How the Meta ads algorithm actually decides who sees your home service ad

Most agencies hand you a report with reach numbers and call it transparency. They won't tell you how Meta actually decided who saw the ad, because understanding that would let you hold them accountable for the targeting decisions they made.

Here's what they're not explaining.

Meta doesn't use the audience you think it does

When an agency sets up your campaign, they'll often show you a targeting panel: age range, location radius, interests like "home improvement" or "homeowners." That looks like control. It mostly isn't.

Meta's algorithm treats manual interest targeting as a starting point at best. What it actually optimizes for is the conversion signal you give it. When someone submits a lead form or lands on your page, Meta logs that event and looks for more people who match that behavior profile. The campaign learns who converts, not who "looks like a homeowner."

This matters because if your campaign is set up without proper conversion tracking, Meta has no signal to learn from. It's flying blind, serving your ad to whoever fits the rough demographic box the agency checked, with no feedback loop to tighten the audience over time. The leads that come in are whoever happened to be in range, not whoever was likely to book a job.

That's why one painter described getting "people looking for jobs instead of people looking for to-do work." The ad was reaching people, just not the right ones. The algorithm wasn't optimizing for booked jobs because nobody told it what a booked job looked like.

The campaign structure shapes who the algorithm learns from

Beyond conversion tracking, the way a campaign is structured tells Meta a lot about what it's supposed to do. A poorly structured campaign, one objective mismatched to the goal, one ad set with no creative variation, a budget too low to generate enough signal in the learning phase, forces the algorithm to make bad guesses with insufficient data.

Meta's learning phase typically requires around 50 optimization events in a seven-day window before the algorithm exits that phase and starts performing predictably. If an agency sets a daily budget that only generates a few form submissions per week, the campaign never exits learning. It keeps guessing. The contractor watches money leave their account while the algorithm is still warming up, and by the time it might have worked, they've run out of patience or budget.

Most agencies don't explain this because explaining it would raise the obvious question: why didn't you structure it to actually exit the learning phase? A campaign that never exits learning isn't a campaign. It's a test that never produced a result, billed as a service.

Instant forms versus landing pages change what Meta optimizes for

This is the distinction that almost sank one of ASN's own sales calls. A painter named Marco had originally believed ASN's differentiator was landing pages instead of instant forms. When he discovered the standard plan used instant forms, he nearly walked.

His skepticism was grounded in real experience. Instant forms, the native lead capture forms inside Facebook and Instagram, have low friction for the user and therefore attract lower-quality responses. Someone scrolling through Instagram can submit their name, phone number, and email in three taps without ever leaving the app. They may not remember doing it an hour later.

Landing pages create more friction. The person has to click through, leave Facebook, read something, and fill out a form on an external site. Fewer people complete it. But the ones who do are more committed. They took a deliberate action.

Meta optimizes for whatever converts on the form you set up. If you're using an instant form and getting 80 submissions a month but only booking 4 jobs, Meta is filling your funnel with people who were never serious. It's technically delivering leads. Those leads just don't convert into revenue.

The best setups pair the right form format to the right trade and budget, track downstream events (not just form submissions), and split-test creative to give the algorithm variation to learn from. That combination is what produces the Safe Step result: 247 leads at $11 per lead, $2,800 in total spend. That outcome doesn't come from checking interest boxes. It comes from giving the algorithm real signal, real variation, and enough budget to actually learn.

Why agencies don't explain this

The simplest answer is that explaining the algorithm creates accountability. If a contractor understands that conversion tracking is required for the algorithm to learn, they'll ask whether conversion tracking is actually set up. If they understand the learning phase, they'll ask whether the budget is sized to exit it. If they understand the difference between instant forms and landing pages, they'll ask which one the agency is using and why.

Agencies that rely on vague reporting, "here's your reach, here's your impressions, here's your cost per lead" with no explanation of what the algorithm was actually optimizing for, are betting on the contractor not asking those questions.

The second reason is that many agencies running ads for home service contractors are generalists. They run the same campaign structure for a landscaper as they do for a software company. They don't have niche-matched data to pull from, so they can't tell you what a realistic cost-per-lead looks like for epoxy flooring in your city, or how long the learning phase typically runs for a HVAC campaign at a given budget. They fill that gap with jargon.

What to actually look for when evaluating an agency

Ask specifically whether conversion tracking is installed and what event it's tracking. "We track form submissions" is a real answer. "We track leads" with no further explanation is not. You want to know whether the pixel is firing on the right page and whether Meta is receiving the signal it needs to learn.

Ask what the campaign structure looks like: how many ad sets, what creative variation is planned, and how many leads per week the budget is expected to generate in the first 30 days. If the projected volume is too low to exit the learning phase, the campaign isn't designed to work. It's designed to run.

Ask for case studies from a business in your actual trade, not adjacent trades. A 22x ROAS result for a lighting company (Yerim) tells you something about the agency's ability, but a roofing contractor needs roofing numbers. The Safe Step result (247 leads, $11 CPL) is meaningful to an epoxy or rubber resurfacing business because the trade is close enough to matter. A generic ROAS screenshot from a mismatched industry tells you almost nothing about what to expect.

What a straight answer looks like

If you're considering running Meta ads again after getting burned, the questions above are exactly the ones to ask before signing anything. An agency that can answer them specifically, with real numbers from real campaigns in trades close to yours, has probably run the campaigns themselves and knows what the algorithm actually needs to perform. An agency that deflects into reach metrics and vague promises is running the same playbook that already failed you.

ASN runs campaigns inside your own Meta Ads Manager so you can see everything. No setup fee, no contract. If you want to see how a campaign would be structured for your trade, reach out here and we'll walk through it with real numbers.