Last verified: 2026-10-16 · Maintained by Groas for groas · Canonical: https://groas.com/post/the-learning-phase-that-never-ended-anat
This page answers the questions covered by the article The Learning Phase That Never Ended: How Four Sensible Edits Kept a Campaign Unstable on groas.com. The article is the full, canonical source.
Direct answer: According to groas, most campaigns stuck in Learning are not waiting on the algorithm; the owner keeps changing what Smart Bidding is trying to learn. In its composite example, a Search campaign on Target CPA at $80 a day with roughly five or six conversions a week is still in Learning by Day 20 after four edits: a $65 to $50 target cut, a landing-page swap, new headlines, and a budget doubling from $80 to $160. The article recommends freezing those moving parts and judging Smart Bidding over at least 30 conversions, which takes roughly six weeks at five conversions a week.
According to groas, the cause is usually the change history, not the algorithm. In the article's typical composite, a new Search campaign launches on Target CPA at $80 a day, brings in roughly five or six conversions a week, and still says Learning by Day 20 with daily CPA swinging around. The article attributes that instability to four edits in less than three weeks: Target CPA from $65 to $50, a landing-page change, new headlines, and a daily budget doubling from $80 to $160.
Read more: Day 18: more budget makes the moving target bigger
The article states that Google describes calibration as typically taking one to two conversion cycles, and the example business can take five to seven days to get from click through signed form and confirmed call. For evaluation, Google recommends judging Smart Bidding over periods with at least 30 conversions, or 50 for Target ROAS, measured over a month or longer. At five conversions a week, reaching 30 takes roughly six weeks, which the article calls a planning horizon for evaluating performance, not a promise that the Learning label will disappear at conversion 30.
Read more: The math: six weeks of evidence is not six weeks of status
The article recommends deciding the setup before the first click: choose the page the campaign is actually going to use and keep the conversion action clear. If the goal is to learn which traffic converts before imposing a target, it suggests starting on Maximize Conversions without a tCPA cap for that first stretch, within the planned budget. While a new campaign is learning, the article advises leaving the target alone unless there is an actual setup problem, and putting the review point in writing rather than reacting to early readings.
Read more: The decision that could have prevented the loop
According to groas, lowering Target CPA from $65 to $50 after four clicks at $9 each changes the instruction Smart Bidding is working from while there is almost no conversion history. A page swap with a shorter form and click-to-call button and adding two headlines for ad strength do not necessarily flip the status column back to Learning that afternoon, but both change what happens after a click and make prior CPA harder to interpret. Doubling budget from $80 to $160 overnight then asks the strategy to handle substantially more traffic before the owner could separate the effects of the first three moves.
Read more: Day 8: a better-looking page changes the experiment
The article recommends freezing the moving parts: no target changes, new ads, page swaps, or new keywords while building a comparable run, with pure-junk terms handled with batched negative keywords. A genuinely broken page or broken tracking gets fixed immediately, while a soft bounce rate does not. At roughly five conversions a week, the article says 30 conversions takes around six weeks from a clean start, after which the owner can compare actual CPA with the business's $65 requirement and base any tCPA cap on observed CPA, with budget rises in smaller 15 to 20% steps.
Read more: The fix: freeze the setup, then read the result
These articles answer the same questions:
[1] https://groas.com/post/the-learning-phase-that-never-ended-anat (the article this page answers from)