Raising the Price of a Lead Can Shrink the Market It Was Meant to Grow

Based on the research ofShi, "The Welfare Effects of Selling Leads in a Two-Sided Marketplace," Management Science, 2026

The Pay-Per-Lead Business Is Bigger and Blunter Than It Looks

Strip away the branding and these platforms run the same transaction. A homeowner describes a job for free. The platform turns that description into a "lead" and charges the contractor to see it, before any job, deposit, or guarantee of a sale exists. Thumbtack's own help center is explicit about this: "On Thumbtack, you pay for leads, new customers who reach out to you directly," and the price is set to "match how valuable the job is," moving with job type, size, and how many other pros are competing for it. Pros can even drag a slider to cap what they're willing to pay per lead in their own service categories. Bark runs the identical logic through a credit system: "we calculate the costs of contacting a customer based on the service, the value of the job, and the supply and demand in the area," with no commission on top. Google's Local Services Ads bill the same way, a contractor is charged only once a customer actually makes contact inside the ad's active schedule and business hours, per Google's own support documentation, with the per-lead price floating based on competition and profile quality.

This is a fee schedule with no natural ceiling baked in. Each platform sets prices to extract as much as the marginal lead is worth to a contractor, category by category, market by market. Modernize goes a step further and sells some leads exclusively rather than to multiple contractors at once, charging a premium for it. Porch blends a subscription with per-lead charges. The result, across every one of these businesses, is the same dial: a fee-per-lead that can be turned up whenever a customer segment looks valuable enough to bear it.

The Threshold Where a Fee Increase Stops Paying Off

Peng Shi's model, published in *Management Science*, formalizes what happens when a platform turns that dial. Raise the fee-per-lead for a given customer type, and, holding the number of quotes a customer receives fixed, those customers end up quoted lower prices, because providers who do buy the lead are effectively bidding against a higher sunk cost and compete harder for the job to make it worthwhile. That part of the story is good for the platform and, on its face, good for customers too.

But the "holding the number of quotes fixed" clause is the entire game. Push the fee past a threshold and providers start declining to buy the lead at all. Fewer providers means fewer quotes per customer, and fewer quotes is its own tax on customer surplus, independent of whatever happens to the price level on the quotes customers do get. The model shows these two effects pull in opposite directions, and past the threshold the volume effect dominates: customer surplus falls even though, conditional on getting a quote, each quote is cheaper. A platform watching only per-lead revenue and average quoted price can walk straight past this point without noticing, because both of those metrics can keep looking fine while the thing that actually matters to customers, the odds of getting matched at all, is quietly eroding.

The real-world evidence for providers actually hitting that exit threshold is not subtle. The FTC's 2023 order against HomeAdvisor, which required the company to pay up to $7.2 million in redress, rests on a record of exactly this kind of unraveling: contractors paying a $287.99 annual membership plus a per-lead charge, then discovering, per the FTC's findings, reported by TechCrunch, that leads were misrepresented in quality, source, and geographic fit, with many leads not matching the services a pro actually offered or the area they served. A contractor who concludes the leads he's buying are bad value doesn't negotiate the fee down; he stops buying leads in that category, exactly the dynamic Shi's model treats as the hinge on which customer welfare turns.

A fee that looks like pure extraction on a spreadsheet is, to the provider deciding whether to buy the next lead, simply a price, and prices have a stopping point.

Designing the Fee Instead of Just Raising It

None of the six platforms named in Shi's own paper currently runs option two or three in a visible, systematic way. Thumbtack's per-pro price slider is the closest analogue to letting the provider side reveal its own clearing price, but the platform still sets the menu of categories and the baseline estimates the slider moves around. Angi's redress funds after the FTC settlement were a quality fix, not a pricing-mechanism fix, they compensated for misrepresented leads rather than changing the fee structure that makes providers vulnerable to bad ones in the first place.

What the Model Implies for Platforms Still Raising Prices

The practical upshot for a platform operator is uncomfortable because it cuts against the obvious move. If a lead-pricing team sees average revenue per lead flat or declining in a category, the instinct is to raise the fee, the per-quote math still works for the leads that do get bought. Shi's result says to check a different number first: how close is this segment's current fee to the price where providers in that segment start walking away entirely. Categories with thin provider supply, high job value, or close substitute marketing channels, the same categories Modernize targets with exclusive, premium-priced leads, are the ones where that threshold sits closest to the current fee, and where raising it carries the most risk of trading a revenue bump for a shrinking pool of quotes.

Google's move toward automated "maximize leads" bidding and Performance Max pay-per-lead campaigns, now expanding into more categories, is itself a tacit acknowledgment that fixed per-category pricing leaves money and matches on the table in both directions. Letting price float with real-time supply and demand is a step toward the market-clearing logic Shi's paper formalizes, but a platform that optimizes that float purely for its own revenue, rather than for the 58.19%-type welfare floor, is still choosing extraction over the guarantee. The paper's deeper point is that the two goals are not identical, and a platform that conflates them will eventually find out the hard way, in a contractor segment that simply stops buying.