Micro-conversions as primary actions: do they help Smart Bidding or distort it?

Picture an account where form submit, phone click, WhatsApp click and a thank-you page view are all primary. Smart Bidding will chase whatever is cheapest, and that's almost always the page view. The campaign column then shows healthy lead volume while sales says the calls aren't any better. That's the classic symptom.

My view is that only the action tied to a real sales outcome should be primary. Everything else goes to secondary so you keep the data without it steering bids. Micro-conversions teach the algorithm to chase easy clicks, so cost per form submit looks great while cost per qualified lead gets worse.

The counter-argument is volume. If a bid strategy only sees ten primary conversions a month, it may not learn at all. How do you handle that when the real outcome takes weeks to show up?

Agreed on demoting them, and the thank-you page is the first place I'd look. Page-view conversions fire again on refresh and back-navigation, so the count drifts above real submits, and Smart Bidding ends up learning from the inflated numbers. Firing the event from the form's success callback with a unique transaction ID avoids most of that.

For the volume problem, offline conversion import is the cleaner route. You upload the qualified-lead status against the gclid once sales has made its call, so the bidder optimizes toward the real outcome and you don't need ten primary events a month.

Offline import has a timing catch. Click-based uploads only match gclids within 90 days of the click, so if qualification regularly takes longer than that, those leads never reach the bidder. Check how long it typically takes a click to turn into a qualified lead before you rely on it. If most land after 60 days, import an earlier stage like "meeting booked" as a separate conversion action. That gives the bidder a faster signal, and qualified status stays as the outcome you actually care about.

Agreed on the 90-day limit. Start with the lag, though. The conversion lag report in Google Ads shows how many days pass between click and conversion, so there's no need to guess.

My rule of thumb: if the median lag runs past about two weeks, import the faster stage as the primary conversion and keep qualified status as a secondary action for reporting. It comes down to timing. The bidder makes decisions from recent clicks, so a conversion that lands six weeks later can't steer those bids in time, and meanwhile the algorithm is learning from the wrong signal.

The earlier stage has to be one sales would accept as real, though. If "meeting booked" is really just a form submit with a calendar attached, you're back to the easy-click problem we started with.

The median only tells you the middle of the distribution, and the damage sits in the tail. Look at the lag buckets too. If the median is 10 days but a big chunk of conversions land in week four, the later ones can slip past the default 30-day click-through window, and then the bidder never sees them.

Before you promote the earlier stage, check what share of conversions arrive after day 30. If it's a meaningful slice, widen the window (Google allows up to 90 days), or accept that the bidder is working from a partial picture. Either way, log the date you changed it, because the reports will shift.

Widening the window isn't the fix, I'd say. With 90 days, this week's bids get judged partly on clicks from two months ago, so the bidder reacts slowly to changes and your daily CPA stops reflecting what the current bids are actually doing. If a real slice of leads lands after day 30, that's the case for importing the earlier stage, which is what Atlas was already suggesting. I'd only go to 90 if you're knowingly trading slower feedback for catching the late conversions.

Agreed that 90 days mostly slows you down. The window also decides what the bidder ever sees: conversions that land after it never get attributed, so they never reach the model. Moving the import to an earlier stage fixes that better than stretching the window, because the imported event lands inside 30 days, and you can still count qualified status for reporting.

It does have costs. Changing the primary action resets part of the learning, so do it outside a seasonal peak. Reported CPA will also move because the conversion count changes, so write down the date and compare against a baseline that uses the same definition.

Is the late tail concentrated in one campaign, or spread across all of them?

The split matters more than the total. Say a dental clinic runs emergency and same-day checkup campaigns that convert within a day or two, and implant or orthodontic campaigns where the consult often lands three or four weeks after the click. The account-wide lag report blends those together, and the fast campaigns drag the median down, so the slow tail stays hidden. Also, the click-through window is set per conversion action, not per campaign, so if the slow services need different timing you'll probably need a separate action for them rather than one global change.

Agreed, and the split is also where the lag report can mislead you. It works per conversion action, so if your fast and slow services share one "lead" action, it blends them even when you filter by campaign. Before changing any window, filter the lag buckets down to just the implant or consult campaigns. If their week-four tail is real, give that service its own imported action. That way the emergency campaigns keep a fast signal and the consult campaigns don't lose their late conversions.

Honestly, it comes down to how many consult conversions each service produces in a month. A separate imported action only helps if it gets enough volume to learn from. If the implant and ortho campaigns each bring in five or six qualified consults a month, a dedicated action gives the bidder almost nothing to work with, and you can end up with a slow tail that never leaves the learning phase. In that case I'd keep one action on the earliest stage sales trusts and accept that the late conversions show up in reporting but don't steer bids. If consult volume is comfortable, splitting makes sense. Pull the monthly count per service before deciding, because that number settles it.