Dayparting with Smart Bidding: cut the night hours or let the bidding handle them?

Say you've got a lead gen account on Target CPA with the schedule wide open, and someone wants to turn off nights and weekends because the cost per lead looks worse there. I'd push back. Smart Bidding already bids down in hours where conversion rates are weaker, so cutting the schedule removes auctions it was already handling cheaply, and the blended CPA can look worse afterwards for no real reason. Dayparting earns its place with manual CPC, or when the daily budget runs out by lunchtime and the best hours get starved. Before cutting anything, check conversion rate by hour and whether budget is actually the limiting factor. Has anyone found hour-of-day data more useful for pacing the budget than for cutting hours?

I don't buy that Smart Bidding makes the schedule decision for you. Target CPA optimizes for whatever the conversion action counts, and a form fill at 2 a.m. counts exactly the same as one at 2 p.m. If the sales team doesn't pick up until morning, those night leads look cheap on paper but are worthless in the pipeline, and the bidding has no way of knowing that.

So pull the hour-of-day data for qualified leads or closed deals, not just raw conversions. If night leads convert to sales at a fraction of the daytime rate, cutting those hours is the right call even on Target CPA.

The bidding can learn this if you tell it. Import qualified leads or closed deals as offline conversions, ideally with their own values, and Target CPA optimizes toward those instead of raw form fills. That way a 2 a.m. lead that never qualifies stops looking cheap to it. Cutting hours is the blunt fix, and it also throws away the night auctions that do turn into sales.

The catch is lag. If sales takes a week to qualify a lead, the feedback arrives late, so check how long your cycle is before you trust the import.

Agreed on the lag. I'd make it a rule: don't cut an hour block until its offline data covers a few weeks of closed deals. Before that the sample is mostly noise, and cutting the hours removes the very auctions that would have given you the answer.

If the gap looks real but you don't trust it yet, a bid modifier of around -20% on the weak blocks is a safer middle step than switching them off. You stay in the auction, Smart Bidding keeps learning from whatever comes in, and you can move the modifier back toward zero if the qualified-lead rate for those hours recovers. Full cuts make more sense once the gap is clear and holds across several weeks, not just one bad stretch.

I wouldn't use the -20% modifier as the middle step. On Target CPA the bidder already sets each auction's bid from its own conversion-rate prediction for that hour, so a flat multiplier just sits on top of that without the bidder knowing why it's there. Then when cost per qualified lead moves, you can't tell whether the modifier, the sales lag or normal variance caused it. If the data isn't clear enough to justify cutting the hours, leave the schedule alone and fix the signal first. Import qualified leads with values, and the bidder will learn the night gap on its own.

I don't think a modifier makes the result unreadable. Yes, it sits on top of the bidder's prediction, but that's exactly why you can test it. Change only the schedule modifier, log the date, and leave budgets, bids and new ads alone for that stretch. Then the before and after windows actually tell you something. The real problem is changing five things in the same week, and that's an argument for being disciplined about changes, not for skipping the test.

The test itself is fine, but I'd measure it differently. If sales takes two weeks to qualify a lead, the "after" window is mostly unqualified leads, so cost per lead can look better or worse for reasons that have nothing to do with the modifier. Run it for at least your full sales cycle plus a week, and compare qualified leads, or qualified-lead cost per hour block, instead of raw CPA. Also keep the offline import rules unchanged during the test, otherwise you'll end up reading changes in how leads get marked as the modifier's effect.

Agreed on the cycle length. Before you trust any hour-level qualified-lead number, check the gclid match rate by hour. If the form only passes the click ID through on a clean page load, or a consent banner blocks the cookie, leads from some hours can drop out of the offline upload entirely. Night traffic often has a different mix, with more in-app browsers and more blockers, so the match rate itself can swing by hour. That means a "cheap but unqualified" night gap might really be a missing-gclid gap, and you'd end up cutting hours based on incomplete data. Compare match rates across blocks over the same window, and if they're uneven, fix the capture first. Enhanced conversions can recover some matches when the click ID is gone.