The Other Customer

An intermediary earns a buyer’s trust. The trust aggregates demand. Then sellers notice.

What sellers notice is usually that reaching the intermediary’s buyers directly would cost more than paying the intermediary to do it. That gap is the whole story, and it is wider than most people expect.

Fact. By the early 1990s, US airlines were paying travel agents roughly $7.5 billion a year in commissions, an amount comparable to what they were spending on fuel. At the same time almost no traveler paid an agent anything.

Fact. A 2011 House Ways and Means report found that royalty payments from for-profit partners made up nearly 46 percent of one large membership organization’s revenue in 2009, while member dues were about 17 percent. Between 2002 and 2009 dues income rose $60 million. Income from business relationships, primarily with insurance companies, rose $417 million.

Fact. ADP publishes an accountant referral program paying 25 percent of a referred client’s first-year payroll billings at one to four referrals, 50 percent at five to nine, and 75 percent at ten or more, plus a 10 percent annual residual for four years. No accountant could charge a client that much for the recommendation itself.

The second customer is frequently the richer customer. Intermediaries do not usually decide to serve it. They grow into it, because the buyer-funded revenue stops keeping pace with the seller-funded revenue and the ratio changes over years without any single decision anyone could point to.

The question is what happens next, and the historical record answers it more precisely than the usual arguments about conflicts of interest do.

The useful question is not whether the seller pays. Nearly every intermediary takes seller money somewhere. The useful question is:

Does the intermediary make more money if the buyer chooses Seller A rather than Seller B?

Compensation variance runs along a spectrum, and the closer it gets to the individual recommendation, the stronger the steering incentive.

That is one axis. On its own it does not predict outcomes, and this is where the simple version of the thesis breaks.

The second axis is what the buyer believes. Variance became a trust failure specifically where the buyer understood the selection to be independent, unbiased, free, exhaustive, or made on their behalf.

The dangerous configuration is both conditions together.

Where both appeared, the cohort produced steering, litigation, regulatory intervention, restructuring, or sustained public scrutiny. Where seller payment existed but could not move the recommendation, trust frequently held for decades. And where compensation varied enormously but the buyer had knowingly agreed to the arrangement in advance, the structure could also remain stable. Franchise vendor programs are the proof of that last point and they are discussed in Section 4.

Evidence-backed inference. Neither condition alone consistently produced failure in this cohort. The combination did.

Wirecutter

Fact. The site launched in September 2011 and generated $150 million in affiliate revenue in its first five years. The New York Times acquired it in October 2016 for a reported $30 million. Affiliate-driven revenue for the parent was estimated at roughly $40 million in FY2023, on reported gross merchandise value exceeding a billion dollars annually.

Seller money on essentially every recommendation, fourteen years, no credible steering accusation.

Fact. It accepts no advertising, specifically so advertisers cannot be perceived as influencing reviews. It returns or discards test products rather than keeping them. Affiliate relationships are disclosed on every page. The commercial team operates separately from editorial, with journalists unaware of commercial agreements in place.

Fact, and this is the load-bearing one. The revenue team stated in 2018 that retailer links are chosen based on convenience to the reader regardless of whether the retailer offers affiliate revenue at all, and that the site links to whichever retailer has the best price and delivery even where that reduces its own payout.

Fact, and a useful qualification. In 2019 a business-side employee at the parent company unilaterally edited a Wirecutter financial product article to insert affiliate links, prompting internal review. The firewall is a control that can be breached rather than a guarantee.

Fact. In 2021 the site added a subscription paywall. It now takes money from the buyer and the seller at the same time.

Healthcare group purchasing organizations

Fact. GPOs are funded by administrative fees from suppliers based on member purchasing volume, typically one to three percent, protected under a safe harbor at 42 CFR 1001.952(j). The safe harbor requires a written agreement with each member stating the GPO may receive vendor payments and specifying the amount or the method of calculation. Where the member is a healthcare provider, the GPO must disclose in writing at least annually the amount received from each vendor on that member’s purchases, and must disclose the same to HHS on request.

