-to-mid-sized lenders unviable. Same mechanism, two levels, and both squeeze from the same direction.
Where the middle gets squeezed
Put stylized numbers on a fixed-cost stack of $9 million annually — a figure meant to demonstrate the shape rather than to describe any institution.
| Accounts | Fixed cost per account | Position |
|---|---|---|
| 5,000 | $1,800 | Impossible at this stack — must operate differently |
| 50,000 | $180 | Severe pressure |
| 250,000 | $36 | Workable but thin |
| 1,000,000 | $9 | Comfortable |
| 5,000,000 | $1.80 | Negligible |
The 5,000-account row is the key to the whole structure. A very small institution cannot carry this stack and doesn't — it operates under different expectations, offers fewer products, holds fewer permissions, and substitutes judgment for modelling. It's a different business.
The 50,000-account row is where the trouble is. Too large to operate as a simple local institution, too small to absorb $180 per account. It has grown into the expectations without growing into the volume, and it's carrying a stack built for someone bigger.
Which is the barbell: viable at the bottom by avoiding the stack, viable at the top by spreading it, and squeezed in between by carrying it without the volume.
How the small end survives
The section that explains why the barbell isn't simply a slide toward concentration.
Small institutions persist because they run a genuinely different model, not a smaller version of the same one:
- Narrower product range. Fewer products means fewer systems, fewer permissions, and fewer compliance surfaces.
- Defined populations. A credit union serving one employer or community knows its members in ways that substitute for modelling — the local knowledge that our cold start analysis treats as expert judgment, held institutionally.
- Proportionate supervision. Regulatory expectations are calibrated to size and complexity, so a small institution's obligations are genuinely lighter.
- Relationship acquisition. Members arrive through affiliation rather than through the paid acquisition that dominates larger firms' cost base.
- Shared infrastructure. Cooperative arrangements, service bureaus, and shared networks let small institutions rent parts of the stack.
- Different objectives. Cooperative and community institutions optimize differently, which changes what counts as viable.
The last two are the load-bearing ones. Shared infrastructure is the small end's answer to fixed costs, and it's the same answer the middle reaches for — with a difference we'll come to.
And the honest limit: the small end survives in the niches, not in the mainstream. A small institution can serve its members well and cannot compete for a national consumer product against a firm spending $1.80 per account on compliance. The barbell's lower end is real and it's bounded.
The middle's three routes
Specialize down. Shrink the addressable market to something a lighter stack can serve — one product, one geography, one population. This works and it means accepting a smaller business, which is why it's chosen less often than it should be.
Partner. Rent what you can't afford to build — permissions from a chartered institution, servicing from a specialist, capital from an investor, technology from a platform. This is the route that has expanded most, and it explains something our chain analysis described without fully accounting for.
That report documented long intermediation chains and attributed them to specialization and permissions. The barbell supplies the driver: chains lengthened because mid-sized participants couldn't fund the fixed-cost stack alone, so they assembled it from pieces owned by others. Every partnership is a rented fixed cost, and the chain is what the rental agreement looks like from outside.
Which also explains why chains are longest in the products serving the most constrained consumers. Those products are served disproportionately by mid-sized and specialist firms — the participants least able to own the stack — while prime products are dominated by institutions large enough to be vertically integrated.
Sell. The most common outcome, and the reason the middle keeps thinning. Consolidation is rational for both parties: the acquirer spreads its stack over more accounts, the seller escapes a stack it can't fund.
The important point about all three: none is a failure of management. A well-run mid-sized institution faces the same arithmetic as a poorly-run one, and no amount of operational excellence changes $180 per account into $9.
What each new requirement does
A consequence worth stating carefully, because it's easily misread as an argument against regulation and isn't.
Every new requirement is a fixed cost, so every new requirement raises the minimum viable scale. A rule requiring an additional analysis, system, or control costs a large institution a negligible amount per account and a mid-sized one a meaningful amount. The rule may be entirely justified on its merits and still thin the middle.
The mechanism compounds:
- A requirement is added, appropriately, to address a real problem.
- Minimum viable scale rises.
- Institutions below the new threshold exit or sell.
- The market concentrates.
- Concentration raises concerns, prompting further requirements.
Two honest observations about this loop. It's not an argument for fewer rules — the problems being addressed are frequently real, and this desk's own analyses of dispute handling and fair lending argue for more rigour rather than less. It's an argument that the distributional consequence should be counted as a cost of the rule rather than discovered afterward as an unrelated trend.
And the design implication follows directly: requirements scaled to institution size preserve the middle where flat requirements don't. Proportionate supervision already does this at the small end. The gap is in the middle, where expectations are frequently applied as though the institution were large.
