The learning phase is the stretch after you launch or heavily edit an automated campaign when the platform is still buying data rather than buying efficiency. Until the system has enough conversion signals, it cannot reliably price an impression, so spend goes to audiences, placements and bid levels it is still testing. Google Ads and Meta Ads Manager both show a “Learning” status, then remove it once predictions settle.
Why early results mislead you
Conversion-based bidding runs on probability. A fresh campaign, a new conversion goal or a changed audience weakens those probabilities. The label is the platform’s way of saying: do not judge this window as steady state.
Meta applies the status at ad set level. Its Business Help Centre guidance says an ad set generally needs about 50 optimisation events within seven days of its last significant edit. For a purchase-optimised ad set, that means actual purchases. If volume is too low, the ad set is flagged Learning Limited. In 2024, vendors reported Meta’s purchase and app-install threshold falling to 10 events; in August 2026, Advantage+ shopping campaigns were reported to exit after 25 purchases.
Google attaches learning to bid strategies. Google says calibration can take up to around 50 conversion events or 3 conversion cycles. Manual CPC has no learning period. TikTok, Pinterest, Microsoft and Reddit all publish their own versions, which is part of the problem.
The numbers vary by platform. TikTok says volatility typically declines after about 25 campaign results or seven days, while its FAQ still points to 50 conversions as the strongest signal. Pinterest says learning lasts about two weeks on average. Microsoft Advertising suggests portfolio bid strategies need a few weeks. Reddit asks for 14 days without changes for Max campaigns. Display & Video 360 says automated bidding can take up to four weeks.
What actually resets the clock
- New campaign, new conversion goal or new bid strategy
- Major targeting, placement or creative changes
- Pausing long enough for the platform to lose signal
- Large budget or bid changes
- AI agents or tools that make frequent automated edits
Every reset forces the advertiser to repay the inefficient spend. That is why the structural advice often comes down to fewer, larger campaigns. Meta recommends bigger budgets across fewer ad sets so they hit 50 events sooner.
One advertiser’s fix
The source highlights a telling example: an advertiser merged three $50-a-day ad sets into a single $150-a-day Advantage+ campaign. ROAS moved from 1.8 to 3.2. The bidding system finally had enough data in one place to optimise against purchases rather than guesses.
Google’s Brandon Ervin, director of product management for search ads, said merging similar ad groups or campaigns should not require much learning because models key on semantic features rather than campaign IDs. Switching conversion actions or adopting new AI campaign types does force a reset.
What to do differently
Build freeze windows into your launch plan. Google asks Demand Gen advertisers on value-based bidding to avoid bid or creative changes for 14 days. Reddit recommends 14 days of stability for Max campaigns. Judge Google’s Target CPA after at least 30 conversions and Target ROAS after 50.
If you cannot reach a conversion threshold, consider optimising toward an earlier-funnel event, but understand the trade-off. A campaign optimised for add-to-cart learns to find people who add to cart, not necessarily buyers.
With AI connectors now editing campaigns, there is a new risk. Meta client partner Brody K. warned: “Just be sure to remind your agent about the learning phase.” Frequent small edits can keep a campaign in calibration indefinitely.
Source: PPC Land
