Skip to main content
Browse topics
Back to Docs

Ad Platform Playbook

Learning Period and Strategy Transitions

What triggers learning, what resets it, change sequencing, and the 20% budget threshold

For · Operators about to change a bid strategy or budget
Time · 5 min read
Next · UTM attribution template

When you change a bid strategy, the platform enters a learning period, a window where it recalibrates using historical data and new auction signals. During this period, performance can fluctuate. How you manage the transition matters as much as the strategy you choose.

The Short Version

  • A learning period starts on a strategy switch, a bid target change, a budget change over 20 percent, or a conversion action change.
  • Google documents the period as roughly 50 conversion events or three conversion cycles, which for a low-volume campaign can mean weeks.
  • Additional changes during learning restart the clock. A campaign that needed two weeks can take six if you keep interrupting it.
  • Sequence changes in phases: structural first, bid strategy second, budget last. Three changes at once leave you unable to attribute the result to any of them.
  • A campaign that never clears the volume threshold never fully exits learning. Enhanced CPC is the fallback that does not require one.

What Triggers the Learning Period

The learning period typically starts when you:

  • Switch to a new bid strategy (e.g., Manual CPC to Maximize Conversions)
  • Change a bid target (e.g., adjusting your Target CPA)
  • Make a budget change greater than 20% of the current daily budget
  • Modify conversion actions (adding, removing, or changing the primary conversion)

Google’s documentation notes the learning period lasts roughly 50 conversion events or three conversion cycles. For low-volume campaigns, that can mean weeks or longer.

What Resets Learning

Making additional changes during the learning period can reset the clock. If you switch strategies and then adjust the budget a few days later, the platform may restart its calibration. The same applies to adding or pausing keywords, changing conversion actions, or modifying bid targets.

The practical consequence: each interruption extends the period of suboptimal bidding. A campaign that needs two weeks to learn can take six if changes keep resetting the process.

Change Sequencing

Because simultaneous changes make it difficult to isolate cause and effect, Trellis audits recommend a phased approach:

PhaseTimingActions
1ImmediateStructural changes: negative keywords, keyword pauses
2Week 2Bid strategy changes (if still needed after Phase 1 impact)
3Week 3+Budget adjustments

The logic: if you change three things at once and performance shifts, you can’t determine which change caused it. Phasing lets you evaluate each change on its own data.

The 20% Budget Threshold

Budget changes during an active learning period carry specific risk. Changes exceeding 20% of the current daily budget can reset learning progress, extending the window of inconsistent performance. If a budget adjustment is necessary during learning, keeping it under 20% reduces the likelihood of a reset.

When Volume Falls Short

Automated strategies depend on conversion volume. If a campaign consistently generates fewer than 30 conversions per month, strategies like Maximize Conversions or Target CPA may never fully exit the learning state. They simply don’t have enough data to optimize.

In one real case, a Microsoft Ads campaign running Maximize Conversions never reached the platform’s documented 30-conversion threshold. The strategy continued running but wasn’t actively optimizing bids. Switching to Enhanced CPC, which adjusts manual bids without a minimum conversion requirement, restored predictable bidding behavior.

Trellis Perspective

Trellis audits flag campaigns where recent changes may have reset the learning period, and where conversion volume is insufficient for the active bid strategy. Recommendations follow the phased sequencing above: structural fixes first, strategy changes second, budget adjustments last.

Sources