Why Today Always Looks Worse Than Yesterday: Attribution Lag and Fake Weekday Patterns

There is a specific complaint that surfaces in every media buying community, every week, in slightly different words. Mondays have tanked three weeks running. Performance collapses every afternoon and recovers overnight. The first few days of the month are always terrible. No campaign changes, no website changes, same creative, same budget. Before you go looking for a cause, it is worth understanding a mechanic that manufactures exactly these patterns out of nothing.

Conversions are credited backwards, not forwards

This is the whole thing, and almost every confusing weekday pattern follows from it.

When someone clicks your ad on Saturday and buys on Tuesday, most ad platforms credit that conversion to Saturday, the day of the click, not Tuesday, the day of the purchase. The platform reaches back into a day that already looked finished and adds a sale to it.

The consequence is that any given day keeps improving for as long as its attribution window stays open. A day viewed on a 7-day click setting is not finished for a week. What you see today is the partial version.

So when you compare today against yesterday, you are not comparing two equivalent things. You are comparing a day with almost no backfill against a day that has had a full extra day to collect it. Today loses that comparison every single time, regardless of how the business is actually doing.

How this manufactures a Monday problem

Now put a weekend in the middle and the pattern assembles itself.

By Monday afternoon, Saturday and Sunday have had one to two days of backfill. Clicks from Friday and the weekend that converted later have landed on those weekend dates and made them look strong. Meanwhile Monday is a few hours old with essentially no backfill at all.

You open the report at 3pm on Monday. The weekend shows a healthy return. Monday shows one sale against thirteen at the same time yesterday. It genuinely looks like something broke overnight.

Come back on Wednesday and look at that same Monday. It will have filled in. Not always to match the weekend, because real weekday demand differences exist, but the catastrophe usually turns out to be a dent.

This is also why the pattern repeats weekly and survives every intervention. People change something on Monday, Monday backfills by Wednesday, and the change gets the credit. Next Monday it happens again.

How to tell a reporting artifact from a real problem

The test is fast and it does not depend on trusting the platform at all. Go to the system that records money.

1. Pull actual orders for the day in question from your own backend. Shopify, your payment processor, your CRM. This number does not backfill, because an order either happened or it did not.
2. Compare that to the same day last week. Not to yesterday. Monday against Monday. Weekday effects are real and comparing across different days of the week creates confusion that has nothing to do with attribution.
3. Decide from the backend number, not the platform number. If your backend says you genuinely took one order today against fifteen last Monday, you have a real problem and it is worth investigating. If the backend says you took twelve orders and the platform is showing one, nothing is wrong with your business. You are looking at attribution that has not caught up.

That distinction, between the platform's view and the reality, is the single most useful habit in ad measurement, and it comes up repeatedly in ad platform revenue versus actual revenue.

The rule that saves the most wasted work

Do not make decisions on data younger than your attribution window.

If you optimise on a 7-day click window, the last seven days are incomplete by definition. Judging a campaign on them is judging a partially reported result and concluding something confident from it.

In practice most teams settle on something like this. Watch the last few days for obvious breakage, such as delivery stopping entirely or spend running with literally zero events. Make actual budget and structure decisions on data that is at least as old as your window. Report to clients on complete periods only, and say plainly that recent days are still filling in.

There is a real cost to ignoring this beyond confusion. Every campaign change resets learning. If you are restructuring weekly in response to a Monday that was always going to fill in by Wednesday, you are paying the learning cost over and over for a problem that does not exist, and the instability you create then becomes a genuine performance problem. That is covered in more detail in what to do about learning limited.

A worked example of the Monday panic

A store runs a 7-day click window. Over a settled week, each day ends up with roughly 14 orders. But a day does not arrive at 14 immediately: it collects them over the following week, with most landing in the first two days and a tail after that.

Say a typical day looks like this as it fills in: about 6 on the day itself, 11 by the next day, 13 by day three, 14 by day seven.

Now it is 3pm on Monday and you open the report.

• Saturday is two days old and shows 13.
• Sunday is one day old and shows 11.
• Monday is six hours old and shows 2.

That reads as a collapse. Sunday to Monday looks like an 80% fall. In reality all three days are heading for 14, and by Thursday the week will look flat.

Now notice what happens if you react. You restructure on Monday afternoon. By Wednesday, Monday has filled in to 13 and the account looks recovered. The restructure gets the credit, the belief that Mondays need intervention is reinforced, and you have paid a learning reset for nothing. Do that weekly and the instability you introduce becomes a real performance problem, at which point the reports will finally show something genuinely wrong.

Platforms differ, and so do your two accounts

The mechanic is general but the shape of the curve is not, which matters if you run Meta and Google side by side.

Backfill depends on your window setting, the platform's default reporting behaviour, and above all how long your customers take to decide. A 1-day click window settles almost immediately and produces very little of this effect. A 7-day window on a considered purchase produces a lot. A Google Ads account with a 30-day conversion window can still be gaining conversions for a month.

So two accounts in the same business can behave completely differently, and comparing "yesterday" across them is not meaningful. It also means a single blanket rule like "ignore the last three days" is right for one account and wrong for another. Measure your own curve, per platform, using the exercise at the end of this article.

The other patterns this explains

• "Performance dies every afternoon." Intraday reporting is the most incomplete data that exists. A day at 2pm has barely started collecting.
• "The start of every month is terrible." A month-to-date figure on the 2nd contains two days of unfilled attribution, compared against a previous month that is fully settled.
• "My numbers changed after I screenshotted them." They did, and that is expected. Reported figures for a given day are provisional until the window closes. Anyone reporting to clients should know this before someone else points it out.
• "The weekend is always our best period." Sometimes true, and worth checking against the backend. Weekends get a full working week of backfill before anyone looks closely at them on Monday, which flatters them.

When it genuinely is not attribution

This mechanic explains a great deal, and it would be a bad outcome if it became the thing you reach for every time a number disappoints. Some signs that you are looking at something real.

• Backend orders dropped too. This is the decisive one. Attribution lag cannot hide orders that your own system recorded.
• The old days never filled in. Go back and look at last Monday now. If it still looks terrible a week later, the window has closed and that was a genuine result.
• Event volume fell off a cliff. If the count of events arriving at the platform collapsed rather than the conversions attributed to recent days, that is a tracking failure, not a lag. The diagnostics are in conversions dropped but spend didn't.
• It coincides with a deploy. Attribution lag does not care what you shipped. If the pattern started the day after a site change, suspect the site change, and see what breaks tracking after a migration.

The five minutes that ends this permanently

Pick a normal week and, for each day, write down two numbers: what the platform reported for that day at the time you looked, and what the platform reports for that day a week later once the window has closed.

The gap between those two columns is your backfill curve. Most accounts find that a day gains somewhere between a fifth and half of its final conversions after the first 24 hours, depending on the window setting and how long the purchase decision takes.

Once you know your own curve, today's number stops being alarming, because you know roughly what it will become. You have converted a recurring weekly panic into a quantity you can reason about, and you will never again restructure a campaign on a Monday afternoon over a number that was always going to fix itself by Wednesday.