Why does changing the conversion window break the historical stats on an existing tCPA campaign?
What to check before you touch anything
Which direction did the window change — wider or narrower? Widening the window (say, from 30 to 60 days) usually adds previously uncounted conversions retroactively, causing a sudden jump in the historical numbers. Narrowing does the opposite — it retroactively removes conversions that used to count.
From what point does the new window actually apply? A conversion window change typically doesn't rewrite past reporting periods consistently across every report, which can create visible mismatches when comparing "old" and "new" numbers.
Did anything else change at the same time (attribution model, the set of conversion actions counted toward the main goal)? Often what looks like "broken stats" is really the combined effect of several simultaneous changes, not the window alone.
Does the team understand that changing the window is essentially changing the measuring instrument itself? Comparing "before" and "after" as if they're the same metric is a methodology mistake, not a sign something's broken.
How different is the new window from the old one relative to the real sales cycle? The right setting should be based on the actual lag, not a number picked at random.
Possible approaches
Standard practice is to treat the point where the conversion window changed as a "cut" in the historical data — compare before/after periods carefully, and ideally flag that moment explicitly in reporting rather than blending the two periods into one trend line.
Google generally warns that the bid strategy itself needs to recalibrate after a conversion window change — it's effectively a trigger similar to changing the target or the conversion actions, and some of the accumulated learning can reset.
If the window was changed to better reflect the real sales cycle, it's worth accepting the temporary instability as the fair cost of getting the setup right, rather than reversing the change at the first sign of a swing.
If the window was changed without first analyzing the real lag, it's better to pull the Time Lag report, determine a well-supported window value, and then change it once, deliberately — rather than trial-and-error adjusting it multiple times in a row.
For business reporting, it's worth explicitly logging the date of an attribution-window change as a system event (on par with a target or structure change) — so that any future anomalies in the trend get checked against this factor first, instead of the cause being sought elsewhere.
Normally, all of these settings — conversion window, attribution model, the set of conversion actions, dynamically assigned conversion values — have to be checked one by one across different parts of the interface, cross-referenced by hand. Our auditor (DataMind) lets you review all of your account's goals in one place, across every category of settings at once — from attribution settings to dynamically assigned conversion values — so mismatches between them show up immediately, instead of surfacing later as unexplained anomalies in your stats.
Related Content
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.
Compare your target CPA to your actual CPA over the last 30–90 days (while running Maximize Conversions or a target close to current performance). A gap bigger than 20–30% usually means the target has drifted away from reality.
Over what timeframe is this "consistent"? A 30–40% miss over 7 days and the same miss over 60 days are two very different situations — short-term noise isn't a reason to act.
What's the campaign's current status — is it already flagged "Limited by budget" or still "Learning"? Lowering the target on a campaign that's already constrained is riskier — the effects will stack.
Google's own rule of thumb is that the learning cycle after a meaningful bid/target change takes about 1–2 weeks (7–14 days), during which performance can swing and won't reflect the strategy's real, settled behavior. But that's a time-based rule of thumb — the real deciding factor is how many conversions you've accumulated since the change (see below).
Look at CPA together with spend, not in isolation — if CPA stayed flat but conversions dropped, spend dropped too. The question is whether the strategy simply "underspent" or whether the budget itself was also cut around the same time.
Break CPA down into its two building blocks: CPC (what you pay per click) and conversion rate (what share of clicks convert). CPA rising because CPC rose while conversion rate held steady points to bidding/auction dynamics. CPA rising because conversion rate dropped while CPC held steady points to traffic quality or the site itself.
How many conversions does the campaign reliably deliver per week/month on Maximize Conversions? At low volume (roughly under 15–30 conversions in 30 days), switching to tCPA risks getting stuck in "Learning" for a long time without a clear payoff.
What's your actual (or expected) target CPA? A sensible minimum daily budget is usually estimated as a multiple of CPA — a common rule of thumb: your daily budget should cover at least 2–4 conversions at target CPA, or you'll accumulate stats too slowly.
Do both flags actually apply to the same time window? Sometimes one status is current and the other is a holdover from an earlier period that hasn't refreshed in the interface yet.
Are these "different conversion types" actually different in substance, or just different paths to the same outcome — for example, a sign-up form, a demo booking, and a phone call that all lead to the same kind of qualified lead?
How many actual changes have been made to the campaign over the last month? Flip-flopping between statuses is almost always a sign of frequent edits (bids, budget, target, conversion actions, structure), not the algorithm "behaving erratically" on its own.
What does Auction Insights show for your core keywords? Overlap rate, outranking share, and impression share relative to competitors give you an indirect read on how aggressively they're bidding — but Google doesn't reveal a competitor's actual CPA.
Is the goal (target CPA and conversion type) really identical across the campaigns you're considering merging? If the targets formally match but the campaigns actually serve different products or audiences, merging could blend inconsistent segments into a single optimization.
Does your business have clear year-over-year seasonality? Compare the current period not just to last month but to the same period last year, to tell a seasonal effect apart from a structural problem.
Is this clustering actually a problem, or just a reflection of real demand? If one segment (geo/device/audience) genuinely accounts for the bulk of paying demand in your niche, concentrating there may be correct algorithm behavior, not a distortion.
Check the campaign's status in the interface — an explicit "Bid strategy constrained by target" or "Limited by search volume" flag, paired with spend running well under budget, is a direct sign the problem is the target, not a lack of demand in the niche.
How long has the campaign been beating target? If it's not a one-off blip but a consistent pattern over several weeks, that's a strong signal of untapped potential, not just statistical noise.
What's the real median (and "tail") gap between click and conversion for this business? Check the Time Lag report in Google Ads (the day-by-day "click to conversion" distribution) — without this number, any judgment about a recent period will be premature.
Is there enough total traffic/conversion volume to split it into two halves (test and control) and still get a statistically meaningful read from each? An experiment needs roughly twice the data to reach the same confidence level as a direct change.
