Over the past two years Meta has handed ad delivery to AI algorithms: Andromeda selects ads, GEM predicts who will act, and Advantage+ automates the settings. Targeting has become a suggestion, and the creative is now the main audience filter. According to Meta, Advantage+ for leads cuts cost per lead by 14% on average. Timelines: ad review usually takes up to 24 hours, first leads come within 1–3 days, learning ends after about 50 events, and cost per lead stabilizes in 2–4 weeks. Learning budget: 50 × cost per lead per ad set per week.
The two most common questions before a launch are “so how do Facebook ads even work now?” and “when will the first leads come in?”. The answers are connected: AI automation has changed both where traffic comes from and how long ads take to reach stable results.
What AI automation changed
Facebook ads used to work like this: a specialist picked interests, age and location, and the algorithm showed ads inside that “box”. Now the box is almost gone. Three systems are at work:
| System | What it does | What it means for you |
|---|---|---|
| Andromeda | Picks a few thousand ads relevant to a specific person out of tens of millions | Different creatives reach different people, so you need a set of varied ads |
| GEM | A large model that predicts who will take action | The more precise the optimization event, the more precisely it searches |
| Advantage+ | Automates audiences, placements, budgets and creative variations | Your audience settings are a suggestion, not a constraint |
According to Meta, GEM delivered +5% conversions on Instagram and +3% in Facebook Feed, and leads campaigns with Advantage+ turned on show on average a 14% lower cost per lead and a 10% lower cost per qualified lead.
How traffic changed
- Audiences are broader. The algorithm looks for people beyond the interests you set. That lowers the price, but also brings random people if the creative doesn't filter them out.
- The creative has become the targeting. The algorithm will show a video for moms of preschoolers to moms of preschoolers even with no interests set. So what matters now isn't the settings but a set of ads with different messages for different segments.
- More placements. Reels, Stories, Feed, third-party apps: ads are distributed automatically, and traffic quality varies a lot between placements.
- Results depend more on data. The algorithm learns from the events you send it. Good data means good leads; “dirty” data means cheap but empty ones.
What advertisers should do in the age of AI
- Give the algorithm 6–10 different creatives. Not variations of one banner, but different ideas for different segments, pain points and formats.
- Choose the right optimization event. A lead, a qualified lead, a booking, a purchase: the algorithm will look for exactly that.
- Connect the Conversions API and CRM data. That way the algorithm sees not only leads but which of them became customers.
- Don't split the budget too thin. Fewer campaigns and ad sets, each with enough budget to learn.
- Don't touch the campaign every day. Significant changes to budget, audience, creatives or objective restart learning.
When to expect the first leads: a realistic timeline
| Stage | Timeframe | What happens |
|---|---|---|
| Preparation | 3–7 days | Audit, offer, creatives, landing page or form, pixel and Conversions API |
| Ad review | Usually up to 24 hours | Meta checks the ads against its policies |
| First leads | 1–3 days after launch | The campaign starts delivering, first leads arrive |
| Learning phase | About 50 events | The algorithm searches for the audience; cost per lead fluctuates |
| Stable results | 2–4 weeks | Cost per lead levels out, and it's clear which creatives work |
| Optimization and growth | Months 2–3 | We turn off what's weak, scale what's strong and expand into new audiences and markets |
The learning budget formula
An ad set exits the learning phase after about 50 optimization events since the last significant edit. Hence a simple formula:
Weekly budget per ad set = 50 × expected cost per lead
Example: if a lead in your niche and market costs about $5, one ad set needs roughly $250 a week to learn within 7 days. If a lead costs $15, you'll need about $750. With less budget, merge ad sets into one or optimize for a more frequent event; otherwise the campaign will get stuck in “Learning limited”.
When results take longer
- A small budget. Fewer than 50 events a week and learning drags on.
- An expensive niche or high ticket. People need time to decide, and there are fewer leads.
- Optimizing for sales or CRM data. There are fewer such events than simple leads, so learning takes longer, but quality is higher.
- A new account and a pixel with no history. The algorithm has nothing to go on in the first weeks.
- A new market or language. Creatives and the offer need to be tested again.
The main rule of the first weeks: don't panic. Cost per lead during learning is almost always higher than the working level, and frequent edits only drag the period out.
Sources
- Meta Engineering: Andromeda, the ad retrieval engine for Advantage+
- Meta Engineering: the GEM ads recommendation model
- Meta for Business: Advantage+ leads campaigns
- Meta Business Help: about the learning phase
- Meta Business Help: about the ad review process
Want a launch that brings leads in the first few days?
We'll build a campaign structure for Meta's AI algorithms, prepare creatives and calculate the learning budget for your niche and market. Free 30-minute consultation.
Get a free consultationFrequently asked questions
How soon do the first leads come in after launching Facebook ads?
Usually within the first 1–3 days. Meta's ad review typically takes up to 24 hours, after which the campaign starts delivering, and with a solid offer the first leads arrive within a day or two. But first leads aren't yet a stable result: the algorithm is learning, and the cost per lead settles after 2–4 weeks.
What is the learning phase and how long does it last?
It's the period when Meta's algorithm figures out who to show your ads to. An ad set exits learning after about 50 optimization events (leads, purchases) since the last significant edit. With enough budget this takes a few days to a week; with a small budget the ad set may stay stuck in “Learning limited”.
What budget does a campaign need to finish learning?
A simple formula: 50 events × expected cost per lead = weekly budget per ad set. If a lead in your niche costs about $5, one ad set needs roughly $250 a week to learn. If your budget is smaller, it's better to merge ad sets than to spread the money thin.
Do you still need a paid social specialist if AI runs the ads?
Yes, but the job has changed. Instead of hand-picking interests, the focus is now on creatives, the offer, campaign structure, signal quality for the algorithm (Conversions API, CRM data) and analyzing which leads turn into sales. AI pursues the goal you set; setting the right goal and feeding it good data is still a human's job.
Does Advantage+ work for lead generation?
According to Meta, leads campaigns with Advantage+ turned on deliver on average a 14% lower cost per lead and a 10% lower cost per qualified lead. In practice results depend on the niche and signal quality, so we test Advantage+ alongside a manual structure and compare the cost per qualified lead, not just cost per lead.