What to check before you touch anything
- How many conversions is the campaign getting per week? Google's own rule of thumb is around 30 conversions in 30 days as the minimum for tCPA to learn reliably. Below that, "Learning" can drag on indefinitely — that's not a glitch, it's just not enough data yet.
- Has anything changed in the last 1–2 weeks: the target CPA itself, the budget, conversion actions (added or removed), campaign structure (merged or split ad groups), or landing page URLs? Any of these resets or extends the learning period.
- Could this simply be a seasonal dip in conversions — fewer conversions happening in general, not the system "failing to learn"?
- Is there a lag between clicks and recorded conversions? If your sales cycle is long, some conversions may have already happened but haven't been counted yet, so Google is working with less data than actually exists.
- Is the campaign also flagged "Limited by budget"? If so, the issue isn't learning per se — it's that the budget physically can't buy enough impressions/clicks to learn from.
- Is this a brand-new account or campaign, and has it had any recent downtime? Pausing a campaign also wipes out part of its accumulated learning signal.
Possible approaches
- If conversions are genuinely too few, one option is to temporarily raise the budget and/or loosen (raise) the target CPA to feed the system more data, then gradually walk the target back down once it has enough signal.
- You could merge a few small tCPA campaigns that share the same goal into one — this can speed up data accumulation, but it's a real trade-off: you lose per-segment budget control, and it won't fit every account structure.
- A common move is to run on Maximize Conversions (no target) for a warm-up period, then move to tCPA once enough data has built up — Google itself advises against starting a brand-new campaign with an aggressive target.
- If the issue is a long, low-frequency conversion cycle, widening the conversion window can help the system see delayed conversions more accurately — but that's also a trade-off: it slows down how quickly the system can react and optimize.
- Either way, avoid stacking several changes at once (budget + target + structure) — otherwise you won't be able to tell what actually worked, and the learning status will reset again.