Why the latest days in your Google Ads account always look bad

The latest days are not bad, they are incomplete

By default, Google Ads assigns conversions to the day of the click, not the day of the purchase, and it reports cost at the time of the click too (Google Ads Help). A click from yesterday is paid for and fully included in the cost. The purchase that this click produces in three days' time is added later, retroactively to yesterday. Until then it is missing. Google describes the consequence itself: because spend is fully reported but the later conversions are still missing, recent days show fewer conversions and a higher cost per conversion (Google Ads Help). Depending on the conversion window you have chosen, conversions can arrive up to 90 days after the click.

This is not a measurement error but booking logic. It becomes expensive when someone looks at the ROAS of the last seven days on a Monday, compares it with the previous month and then raises the target ROAS or cuts budget. The previous month is fully booked, the last week is not. The comparison is skewed before it starts.

So the only question is how many days you leave out. The answer is in your account, not in a flat rule.

How to read your own delay

The report exists in every account: campaign or ad group view, segment Conversions, then Days to conversion. Google splits the conversion columns into up to 19 rows by delay. Choose a date range that ends at least 30 days ago, or longer if your conversion window is longer, so that the data is complete (Google Ads Help).

This is how I evaluate the report:

• Conversion value, not count. In a shop you judge performance by ROAS, so what matters is when the revenue arrives. In my assessment, expensive baskets often take longer than small purchases, in which case the count underestimates the revenue still to come.
• Cumulate. The average is of little help. You need the share that has arrived after x days. To get it, add up the rows from top to bottom and divide by the total.
• Split by campaign type. I expect brand searches to settle faster than generic searches, where buyers are still comparing. An account-wide average hides such differences.

Worked example with invented, round numbers for a shop, cumulative share of conversion value:

At the time you look, this shop sees about two thirds of the revenue that yesterday will eventually bring. For three days ago it is four fifths.

The rule: leave out days until 90% has arrived

My rule of thumb for evaluations: the evaluation window ends on the day from which your cumulative share reaches roughly 90%. In the example that is seven days. If you evaluate on the 20th of a month, you use data up to the 13th. The remaining ten per cent rarely changes a bidding decision, whereas a shortfall of 30 or 40% does.

For reports that have to show current days, you extrapolate instead of leaving days out:

Expected value = reported value ÷ cumulative share for the age of the day

Suppose a click day from three days ago shows €1,250 in cost and €4,000 in conversion value. Reported, that is a ROAS of 3.2. Extrapolated with 80%, it is €5,000 and, because the cost is already complete, a ROAS of 4.0. With a target of 3.5, that is the difference between "throttle" and "let it run".

In practice, every report should carry a "complete from" column per day, derived from your own profile, instead of a blanket waiting time for everything. This fits Google's recommendation to evaluate bid strategies over at least two full conversion cycles, ideally over a month or at least 50 conversions, and to wait one to two cycles after major changes to bids or budget (Google Ads Help). Your 90% value is a workable definition of a cycle. How this relates to the learning period is covered in The learning period in Smart Bidding.

If you use Smart Bidding, there is a quick hint: in the bid strategy report, Google shows the average delay and the estimated number of missing conversions when you hover over the conversion columns, and the chart marks the incomplete period (Google Ads Help). A Google employee advises taking the days marked in the chart out of evaluations (PPC Land). For my own rule I still use the cumulative profile, because an average does not tell you how much revenue is still missing on a particular day.

Click date or conversion time: which column for what

Alongside the standard columns there are "Conversions (by conv. time)" and "Conv. value (by conv. time)". They assign the purchase to the day it took place. According to Google, they are suited to reconciling with your own sales figures (Google Ads Help).

This is the split I recommend:

• Click date for everything that relates cost to return: ROAS, cost per conversion, target ROAS. Only here do spend and return belong to the same click.
• Conversion time for reconciling with the shop back end, for example "How much revenue did Google Ads report for October?" You then compare purchases in October with purchases in October. More on reconciliation in Is your conversion data correct?.

A ROAS built from cost by click date and revenue by conversion time is a hybrid figure. In quiet weeks you hardly notice, in promotion weeks you notice a great deal.

Black Week: the week when both errors happen at once

Many buyers click before Black Friday and Cyber Monday and only buy on the promotion days. By click date, part of the promotion revenue therefore lands on the days before (Metric Theory). Two distortions follow, and they do not cancel out:

1. The lead-up days look better in hindsight than they felt on the day. If you cut in real time, you cut clicks that would have bought later.
2. The promotion days themselves are still incomplete on the Monday after. It is plausible that stragglers from promotion clicks still buy in the days that follow, while discounts run or the basket is still open. If you cut budget on the Monday after Cyber Monday, you save exactly when those purchases arrive.

On top of that, your normal profile does not apply that week. How the delay shifts depends on assortment and promotion, so I would evaluate the profile of last year's promotion week separately, if you have one. A sensible Black Week rule: evaluate the promotion days no earlier than your own 90% value, counted from the last promotion day. Until then, you only decide whether the week as a whole pays off, using the conversion time columns against the shop back end. What needs preparing in advance is in Black Week with Google Ads, and winding things back afterwards in After Black Week.

According to Google, seasonality adjustments are meant for short events of one to seven days and only make sense when the conversion rate changes sharply (Google Ads Help). They do not change the fact that you can only evaluate the week once the purchases are booked.

When the 90% rule does not hold

• Few conversions. With a few dozen purchases a month, the rows in the delay report scatter widely. Use a longer period or the account-level profile, and evaluate only over longer windows anyway.
• Changed conversion window or new conversion action. The old profile no longer fits. Only measure again once enough new data is complete.
• Tracking outage. If conversions suddenly go missing even on days that should long since be complete, that is not delay. In its explanations Google distinguishes between delay as a cause and drops in tag fires (Google Ads Help). A look at older days separates the two quickly.
• Long purchase decisions. For furniture or technology with a high basket value, 90% may only be reached after weeks. Leaving days out is then impractical and extrapolation is the better route, knowing that it is imprecise with small numbers.

Redrawing the profile once a quarter is enough. If you derive a fixed number per campaign type from it and write it into your reporting, you no longer make budget decisions on half-booked days.