The fee is earned across the contracted portfolio rather than on which contracted supplier a member ultimately buys from. Disclosure is a condition of the arrangement rather than a courtesy. The structure has been operating since 1991.

Criticism has concentrated where the variance actually lives, which is the decision about which suppliers get contracts at all. GAO examined GPO contracting practices in 2003 and again in 2014, reporting manufacturer complaints that GPOs are slow to place products on contract and that fee levels make obtaining a contract difficult for some firms.

Franchise approved-vendor programs

This case matters because it disproves the tidy version of the thesis.

Fact. Under the FTC Franchise Rule, franchisors must disclose in Item 8 of the Franchise Disclosure Document any requirement that franchisees purchase from designated or approved suppliers, including whether the franchisor or its affiliates are approved suppliers and whether they are the only ones. The Amended Rule requires disclosure of rebate programs including amounts received, of any officer or owner interest in required suppliers, and of the process for approving alternate suppliers.

Here compensation variance is at a maximum, because the franchisor takes rebates on franchisee purchase volume. Control over buyer behavior is also at a maximum, because the purchase is contractually mandated. By the simple version of the thesis this should be unstable. It is not.

What makes it stable is that Condition 2 is absent. The franchisee agreed in advance, in writing, with the economics disclosed, as a term of a relationship they chose. Nobody told them the vendor selection was independent.

SMB referral programs, as short corroboration

An accountant can earn substantial revenue share from payroll referrals without producing observable trust problems, because the business owner retains real comparison ability in a market with published pricing and understands that a relationship exists. Conflict severity scales with how much the buyer is relying on the intermediary’s search instead of doing their own.

Mortgage lending is the cleanest evidence in the whole study. Disclosure regimes were in place for decades. The incentive persisted anyway. Regulators eventually stopped adding disclosure and changed the shape of the compensation.

Disclosure tells the buyer a conflict exists. It does not remove the incentive.

It also does not tell them whether this particular recommendation was affected by it, and it cannot, because they have no counterfactual.

But the franchise case blocks the stronger claim. Where the buyer knowingly agreed beforehand that the intermediary controls vendor selection, disclosure does legitimize a highly variable structure, because independence was never part of what was promised.

The precise finding: disclosure cannot rescue an incentive that contradicts the service the buyer believes they are receiving. Where the service was never described as independent, disclosure does most of the work.

A private example of the same logic. Fact. TrueCar charged dealers $299 per new vehicle and $399 per used vehicle sold through its platform while advertising “No Surprises... No hidden costs or surprise fees. Ever.” Litigation alleged the per-sale fee was undisclosed to consumers. In late 2017 it settled the California New Car Dealers Association suit by switching from per-sale compensation to a flat subscription. No regulator required that. The remedy was removing the variance.

Sellers were not all purchasing the same thing, and what they purchased predicts how much governance the arrangement required.

Seller willingness to pay generally rises as the intermediary gains more influence over what the buyer does next. So does the governance burden.

The franchise evidence adds the refinement. Control itself is not the instability. A franchisee can knowingly agree to required vendors. A health plan member does not knowingly agree to a formulary incentive steering them toward a more profitable drug. What separates those is whether the buyer consented to the selection architecture before they were inside it.

The more an intermediary’s economics change with the buyer’s selection, the more it behaves economically like a distribution channel rather than a neutral adviser, whatever it calls itself.

Seven questions, in rough order of how easily an outsider can check them.

Worth noting that regulators in three separate industries converged on remedies of the same shape. Reg Z prohibits compensation based on the terms of the transaction, not lender payment. The NAR settlement prohibits compensation offers on the MLS, not seller contribution. The GPO safe harbor protects a portfolio-level fee with per-vendor disclosure. TrueCar reached the same structure privately under litigation pressure. Marsh reached it voluntarily inside its own firm.

Hypothesis. That convergence may reflect regulators borrowing from one another rather than four independent discoveries. The private settlements and the voluntary internal split cut the other way. I would not claim more than that.