I spend a good part of every month looking at accounts where the last two weeks of the month report worse than the two weeks before them, with a client on the call asking Idris what has broken in the account. The thing is nothing broke; the conversions are still in transit. Google attributes a conversion to the date of the click, so the most recent window in your report is always the emptiest one, and it always fills in later.
We walked through that mechanic in an earlier post: what conversion lag is, why it happens, how to read around it. The takeaway most people leave with is patience and better reporting habits.
We wrote that article in 2025, and the advice still holds up. It just stops short of the part that costs money, which is what Smart Bidding is doing while you sit there being patient. While you are waiting for the data to finish arriving, the bid strategy is already acting on the most recent available data.
Smart Bidding reads the same half-empty window you do
Smart Bidding adjusts continuously, on whatever conversion signal is present at that moment. There is no end-of-window batch where it steps back and reconciles. In an account with meaningful lag, that means the newest click cohorts always look like the weakest ones, because a chunk of their conversions has not landed yet.
The system treats that shortfall as performance information and bids accordingly. Then the lagged conversions arrive, the window fills in, and the strategy has already spent a week or two pulling back on auctions it was winning. The correction happens first, and the evidence arrives second.
In the B2B lead gen accounts, three to four weeks between click and a recorded conversion is ordinary. However, I do not want you to borrow that number as a benchmark for yourself. Look at your own Days to Conversion report and have the data tell you what the benchmark is; the whole point is that your lag is a property of your sales cycle and nobody else’s.
When the window is shorter than the cycle, the gap stops being temporary
Lag is survivable when the data eventually arrives. It stops being survivable when your conversion window is set shorter than your actual conversion cycle.
Take a B2B SaaS account with a 30-day average close cycle and a 7-day conversion window. In this scenario, conversions that land on day 12 never enter the learning loop, not because they’re “late”; it’s because they’re outside of the parameters. Google’s default click-through window for conversion actions is 30 days, so the accounts that get screwed with the misalignment are the ones that were set up by someone who shortened the window, inherited it from a previous agency, or, potentially, forgot about it while the sales cycle stretched past a month.
If we want Smart Bidding to deliver us great results, we need to ensure it’s calibrating on conversion history it can see. If that history is half-complete, it will optimize toward that signal confidently because it has no way of knowing what a real conversion looks like to your business.
The optimization toward the half-baked signal is what keeps account dashboards looking steady while the business feels a slowdown in leads and closed/won business. Aligning the conversion window to the real sales cycle is a high-impact change that is readily available in lead generation accounts.
Where modeled conversions come in
When direct measurement is thin, Google fills the gap with modeled conversions: estimates built from observable patterns in users who never triggered a trackable event. Consent gaps, mismatched attribution windows, or, even, when a recent window where we’re just waiting for the conversion data to arrive produces thinness in that data.
Modeled conversions sit in your reporting next to measured ones, with no visual distinction in most views. So the practical situation is that your bid strategy is drawing on a blend of measured and estimated outcomes, and you have no clean read on the ratio.
A note: Google has not published documentation of exactly how modeled conversion data feeds Smart Bidding’s real-time learning. What is documented is that Smart Bidding uses available conversion data, and that modeled conversions are part of that dataset.
A reasonable inference from how the system is described is that the estimates and measurements are weighted together rather than kept in separate buckets. So, if your dashboard conversion volume won’t reconcile with your CRM and your bid strategy looks placid, that gap is worth an afternoon.
The corrections are at the input layer
Treat lag as a reporting problem, and get your answers from different columns, longer lookback windows, and a slower optimization cadence.
What changes what the system learns:
Align the conversion window to the real sales cycle. If the average sale takes 21 days, the window needs to be 30 days. Check what yours is actually set to rather than assuming it was ever set deliberately.
Import offline conversions. Where lead quality is the thing that matters, which is most B2B, a form fill is a weak proxy. Connecting the CRM so qualified leads, opportunities, or closed deals come back in gives Smart Bidding something worth optimizing toward, and something the modeling gaps don’t touch in the same way.
Use data exclusions when tracking breaks. A tagging error or a consent change produces a burst of corrupted signal. Data exclusions tell the system to disregard that period rather than learn from it.
Read the Days to Conversion report before you touch bid strategy. If the last 14 days of your window sit inside your typical lag, any performance read from that window is incomplete by definition, and adjusting bids against it is adjusting against noise.
Lag gets treated as a reason to read your data differently. It is more useful as a signal quality problem, and signal quality is the whole substrate your bid strategy is built on. Patience is getting the inputs right before the system starts learning from them.