The same logic applies to the fragmentation our definitional analysis describes. Fifty state regimes is a fixed cost paid fifty times, and it falls on the middle hardest — which means definitional uncertainty and jurisdictional fragmentation are barbell forces as much as they are compliance problems.
What it means for consumers
The structure determines what's available:
- Products come from a giant or a specialist, with little in between. The mid-sized relationship lender that once served the gap between a large bank's cutoff and a specialty lender's pricing has largely gone.
- The gap is filled by the chain, with the accountability costs our chain analysis documents.
- Small institutions serve their members well and can't be reached by non-members, so the small end's advantages are unavailable to most people — which is why the credit union recommendation in our account guide comes with an eligibility check attached.
- Price dispersion persists, since the search costs in our search analysis mean few consumers compare across the ends of the barbell.
- Thin-file and small-dollar borrowers are served by the most intermediated structures, because those are the segments the middle used to serve.
The consumer-facing summary: the barbell doesn't reduce the number of products, it changes who provides them and through how many hands.
The strongest objections
"Technology is lowering the fixed-cost stack, so this reverses." The best objection and partly right. Cloud infrastructure, purchased models, and platform services genuinely reduce what a firm must build — which is the buy-side case in our vendor analysis. Two qualifications. Renting is the partnership route, not an escape from it — it converts fixed cost to variable cost while creating the dependency and the chain. And compliance and supervisory costs have not fallen, and they're the largest component. Technology has made the stack cheaper to assemble and not cheaper to be accountable for.
"Concentration reflects better firms winning." Partly, and the analysis doesn't deny that large firms are frequently good at what they do. But the mechanism described here operates on well-run and poorly-run institutions identically, which means exit from the middle is weak evidence about quality. A mid-sized lender with excellent credit performance faces the same per-account fixed cost as a mediocre one.
"The figures are asserted." Correct. The $9 million stack and the account thresholds are stylized, chosen to demonstrate a shape rather than to estimate any institution's costs, and credible estimates aren't publicly available. The result depends on the structure rather than the magnitude — at any fixed-cost level, cost per account falls hyperbolically with volume, and the squeeze appears somewhere. Where it appears is an empirical question this report doesn't answer.
Testable implications
- Exits and consolidations should concentrate in the middle of the size distribution rather than at the bottom, which is checkable against institution counts by asset band over time.
- New fixed-cost requirements should be followed by exits concentrated just below the new viable threshold.
- Surviving small institutions should hold narrower product ranges and fewer permissions than mid-sized ones, evidencing the different-business claim.
- Chain length should correlate inversely with participant size, since partnering is the middle's route.
- Proportionate supervision regimes should show thicker middles than flat ones, comparing across jurisdictions.
- Loss volatility should not differ materially between mid-sized and very large lenders at comparable portfolio composition — the direct test of the diversification claim, and the one that would confirm scale's advantage lies entirely elsewhere.
The sixth is the one that matters most for how the industry talks about itself. If loss stability is genuinely comparable across sizes, then every consolidation justified on risk grounds was justified on the wrong grounds — and the real case, which is about fixed costs and data, has different implications. It suggests the answer to a thin middle is shared infrastructure and proportionate requirements rather than more consolidation, since consolidation solves a cost problem by eliminating the firm that had it.
Frequently asked questions
Compliance, technology, model development, and permissions cost broadly the same at any size, so cost per customer falls steeply with volume. Scale also produces the data models need, which can't be bought.
Much less than assumed — the independent component is averaged away within a few thousand accounts, and what remains is driven by common factors that don't diminish with scale.
By not carrying the full stack — narrower products, defined populations where local knowledge substitutes for modelling, proportionate supervision, and shared infrastructure.
Three routes: specialize down, partner to rent capabilities, or sell. The second explains why so many consumer products now involve several firms.
Key takeaways
- Scale in consumer lending buys fixed-cost absorption and data, not risk diversification — which plateaus within a few thousand accounts.
- The compliance, technology, and permissions stack costs roughly the same at any size, producing $1,800 per account at 5,000 customers and $1.80 at 5 million.
- Small institutions survive by running a different business, not a smaller one — narrower products, defined populations, and shared infrastructure.
- The middle's partnership route explains long intermediation chains: every partnership is a rented fixed cost.
- Each new requirement raises minimum viable scale, which is a distributional cost worth counting rather than an argument for fewer rules.
- The mechanism operates identically on well-run and poorly-run institutions, so exit from the middle is weak evidence about quality.
This report presents an analytical framework and the authors' interpretation; it is not financial, regulatory, or investment advice. The cost figures and account thresholds used are stylized illustrations chosen to demonstrate the structure, not estimates of any institution's costs; credible public data on institutional fixed-cost stacks is not available, and the location of the squeeze in any real market is an empirical question not answered here.