Your CPC Is Supposed to Go Up

There is a conversation we have with almost every new client inside the first ninety days. We take the bidding off a shallow event, a button click or a page view, and point it at the thing that funds the business. Within a few weeks the cost per click climbs. The cost per conversion on the dashboard climbs with it. Someone on the client side gets nervous, because the old numbers looked cleaner. Then we look at pipeline, and cost per opportunity has fallen.

This piece lays out why that happens. The short version is that the price of a click is set by other advertisers' valuations of the same user, so when you start bidding for the users worth the most, you meet the competitors who already knew that. The long version covers the economics, the market data, and what it looked like inside our own accounts this year.

What the bidding engine buys

Smart Bidding is a machine that buys whatever outcome you name, as much of it as your budget allows, at whatever average cost your target permits. Google says so in its own documentation. Target CPA "optimizes for conversion volume irrespective of value," and the guidance for value-based strategies is to judge them on "the conversion value your campaign is generating," not on cost per conversion (Google Ads Help).

Hand the machine a shallow event and you have handed it an abundant target. A click on "Book a demo" happens constantly, from students, competitors, job seekers, and people who misread the ad. The machine finds the cheapest ones, fills your account with them, and your cost per conversion looks excellent.

Point the same machine at a closed deal and the target becomes rare. Out of a hundred clicks a handful become opportunities and one or two close. The engine stops buying the cheap volume that padded the count, so clicks fall. It starts competing for the specific users who look like buyers, and those users are the ones every serious advertiser in your category is also bidding on.

The theory

Google's search auction is a generalized second-price auction. Hal Varian, Google's chief economist, worked out its equilibrium in a 2007 paper, and the result that matters here is simple. "The price each agent must pay per click is determined by the bid of the agent below him in the ranking" (Varian, 2007, *International Journal of Industrial Organization*). In equilibrium, advertisers with higher value per click take the higher positions and pay strictly more per click. Google's help center says the same thing in plainer words. You pay only "what's minimally required to clear the Ad Rank thresholds and beat the Ad Rank of the competitor immediately below you" (Google Ads Help).

Your CPC measures what the next advertiser thinks that click is worth. When you optimize to a button click, the engine sends you into auctions for users whose next-best bidder values them at very little, because the users are worth very little. When you optimize to revenue, it sends you into auctions where the next-best bidder is a competitor who has already connected their CRM and knows this user is a buyer. Varian put it precisely in a 2009 follow-up, writing that "it is advertiser s + 1's value for those clicks that is relevant since that is the bid that the advertiser in slot s must beat" (Varian, 2009, *American Economic Review*).

A rising CPC after a bidding change is therefore evidence that you have moved into inventory where informed competitors are present. If the CPC does not move, one of two things is true. Either you were already competing in those auctions, which is common in enterprise categories where clicks cost two or three digits, or the engine has not found any new auctions to enter, which means the new signal is too thin to learn from.

There is a second-order effect that George Akerlof described in 1970 for used cars, and it applies to clicks without much adjustment. When buyers cannot tell good from bad, "the bad cars sell at the same price as good cars," and the bad drive out the good (Akerlof, 1970, *Quarterly Journal of Economics*). Ad auctions have partially solved that problem. The advertisers with CRM data attached to their bids can tell good from bad, and they bid accordingly. What remains cheap is the inventory those advertisers have declined. A cheap click is cheap because the advertisers who know what that user is worth have already decided not to pay for it.

Why the cheap volume was never yours

This matters for CPC because the cheap clicks a shallow target buys are disproportionately people who were coming anyway, and the evidence for that comes from eBay. Blake, Nosko, and Tadelis ran a field experiment in which eBay turned paid search off in randomized markets and measured what happened to sales. Conventional regression on the observational data showed a return of "over 1,400%." The experiment showed "a ROI of −63%, with a 95% confidence interval of [−124%, −3%]." The authors' explanation is the point. In their words, "most paid search traffic and attributed sales are high volume, frequent purchasers" who would have bought anyway (Blake, Nosko & Tadelis, 2015, *Econometrica*).

That is the population a shallow proxy event recruits. People who were coming regardless click the button, the machine credits the click, and the dashboard reports a cheap conversion that added nothing. A study of 15 Facebook experiments by Gordon, Zettelmeyer, Bhargava, and Chapsky found the same failure across methods, with "in 50% of our studies, the estimated percentage increase in purchase outcomes is off by a factor of three" (Gordon et al., 2019, *Marketing Science*). Lewis and Rao, working with 25 display advertising experiments at major US retailers and brokerages, concluded that "selection bias, due to the targeted nature of advertising, is a crippling concern for widely employed observational methods" (Lewis & Rao, 2015, *Quarterly Journal of Economics*).

Google's own researchers reached a related conclusion about their auction. Hohnhold, O'Brien, and Tang found that "the short-term effect is not always predictive of the long-term effect," and the finding led to "a conceptual change to our search ads auction that further increased the importance of ads quality, and a 50% reduction of the ad load on Google's mobile search interface" (Hohnhold, O'Brien & Tang, 2015, KDD).

The arithmetic

Round hypothetical figures, to make the shape legible. These are not benchmarks from any account.

Spend $20,000 optimizing to a button click. You buy 10,000 clicks at $2 and collect 1,000 CTA events at $20 apiece. Five percent of those are real opportunities and 20 percent of the opportunities close. That is 10 deals at $2,000 each.

Spend the same $20,000 optimizing to closed-won. CPC rises to $5, so you buy 4,000 clicks. You collect 300 CTA events at about $67 each, which looks like a disaster beside the old $20. But 20 percent of these become opportunities and 30 percent of those close. That is 18 deals at about $1,100 each.

