Meta's AI Found Your Buyer. Your Funnel Lost Them.

For most of the last decade, running Meta ads meant managing targeting. You picked interests, built lookalikes, split audiences, and the quality of your results tracked the quality of your guesses.

That job has largely gone. Meta’s own systems now do it better than you can, at a scale no media buyer can match. Which creates a problem most businesses have not adjusted to. When the platform gets better at finding people who might buy, the constraint moves. It stops being can we reach the right person and becomes does the right person believe us when they land.

What actually changed inside Meta

In December 2024, Meta published the engineering detail behind Andromeda, the retrieval system that decides which ads are even eligible to be shown to a given person. Three figures from that paper are worth reading carefully:

• A 10,000x increase in model complexity for personalisation.

• A 6% recall improvement to the retrieval system, producing an 8% ad quality improvement on selected segments.

• A 22% increase in ROAS reported by advertisers who enabled Advantage+ AI-driven features.

Two things follow from that.

First, the retrieval model works backwards from how media buyers think. It does not start from an audience you defined. It evaluates your creative, copy, format and historical engagement, then predicts who is likely to respond. Your targeting is a hint now, not an instruction. This is why accounts that were built on tight audience segmentation often perform worse than accounts that give the system room and spend their effort on creative instead.

Second, and this is the part that gets missed: a 22% ROAS improvement is a delivery improvement. More of the right people see the ad. It says nothing about whether they believe you once they arrive.

The bottleneck moved to trust

The same technology that improved ad delivery also made every claim online cheaper to manufacture. Landing pages, reviews, before-and-after photos, case studies, testimonials, entire brands can now be produced in an afternoon by someone with no business behind them.

Buyers adjusted. They did not become more sceptical of your business specifically. They became more sceptical of the entire category of things that look like your business. A prospect landing on your quote form has no way to tell you apart from the six other companies that ran a similar ad that week, and they know it.

So the ad works and the funnel does not. Cost per lead looks fine. Cost per booked call is climbing. Cost per closed deal is climbing faster. Most operators respond by changing creative or lowering the bid, because those are the levers they know. The problem is one step downstream.

The counterintuitive fix: add friction

Standard funnel advice is to remove friction. Fewer fields, faster form, get the phone number and let a salesperson do the rest. That advice was written for a market where attention was scarce and trust was cheap. Both have inverted.

When you strip a funnel to name, email and phone, you have optimised for volume of contact details and against quality of intent. The prospect gives up the minimum, learns nothing, and is then called by a stranger who opens with the questions they were not asked online. From their side, that call is an interrogation before they have received anything.

Add friction deliberately instead. Ask the questions that let someone qualify themselves, and give them something in return for answering.

Give the price first

The most useful thing you can hand a prospect before a call is a number.

Buyers behave consistently here. Price is one of the first things somebody looks for on a supplier site, and a site that will not show it sends them to one that does. Hiding cost reads as evasive, and the trust environment of the last few years has only sharpened that.

There is a second effect, and it is the one that decides deals. The vendor who names a price first is usually the one who is not afraid of the conversation. Everyone else is perceived as waiting to see how much the buyer can afford. In a market where a prospect cannot verify anything else about you, willingness to quote is one of the few honest signals available to them.

Most businesses refuse on the grounds that every job is different. Every job is different. That is an argument for an estimate with a stated range and stated assumptions. It is not an argument for silence.

What the funnel looks like

Take a kitchen remodelling company running Meta ads. The conventional version sends traffic to a page offering a free consultation and captures a phone number. The version that works in this market does something else:

• The ad sets expectation, not curiosity. It says what the funnel does: answer a few questions, see what a project like yours costs. Curiosity gets clicks. Expectation gets the right clicks.

• The funnel asks what a good estimator would ask. Scope, size, finish level, timeline, whether it is a full replacement or a refresh. Six to ten questions, one per screen, plain language.

• An estimate is returned on screen. A range, with the assumptions behind it visible. Not a brochure, not a callback promise. A number they can act on.

• The exact figure is the reason to book. The range is honest and the caveat is real: the precise price depends on things nobody can assess through a form. Booking a call is now the obvious next step for someone who liked the range, rather than a hurdle in front of information they wanted.

• Everything is captured either way. Answers, estimate shown, and whether they booked all land against the contact record, so the follow-up knows what the prospect already saw.

What this changes on the sales side

The prospects who think the number is too high leave without a call. That is the point, not a loss. A remodelling company that runs forty consultations a month to close six is spending most of its selling capacity teaching people they cannot afford the work. Filtering that out at the funnel is worth more than a lower cost per lead.

The prospects who book have already seen the price and booked anyway. The first call stops being a discovery interrogation and becomes a scoping conversation, which is a different meeting with a different close rate. Your salesperson is no longer establishing trust from zero. The funnel did that by being willing to answer the question everyone else dodges.

What to measure

Cost per lead becomes an unhelpful headline number once a funnel qualifies people out on purpose. It will get worse while the business gets better. Track instead:

• Funnel completion rate. What share of people who start the questions finish them. Below roughly half, the questions are too many or worded badly.

• Estimate-to-booking rate. Of the people shown a number, how many book. This is the honest measure of whether your price and your positioning match.

• Booked-to-held rate. No-shows collapse when someone has seen a price and booked anyway.

• Cost per held call, then cost per closed deal. These are the only two numbers that connect ad spend to cash.

• Sales hours per closed deal. The saving that never shows up in an ads dashboard.

Where this goes wrong

Three failure modes are worth naming.

The estimate is dishonest. A range engineered to look cheap produces booked calls and furious prospects. The range has to be one you would actually quote.

The questions serve you instead of them. If the funnel asks budget, timeline and decision-maker status before giving anything back, you have rebuilt the interrogation with extra steps.

Nothing behind it is connected. An estimate shown on screen and then lost is worse than no estimate, because your salesperson calls someone who already knows a number and asks them what their budget is.

The short version

Meta’s systems got materially better at putting your ad in front of a plausible buyer, and they will keep getting better. That capability is now roughly equally available to you and to every competitor in your market. It is no longer where the advantage sits.

The advantage sits in what happens in the ninety seconds after the click, in a market where nobody has a reason to believe anyone. Answer the question they came to ask, before you ask them for anything. That is most of it.

AI Optimize builds these systems for businesses that need client acquisition to be predictable rather than seasonal. The ads, the funnel, the qualification logic, the estimate engine and the CRM behind it, run as one system. That work sits under Paid Ads Management.

Sources

• Meta Engineering, “Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine”, 2 December 2024.