How to Train the Algorithm: A Modern Approach to Paid Ads and AI

I spent six years on the agency side running paid ads for small and mid-sized businesses before moving in-house at Calendly for four years, and now I lead paid acquisition at Luxury Presence. Over that decade, the advertiser’s job has changed fundamentally: less manual targeting, more training the algorithm to find the right people for you. 

Google and Meta (Facebook, Instagram) are both fully leaning into  machine learning now. The platforms want to control more and more of the experience, from who you target to what you say, to where you spend your money. Some of that automation is genuinely helpful. But if you just accept every default recommendation and chase the “account quality score” up to 100, you’re handing over the keys without teaching the driver where to go. 

The advertisers who win are the ones who understand that their real job is feeding the algorithm the right data so it gets smarter over time.

TLDR

• Your most important job as an advertiser is feeding quality conversion data back into the ad platforms, because that’s the only way the algorithm learns what “good” looks like.
• Optimizing to clicks or even leads is a starting point, but the real gains come when you can attribute ads to meetings and closed deals.
• The post-click experience (your landing page, your follow-up) is where most conversions are won or lost, and most advertisers barely think about it.
• Start tight with targeting and spend, learn for 90 days, then scale based on what the data tells you.
• Every step in the funnel has one job. When you try to make a landing page or an ad message do five things, it does none of them well.

The algorithm only knows what you tell it

Google and Meta are learning systems, and the quality of what you feed them determines the quality of what you get back. It’s the same principle as working with AI chat tools: garbage in, garbage out.

If you’re only optimizing to clicks, the platforms will happily get you clicks all day. Cheap ones, too. But clicks from people who will never convert don’t help your business. If you step up to optimizing for leads, that’s better, but you can still end up with a flood of form fills from people who never show up to a meeting.

The sweet spot is optimizing to a conversion event that lets you know a lead is quality, but also happens often enough that the ad systems can learn from it. When you’re able to feed offline conversions back into the ad platforms (meetings booked, deals closed), the algorithm starts to understand what a high-quality prospect actually looks like. It builds a profile based on real outcomes, and it gets better at finding more people who match that profile.

Start by looking at your CRM data or analytics tool, and check which ad-sourced leads actually booked meetings in the last 90 days. Make sure you have UTM tracking parameters set up on your ads so you can see which campaigns and creatives are actually earning you listings.

Ideally, integrate your CRM with the ad platform itself, or use an all-in-one growth platform that handles it for you. You can also set up offline conversion tracking in Google or Meta if you have the technical resources to do so. Even a manual CSV upload is better than nothing. This will allow you to see not only which ads are delivering your leads, but which are resulting in meetings and closings.

Start tight, learn for 90 days, then expand

One of the most common mistakes I see is going too broad too fast. Google’s default recommendations push you toward broad match keywords, where a search for “HVAC service in Austin” might match your ad targeting “homes in Austin.” Meta’s Advantage Plus wants to decide everything for you. For some advertisers with massive budgets and years of data, that can work. For someone starting out or working with a focused audience like real estate, it’s a recipe for wasted spend.

I always start with the tightest possible targeting and expand from there. On Google, that means exact match and phrase match keywords with a strong negative keyword list from day one. You’re essentially whitelisting who can see your ads, and then continually monitoring search terms and audience data to prune out anything that’s not relevant.

On Meta, it means location targeting around your service area and, at minimum, some interest-based filters. Meta also analyzes your ad copy, images, and videos to decide who to target. Ensuring your ideal audience is clearly represented in your ad creative is a great way to tighten the quality of your audience.

Give yourself three months. The first month or two is learning: what hooks resonate, which keywords convert, where the gaps are. The ad platforms need that time too, to build conversion data and learn who they should be serving your ads to. If you’re looking for proof of return in the first 30 days, you’re measuring a marathon by the first hundred meters.

This week: Audit your keyword match types in Google Ads. If you’re running broad match on anything, switch the highest-spend keywords to phrase or exact match. Add negative keywords for common irrelevant terms in your market (careers, jobs, rentals, commercial, if those aren’t your business).

One job at every stage of the user experience

I like to think about the entire ad experience as a series of stages, and each stage has exactly one job. The targeting gets the ad in front of the right person. The ad earns the click-through to the landing page. The landing page captures the lead with a simple form. And once you have the lead, your nurture sequence builds the relationship toward a meeting.

When you start asking one stage to do multiple jobs of another, everything gets diluted. A landing page that tries to book meetings and showcase your bio and link to your blog and offer a home valuation is doing four jobs, which means it’s doing none of them well. An ad that tries to close the sale on the first impression is proposing on the first date.

This matters especially in real estate, where deal cycles are long. Someone might find you and fill out a form, but not be ready to buy or sell for another 12 to 24 months. The agents who ultimately win that deal are the ones who built the relationship over time through retargeting, through email nurture, through consistently offering something valuable rather than constantly asking for the meeting. You’re filling a bucket, and your job is to keep filling it while preventing leaks as much as possible.

Map out your current funnel and write down the one job of each stage. If any stage is trying to do more than one thing, simplify it. And if you don’t have a lead nurture email sequence set up, that’s a bigger priority than any change you could make to your ad copy.

The long game is the only game

The unit economics of advertising for real estate agents are actually in your favor. If you’re spending $500 a month on ads and your commission on a single closed deal is $10,000 to $15,000, one additional deal from your ad campaigns could cover one to two years of spend. You don’t need to close a ton of deals to see a strong return.

But that math only works if you treat this as a long game. Feed the algorithm real conversion data so it gets smarter. Build landing pages that do one thing well. Use AI tools to sharpen your ad copy, and give every lead the follow-up they deserve, because the person who filled out your form today might be your biggest deal eighteen months from now.

For your own advertising, calculate your break-even point. Take your average commission, divide it by your monthly ad spend, and figure out how many deals per year you need from ads to come out ahead. For most agents, that number is surprisingly low, and seeing it written down makes the 90-day learning curve a lot easier to commit to.