The Home-Services Bid War: Auction Theory Meets Emergency Demand

The home-services query is an auction, not a listing

The paid results above an organic map pack are not a rate card. Every time someone searches 'water heater repair near me', the platform runs a real-time auction among the advertisers targeting that query, orders them by a combination of bid and quality, and sells the positions. The pricing rule is the important part. In a generalized second-price (GSP) auction, the advertiser who wins a slot does not pay their own bid; they pay just enough to hold their position against the next advertiser below them. Edelman, Ostrovsky and Schwarz documented this as the mechanism behind essentially all keyword advertising, the system selling billions of dollars of keywords by ranking bidders and charging each the price of the bid beneath it.

For a home-services advertiser this reframes the whole cost question. You are not paying a fixed price for a click; you are paying a function of what your closest competitor was willing to pay for the same homeowner at the same moment. When the trade is fragmented across many local firms all chasing the same urgent job, the price you clear at is set by the second-most-aggressive bidder in your ZIP code, not by any list rate you could look up in advance.

Why emergency demand is the accelerant

Auction mechanics alone do not make a category expensive. Demand does. Home services carries a demand profile that pushes hard on the auction in three ways, and it is worth being precise that this section is analytical reasoning built on the established GSP mechanism above, not a cited measurement of any single vertical.

First, the demand is often non-deferrable. A burst pipe, a failed furnace in winter, a roof leaking into a bedroom: these are jobs the homeowner cannot postpone to comparison-shop for a week. Time-pressed buyers tend to act on the first credible option in front of them rather than exhaustively scanning alternatives, which concentrates value on the highest, most visible positions and makes those positions worth bidding up.

Second, the lifetime value of a won job is high relative to a click. When a single booked repair or install is worth hundreds or thousands of dollars, an advertiser can rationally tolerate a large cost per click and still profit, and every competent competitor reasons the same way. Rational high bids from many firms are exactly the input a second-price auction converts into high clearing prices.

Third, the demand recurs and is geographically dense. The same head terms surface every day across a metro, so the auction runs at high frequency with a stable, motivated set of bidders. That combination, a repeated auction against inelastic urgent demand with high per-job value, is the structural reason home-services keywords sit among the most contested in local search. What the mechanism does not hand you is a number: the specific cost-per-click level in your trade and market is a measured quantity, not a theoretical constant.

'Just bid higher' is naive advice

The most common instruction a home-services owner hears is to raise the max bid until the phone rings. The auction theory says this misreads the game. GSP looks like a Vickrey auction, where bidding your true value is the dominant strategy, but it does not behave like one. Edelman, Ostrovsky and Schwarz, with the companion equilibrium analysis in Varian's 'Position Auctions', showed that GSP is not incentive-compatible: it generally has no dominant-strategy equilibrium, and simply declaring your true willingness to pay is not an optimal strategy.

The practical translation is that there is no single correct bid you can compute from your own economics in isolation. Your best bid depends on what everyone else is doing, which position is actually worth paying for once the price of the rung above it is accounted for, and where the tradeoff between volume and cost per acquisition sits for your specific job mix. That is a modeling problem that shifts as competitors enter, pause, and re-price. It is precisely why paid search rewards continuous, strategist-led management over a set-and-forget max bid, not as a vendor preference but as the direct consequence of a documented game-theoretic property.

The road not taken, and what it reveals

There is a mechanism that is provably truthful. The Vickrey-Clarke-Groves (VCG) auction charges each bidder the cost their presence imposes on everyone else, which makes bidding your true valuation a dominant strategy and improves allocative efficiency under collusion. Vickrey, Clarke and Groves built the theory across three foundational papers, and by the textbook criteria VCG is the 'better' design.

The advertising industry adopted GSP anyway, largely for its simplicity and its favorable revenue properties for the auctioneer under many conditions. For an advertiser the lesson is not to lobby for a different auction; it is to accept that the surface you buy on was engineered with known, non-neutral incentives baked in. The auction is not a neutral meter. It is a designed market whose rules reward those who model it and quietly penalize those who treat it as a vending machine.

Where the money leaks: branded search and the incrementality gap

High clearing prices are only half the cost story. The other half is paying for clicks that would have converted anyway, and here the field-experiment literature is unusually direct.

In a large randomized experiment at eBay, Blake, Nosko and Tadelis found that paid search on branded and trademark keywords produced no measurable short-term incremental benefit; for non-brand terms, less-frequent and new users were positively influenced, while frequent users, whose purchases were unaffected by the ads, absorbed most of the spend and dragged average returns negative. For a home-services firm the direct analogy is brand-defense bidding: paying to appear when someone already searched your company name. Sometimes that is genuine protection against a competitor poaching your traffic; often it is spend on customers who were already walking through your door. The eBay result is a single-firm study, so treat it as a well-documented mechanism to test in your own account rather than a universal constant, but the mechanism, that ad exposure correlated with existing intent tends to look effective while adding little, is exactly what inflates a home-services bill.

The deeper reason platform-reported returns overstate reality is activity bias. Lewis, Rao and Reiley showed across three controlled experiments that people who happen to be browsing are also more likely to be searching and clicking, ad or no ad, so naive before-and-after or last-click measurement credits the ad for conversions it did not cause. This is why practitioner analyses report measured incremental ROAS running well below platform-reported ROAS, most severely on branded search. Those specific percentage ranges come from unaudited vendor case studies rather than academic replication, so we flag them as industry-reported and not something to publish as a fixed figure without a firm's own measured data behind it.

You cannot manage a bid war you refuse to measure

If the auction rewards modeling and the returns are systematically overstated, the antidote is causal measurement rather than dashboard faith. Two methods from the literature make this affordable at local scale.

Geo experiments, formalized by Vaver and Koehler, randomize non-overlapping geographic regions into ad and no-ad conditions to measure true causal lift without any individual-level tracking, and were explicitly designed to inform bidding, budgeting and campaign decisions. For a multi-location or multi-metro home-services operator this is the closest thing to a randomized controlled trial for advertising. Ghost ads, introduced by Johnson, Lewis and Nubbemeyer, record the counterfactual impressions a control group would have seen, letting an advertiser measure incrementality at a fraction of the cost of older public-service-announcement holdout tests; the method demonstrated a 17.2 percent lift in site visits and a 10.5 percent lift in purchases on a retargeting campaign while working natively with real-time ad delivery.

The reporting philosophy that follows is deliberately blunt. A single blended efficiency number, total revenue over total marketing spend, resists the attribution-gaming that channel-level ROAS invites, because it cannot be improved simply by moving credit between line items. None of this promises a specific outcome. It replaces a comforting platform metric with a causal one, which is the only kind of number worth managing a bid war against.

What the mechanism tells you, and what it cannot

The summary divides cleanly by certainty. The auction structure is established: GSP prices your click against your closest competitor, it has no truthful dominant strategy, and inelastic emergency demand with high per-job value is the structural reason home-services positions are contested and expensive. The field experiments are established too: branded-search incrementality is frequently near zero, activity bias inflates naive measurement, and geo experiments and ghost ads give you a real way to see through it.

What is not settled, and what no responsible article should assert as fact, is the exact cost per click, waste percentage, or incrementality gap for plumbing versus HVAC versus roofing in your specific market. Those are vertical figures that need a firm's own primary data before they can be published as anything more than a hypothesis. The theory tells you the shape of the problem with confidence. Only measurement tells you the size of yours.