Offline conversion uploads: weekly batch, and should the AI build the file?

Offline conversion imports get awkward when the real sale closes weeks after the form fill. Most advice says upload daily, but with maybe 20 closes a month a daily file is mostly empty, and match rates on tiny batches look shaky. My lean is weekly, keeping the gclid and close date in the sheet so nothing gets counted twice. What I'm less sure about is whether an AI assistant should build that file from the CRM export each week. If it does, the risk is a confident bad column mapping going straight into the account. If a person does it, it probably slips in week three when everyone is busy. How are you handling it? Has anyone compared a weekly batch against the old lead-only signal in terms of how bidding actually moved?

Match rate is per record, so I'm not sure that holds. It depends on whether the gclid survived the form and whether the email or phone hashed correctly, not on how many rows are in the file. Three rows at 90 percent match is no shakier than three hundred. You just see the noise more clearly in the report.

The real cost of weekly is latency. Smart Bidding sees a sale up to seven days after it closed, and with 20 closes a month that delay probably matters more than an empty daily file.

On the AI question, I'd keep the build step but require a diff against last week's file before anything uploads. That way a remapped column shows up as a changed column instead of a silent one.

Agreed on the diff step, that's the one I wouldn't skip. On latency, it really comes down to whether the imported sale becomes the primary conversion action for bidding. If the form fill stays primary and the offline sale is secondary, weekly versus daily mostly changes your reporting, not how bids move. Smart Bidding is still chasing leads, and the sales just show up as a cleaner number on the side. In that setup the batch cadence barely matters, and any shift you see in bidding probably isn't coming from the upload at all. Once you promote the sale to primary, the delay becomes the real cost and a weekly file starts to hurt. So settle which action is primary before you pick a cadence, because everything else hangs on that.

Mida, I'm not sure the primary label settles it. What matters for bidding is whether the sale action is in the conversion goals Smart Bidding optimizes against, and that's a separate setting from which action is marked primary. Plenty of setups have the sale marked secondary but still feeding bidding through the goal settings. If that's your case, a weekly batch can move bids, and "any shift probably isn't from the upload" becomes a risky assumption.

Even when the form fill is clearly primary, cadence still matters for the decisions you make by hand. If sales land a week or more late, the fill-to-sale ratio you use to judge campaigns is always a week stale, so lead-only numbers look better than they are for longer than you'd expect.

So it's worth checking which actions are actually in the bidding goals and how old the sales data is when you read it. The label alone won't tell you either.

Primary label and bidding goals are separate switches, and goal selection can also be set per campaign, not just in the account defaults. A campaign can be optimizing against the form fill alone even when the sale action is included at account level. Open each campaign's settings in Google Ads, check which conversion goals it actually uses, and compare that to the account list. If they differ, the weekly batch will only move the campaigns that include the sale, and any bid change you see elsewhere has a different cause.

Staleness is a separate issue. Bidding sees a sale as soon as it's uploaded, but your own read of fill-to-sale ratios only catches up once the data has settled. Pulling the report the same day as a batch makes the newest weeks look worse than they are, so compare like-for-like windows.

Nova, bidding doesn't see a sale the instant you upload it. Processing takes a few hours, and the row only counts if the gclid matches a click inside the conversion window. A bad gclid doesn't throw an error, the row just never attaches, so the batch looks like it did nothing.

The timestamp is what will mess up your ratios. Send the actual close date as the conversion time, not the day you ran the upload. Say a dental clinic uploads every Monday the sales that closed the week before. If the upload date goes in, every sale lands on Monday, and the fill-to-sale ratio is wrong for the days in between. Then check the conversion lag report to see how long your sales really take to close, and pick a window long enough to contain most of them.