Google Ads Conversion Tracking: the inputs that matter

Smart bidding only optimises against the conversion data flowing into it. Poor match rates on first-party data, data loss on the conversion event, signal mismatched to the campaign’s objective: each one poisons the bidding model and caps performance regardless of what strategy is layered on top. The accounts that get the most out of Google Ads’ smart bidding are not the ones with the cleverest bid strategy. They are the ones whose data layer is feeding the algorithm signal worth bidding against.

If your account is running a product feed pulled directly from Shopify, a conversion tag firing on the order confirmation page, and a customer list uploaded once into Google Ads, your setup is identical to almost every other ecommerce account in the auction. That is the floor of conversion tracking, not a competitive position. What gets accounts past the floor is data discipline operating underneath the tag: a CRM broken into multiple audience segments instead of one master list, lifetime value calculated per customer to inform how those segments are built, value rules adjusting bids for new versus lapsed customers even though both fire the same tag. The data the algorithm sees on accounts doing this is materially richer than the data it sees on competitors, and the auction outcomes follow.

The reason this matters more now than it did three years ago is that the platforms changed underneath operators. Privacy legislation tightened through GDPR and iOS 14, and platform-side tracking capability fell with it. Across 2021 and 2022 specifically, average CPAs on Meta doubled across our portfolio. Google and Meta both responded by moving away from tight granular targeting and leaning more heavily on modelling and machine learning, both of which require large clean data sets to perform. The accounts that built their data layer to feed that change kept scaling. The ones that didn’t are hitting ceilings that look like creative or bidding problems but trace back to what the conversion tag is sending.

Primary vs secondary conversion actions

The first decision the bidding model sees is which conversion actions you have told it to bid against. On an ecommerce account, only one event belongs in Primary: a single purchase conversion. Everything else the site fires gets tracked as a secondary signal, which covers page views, add-to-cart, form submissions, and any other on-site event the operator wants visibility into.

Secondary tracking is doing real work. Funnel events feed remarketing audiences and give operators a read on where customers are dropping off before purchase. Those are legitimate uses for the data, and they live on the audiences side of the system. The bidding model should be working off the purchase signal alone. A campaign asked to optimise for multiple objectives at once gives the algorithm conflicting instructions instead of one clear one to bid against.

Should add-to-cart be a primary conversion?

This is the question that comes up on every account. The answer is no. Add-to-cart belongs in Secondary, the same as page views and form submissions. Track it so you can build remarketing audiences off it and so it shows up in reporting. Promoting it to Primary tells the algorithm to treat add-to-cart events as equivalent to purchases, and the bidding model loses the clean purchase signal it needs to bid toward actual business outcomes.

Multiple primary conversions and confused signal

When two or three different events all get marked Primary in the same account, the bidding model is being asked to optimise for several things at once instead of one. The clean purchase signal smart bidding needs to perform against gets lost in the mix.

First-party customer segments

If Primary versus Secondary tells the bidding model which event to bid for, customer segments tell it which audience profile to bid against. First-party data is the strongest signal an operator can feed the algorithm. It is unique to the business, drawn from people who have already completed an action on the site, and it tells Google’s bidding model what an actual customer on this account looks like at the front of the auction.

The customer segments to import:

• Newsletter subscribers
• Loyalty programme members
• Purchasers split by value
• Purchasers split by timeframe (recent versus lapsed)
• Purchasers split by frequency
• Purchasers split by category

Where analytics depth allows it, basket abandoners and demographic splits across age, gender, or location add another layer of granularity. The more discrete segments the algorithm can see, the more accurately it can bid against the customer profile that actually converts on this specific account, and the less weight it puts on generic prospect lookalikes built off lower-intent audiences.

Search-side activation needs a minimum of 1,000 users per list. Anything below that is invisible to the campaign, so smaller lists should be consolidated until they cross the threshold, rather than split further to look more granular in the interface.

Conversion value rules: bidding higher for new and lapsed customers

In the bidding model, a new-customer purchase and a repeat purchase fire the same tag. What separates them at the auction is the conversion value rule applied to the event. This is the third layer of data discipline, and the feature inside Google Ads where it lives is Conversion Value Rules.

Google’s Original Conv Value column shows the raw purchase value separately from any proxy adjustments the rules are applying. That separation is what makes it possible to verify the rules are actually changing the bidding the way they were configured to.

The adjustments worth setting up on an ecommerce account:

• Bid higher for new customers
• Bid higher for high-value new customers
• Bid higher for lapsed customers
• Bid higher for high-value lapsed customers

What these do is tell the bidding model to weight conversion types unequally. New-customer acquisition is the growth bottleneck on most ecommerce accounts, so a new-customer purchase carries more strategic weight than a repeat one even though the platform sees both as the same tag firing. Lapsed high-value customers earn weight on a different basis: they have a demonstrated spend pattern, and bringing one back is usually cheaper than acquiring a comparable prospect from cold. Value rules encode this hierarchy at the bidding layer, so the campaign doesn’t have to be split into separate new-customer and BAU shells to express it.

Value rules depend on the customer segments in the previous section. The 1,000-user Search activation threshold applies to those segments, so the audience work is the precondition for the value-rules work.

Attribution lag and short reporting windows

The fourth piece of data discipline is the gap between what’s actually happening and what the bidding model is being shown. Conversions don’t all attribute back to the click that drove them on the same day, or even the same week. Google’s in-platform reporting attributes conversions as they arrive, which means any short reporting window is by construction looking at a partial picture of what the campaign is actually delivering.

The platform doesn’t surface the lag directly, so operators have to track it themselves. The number a week after a bidding change is not the number that same change will be producing a month later, and accounts reacting to short-window reports are usually reading noise.

This is the failure mode behind almost every panicked “the campaign stopped working” message. The conversions are still being attributed back to the campaign; they just haven’t been reported yet. A separate tracker that captures how conversion volume on each major campaign matures across the 30-day window after the conversion event is the practical fix, and it gives the operator a real picture of what is moving instead of what is being shown.

The data-layer audit

The five checks below run in order, because each one determines what smart bidding can see at the next level:

1. Primary conversion action. A single purchase action is the only Primary on the account. Nothing else in the audit matters until this is true.
2. Funnel events tracked as Secondary. Page views, add-to-cart, and form submissions are firing and tagged Secondary, so remarketing audiences can be built off them without polluting the bidding signal.
3. Customer segments live and above threshold. Newsletter, loyalty, and purchaser segments split by value, timeframe, frequency, and category are imported, with each list above the 1,000-user activation minimum on Search.
4. Value rules configured and verifiable. New-customer and lapsed-customer adjustments are running on top of the active segments, and the Original Conv Value column confirms the adjustments are being applied.
5. Attribution lag tracked externally. A separate tracker captures how conversion volume on each major campaign matures across the 30-day window after the conversion event.

None of these is replaceable by tuning at the campaign level.

FAQ