Days to conversion: how I project Google Ads ROAS

Every Monday, someone looks at last week next to the week before and decides last week was worse. A lot of the time it wasn't. It just hasn't finished reporting yet.

I've tracked this by hand, month by month, for years. Google's Days to conversion segment gets me to the same place, and I can build it from whichever season I need. I think it's one of the most underrated reports in Google Ads.

Why recent days always look worse

Google Ads' main conversion columns report conversions on the day of the click, not the day of the conversion. Google's example: if your ad was clicked last week and that person converted this week, "both the click and the conversion are reported back to last week." That keeps spend and conversions on the same date, which is what makes CPA and ROAS line up.

The catch is that last week's spend is fully counted while some of last week's conversions haven't happened yet. Google says so directly.

Depending on your conversion window, conversions can keep coming in for up to 90 days after the click. So a day from last week and a day from a month ago aren't at the same stage. Comparing them as if they were is how good weeks get called bad ones.

Where to find Days to conversion

On the Campaigns page, click Segment, then Conversions, then Days to conversion. Google says it splits your conversion columns into up to 19 rows, one for each range of days between the ad and the conversion. The rows run from less than one day out to 90 days, the longest conversion window Google allows.

Google gives one instruction worth following exactly: make sure the date range "ends at least 30 days ago (or longer if you have a longer conversion window)." If the range is too recent, the curve you build from it is incomplete too.

Turn on Conv. value next to Conversions. If you manage to ROAS, value is the number you're projecting.

Build your account's curve

Take a fully reported period and add up the share of conversion value in each row, from the shortest lag to the longest. That running total is your curve: how much of a day's final value is usually in by the time it's one day old, seven days old, fourteen days old, and so on.

Here's a made-up example to show the shape:

To project a recent day, divide what's reported so far by the share you'd expect at its age. A day that's seven days old showing $8,400 in conversion value would land around $10,000 ($8,400 ÷ 0.84).

This is where it gets confusing, so take it slowly. It isn't one static number. Something from last week is less mature than something from two weeks ago, so every day gets its own adjustment based on how old it is. That's the point: one flat percentage across the whole week would get you a rougher estimate.

Make week over week and year over year fair

I use this most for week-over-week checks and month pacing. Take an account where a big share of conversions arrive in the seven days after the week ends. Its week-over-week numbers will always look poor on Monday, because the week before has had seven more days to fill in.

Here's a made-up Monday:

What looked like a drop of half a point is really about flat. The same goes for year over year. Last year's version of this week has had a full year to report. This year's has had a few days. Project this year's numbers first, then compare.

For the curve itself, I use a longer lookback. For peak season or Q4, I build it from last year's peak season, because in my experience the lag is shorter then, or I use an aggregate of the last year. Either way, I'd rather under-deliver on a projection than over-promise.

Set the month-end target backward from the goal

Say the client's goal is a 4.0 ROAS. On the first day of the next month, the last day of the month is a day old and the first day is about a month old. Weight each day's maturity by its spend, and in this made-up account the month is about 89% reported when it closes.

So the number to hit at month end isn't 4.0. It's about 3.6. If the month closes at 3.6 and the curve holds, it should keep climbing toward 4.0 as late conversions come in.

The opposite case matters too. If you hold the account to 4.0 on the last day of the month, it will likely end up above the goal once everything reports. Don't shortchange it by judging it on numbers that haven't matured.

How close the projections get

A projection is a projection. Don't assume that's what you'll get. But I think it's the most educated estimate you can make, because it comes from your own account's data.

I go back to the numbers I projected and check them once the month has fully reported. In my experience they've usually landed within 0.3 ROAS points, and sometimes within 0.03 (I know, specific).

Google applies the same caution to its own tools

Smart Bidding already adjusts for this. Google says its algorithms lean on recent data "while also accounting for the length of your conversion cycle," which "helps prevent overreactions to recent clicks that are experiencing conversion delays." Google's own guidance on judging performance follows the same logic:

• When setting a target, Google's example is to look at your historical CPA or ROAS over a 28-day period, leaving out the past 2 days, if you have a 2-day conversion delay.
• For value-based bidding experiments, Google says to exclude recent days "where less than 90% of your conversions have been reported."

Where it helps most

Every business type can use this. E-commerce is the easiest place to start, but that's partly because e-commerce is what I prefer to work on.

It also helps manage expectations. I use it to have a better conversation about performance with hard numbers: actual next to projected, and why that's the projection. It comes from their own account, not out of thin air.