CPC up 150 percent. Cost per CTA up more than 200 percent. Cost per customer down 45 percent on the same budget.

What it looked like in our accounts

These are two anonymized client accounts with real numbers, from the past year.

Client A. When we took over the account earlier this year, the primary conversion the bidding engine was learning from was a pricing-page view, an event it could buy for almost nothing. We replaced it with a set of actions that together describe the funnel: a product signup and a demo request as the two ways a buyer raises a hand, and two Salesforce stages (Engaged and Qualified) uploaded back into Google Ads with dollar values attached, so the engine could learn which hand-raises turned into pipeline. In June we weighted a demo request at 100 times a signup and moved the campaigns to Maximize Conversion Value.

CPC went up 2.6 times. Clicks fell 88 percent. Dashboard conversions fell 83 percent. Cost per qualified opportunity fell 29 percent, and July's figure will improve as the 30-day-plus stage lag matures.

Before the June change, when a signup and a demo request each counted as one conversion, the engine had learned to buy the cheapest version of each: demo requests from personal email addresses and signups that never reached a Salesforce stage. Marilyn Strathern's formulation of Goodhart's law applies, that "when a measure becomes a target, it ceases to be a good measure" (Strathern, 1997, *European Review*). Putting dollar values on the Salesforce stages is what pushed the engine out of that inventory.

Client B. The account bids to demo requests, with a conversion value attached to each lead based on the size of the prospect. From September 2025 through April 2026 the blended CPC sat between $7 and $10. In May we pushed the value model hard toward the largest opportunities, with a 25x value multiplier on the top tier.

CPC went up 2.8 times, and peaked at $37 a click in the third week of June. Conversion rate did not triple with it, and the raw count of qualified form submissions was lower than the January to April run rate. The sales team's own tally is the number that mattered. Before the change, no top-tier prospect had ever been attributed to paid search. In the ten weeks after, seven came through. Cutting the multiplier to 2.5x in July brought CPC back near $20 without losing them; two top-tier prospects arrived the same week.

What the market data says

The whole market has been paying more per click for a decade, and the cost of a lead has not followed. LocaliQ's benchmark study, drawn from more than 13,000 search campaigns across 23 industries, puts the 2026 average CPC at $5.42, "over twice what it was 10 years ago ($2.32 versus $5.42)." The average cost per lead in the same study is $66.69, and "across all industries, cost per lead decreased overall for the first time in five years" (LocaliQ, 2026 Google Ads Benchmarks). The two prior editions showed CPC rising 10 percent and then 12.9 percent year over year, with CPC up in 86 and 87 percent of industries respectively (2024, 2025).

Clicks are getting more expensive and leads are not, which is what you would expect if advertisers as a group are paying more per click to buy fewer, better clicks. It is consistent with the auction logic above and it is not proof of it. Benchmark averages blend thousands of accounts with different goals. Treat it as the market moving the same direction we push individual accounts.

Google's published case work points the same way. When Volkswagen Middle East assigned values to online and offline actions and bid to value, the result was a "19% lift in lead qualification, an 18% drop in cost per test drive, and a 28% decrease in cost per qualified lead" (Think with Google, October 2025). Google internal data from 2021, reported by Search Engine Land, found advertisers who moved from Target CPA to Target ROAS saw a 14 percent increase in conversion value at similar return on ad spend (Search Engine Land, 2023). Neither source reports what happened to CPC.

Where it stops being true

Higher is the correct direction, with two limits.

The first is a value model that flatters you. In any auction where bidders estimate a common value, the winner tends to be the bidder who overestimated most; Richard Thaler's 1988 summary still holds, that "the winner of the auction is likely to be a loser" (Thaler, 1988, *Journal of Economic Perspectives*). If you tell Smart Bidding a signup is worth $500 and half of signups are students, it will outbid competitors whose data is better than yours, and it will win the auctions it should have lost. Client B's 25x multiplier was exactly that, which is why it came down to 2.5x.

The second is that the platform has no reason to show you the ceiling. Google earns revenue on spend. Point Smart Bidding at a real revenue event and it will chase those buyers as far as the budget allows, sometimes past the point where the next click is still incremental. Google's guidance concedes the transition is bumpy. After a goal change, "Smart Bidding will take some time to learn (1-2 conversion cycles in most cases)," and "you may see volatility if a large change in targets causes you to enter an entirely new set of auctions" (Google Ads Help, Google Ads Help). Entering a new set of auctions is the point of the change, and it is the moment to watch closely.

The data behind the bid is what keeps a higher CPC honest. Feed the machine accurate values for the events that carry revenue and the extra money goes toward people worth reaching. Feed it a flattering value model and the same higher CPC buys more expensive junk.

How to judge the program instead

Judge the program on three numbers, none of which live on the ads dashboard. They come from the CRM and the invoice.

1. Cost per qualified opportunity, using the CRM stage your sales team would defend in a pipeline review.

2. Cost per dollar of pipeline for accounts with value-based bidding, measured on the same lag as your sales cycle. Google's own advice is to wait for one or two conversion cycles before reading results, and for a 60-day cycle that means two to four months of patience.

3. A holdout, geographic or audience-based, at least once a year. The eBay, Facebook, and retailer experiments above all found the same thing: the observational read overstates the effect, often by multiples. The only way to know your paid search is incremental is to turn some of it off and measure.

If those three are moving the right way, the CPC is doing its job. If cost per opportunity is not on the weekly report, ask for it.