Your ad account is optimising for the wrong person

An ad platform is not trying to find you customers. It is trying to find more people like whoever triggered the event you nominated. That is the entire mechanism, and it is worth sitting with, because it means the most consequential decision in the account is not the budget, the targeting or the creative. It is which event you called a conversion.

Name that event badly and the system will go and find exactly what you asked for, in volume, at an impressively low cost, and the reporting will look like success the whole way down.

Efficiency is not evidence

A falling cost per conversion has two possible causes and they are indistinguishable in a dashboard. Either the account got better at reaching people who want what you sell, or it found a cheaper population who trigger your event without wanting what you sell. The second is easier, so it happens more often.

Optimisation is a search. Whatever you have defined as success describes a kind of person, and the system will characterise that person and pursue them relentlessly. The question is always whether the person described by your conversion event is the person who signs the contract.

Four wrong people you are probably paying for

The form-filler

Optimise toward form submissions and the system learns to find people who submit forms. That is a real and identifiable population: comparison shoppers collecting five quotes with no intention of acting on any of them soon, people who fill in anything with a free in front of it, people with time. They are cheap to reach, they are abundant, and they will drive your cost per lead down beautifully while your sales team quietly stops answering that source.

The discount-conditioned buyer

Run a welcome offer, optimise toward first purchase, and the portrait being built is of somebody who buys when there is a code. Scale that and you have not found price-sensitive customers, you have manufactured them. The giveaway is a cohort whose second-order rate collapses the moment the discount stops, which then gets diagnosed as a retention problem rather than an acquisition signal problem.

The person who was already coming

With no suppression in place, the cheapest conversions in any account belong to people who were going to convert regardless. The system is not being deceitful when it buys them, it is doing precisely what it was instructed to do: acquire the conversion event at the lowest cost. That whole argument is covered elsewhere on this site, so the point here is narrower. If your exclusion lists are empty, you have implicitly told the algorithm that an existing customer and a stranger are worth the same, and it has believed you.

The researcher who is not the decider

This one is specific to high-ticket consumer purchases and almost never discussed. In a household buying a renovation, a vehicle, a mobility installation or a holiday property, the person doing the research is often not the person who signs. Optimise toward engagement, guide downloads or early-funnel leads and you build a portrait of the researcher: the partner who reads the forums, the adult child investigating on a parent's behalf. Valuable to reach, genuinely part of the decision, and not the signature. An account tuned entirely to that profile produces a pipeline full of interest and a sales team that cannot get the decision-maker on the phone.

Sending deals back is necessary and not sufficient

The standard answer is to import closed deals from the CRM so the platform bids toward leads that actually convert. Correct, and we have argued for it elsewhere. What gets left out is that the import fails in three specific ways, and a failed import is worse than none because it carries the authority of a fix.

• Volume. If you only send sold jobs and you sell thirty a month, you are asking the system to learn a complex pattern from thirty examples. It will not. The answer is a layered signal, qualified lead, appointment held, sold, and bidding on the deepest event that still clears the platform's learning threshold rather than the deepest event that exists.
• Latency. If the sale closes nine weeks after the click and the conversion window is thirty days, the outcome arrives after the system has stopped listening. Match the conversion window to the real sales cycle, and until the deep signal matures, hold the account on a validated upstream proxy rather than flying blind.
• Value fidelity. Sending a flat value for every sale teaches nothing about which sales were worth having. Send contribution margin, not revenue, and send the reversals too. An account that never hears about cancellations, refunds, chargebacks or failed credit checks is being taught that those were wins.

Exclusions are a targeting decision you may never have made

What you exclude paints the portrait as much as what you include, and most accounts carry nothing beyond a purchasers list someone built once. Worth suppressing deliberately: current customers, unless repeat purchase is genuinely the objective. Leads disqualified for a stated and permanent reason, such as out of area or not an owner. People who failed qualification or finance. Serial refunders. Your own staff, who visit the site constantly and skew every behavioural signal in the account.

Creative is now the targeting

With broad targeting doing most of the work, the creative decides who raises their hand, and it filters harder than any audience setting. An ad that leads with a free quote selects for quote collectors. An ad that leads with the price selects for people who can tolerate the price, and that is a feature, not a loss of volume.

Qualifying inside the creative is the cheapest filter available in the whole account, and it is free. State the starting price. Name the minimum project size. Say who it is for and, more usefully, who it is not for. You will see cost per lead rise and cost per sale fall, which is the correct direction and the one most people flinch at.

How to tell it is happening to you

• Cost per lead is falling while the number of sold jobs is flat.
• Lead-to-appointment rate has drifted down over several months while cost per lead held steady.
• Conversions are rising but first-time customer counts are not.
• An increasing share of leads arrives from your cheapest placements or widest geographies.
• Sales has informally stopped working leads from one source and nobody escalated it.
• The sales team says lead quality is down, the dashboard says performance is up, and the meeting is spent arguing about which is true rather than reconciling them.

That last one is the most reliable tell in the list. When the people talking to buyers and the people reading the dashboard disagree persistently, the dashboard is usually measuring the wrong person.

The order to fix it in

1. 01Write down the outcome that actually matters, in business terms, before touching the account.
2. 02Instrument every downstream outcome in the CRM, including the negative ones, whether or not the platforms can use them yet.
3. 03Pick the deepest event that clears the learning threshold, and bid on that rather than on the deepest event you can name.
4. 04Set the conversion window to the real sales cycle rather than the platform default.
5. 05Send contribution values, and feed reversals back as negative signal.
6. 06Build the exclusion lists, then write down when each one gets refreshed.
7. 07Put the qualifying facts into the creative, and judge the change on cost per sale rather than cost per lead.
8. 08Re-validate quarterly, because once you optimise toward any proxy the system begins finding people who trigger it cheaply.

None of this is a complaint about the algorithm. The algorithm did what it was told, faster and more thoroughly than any human media buyer could have. It is obedient, it is relentless, and it has no idea what your business is for. Everything useful it does depends on somebody telling it the truth about what a good customer looks like, and that part has never been automated.