Better campaign performance starts with better measurement

Google Ads can only optimize toward the signals it receives. Feed it a clear definition of what creates value and automated bidding gets better at finding it. Feed it noise and it gets efficient at producing noise.

The second case is more common than it looks, because conversion tracking is usually treated as a setup task. Install the tag, confirm it fires, move on. But the conversions you choose to count are not a technical detail. They are the instructions the platform optimizes against.

What Google Ads actually does with what you count

Google Ads does not know which actions matter to your business. It learns that from the conversions you mark as primary. If a form submission, a phone call, a content download, and a page view all count as conversions of equal weight, the platform cannot tell a sales-ready lead from someone who skimmed a page. With no priority to work from, automated bidding does the predictable thing. It chases the cheapest, highest-volume action. You end up with more conversions and fewer customers.

Smart Bidding strategies like target CPA and target ROAS take your primary conversions as the definition of success and spend against it. Every bid the system places is a prediction about the likelihood of one of those actions happening. Mark a low-intent action as primary and the system will happily find you more of it, because you told it that is what winning looks like.

Secondary conversions are the other half of the setting, and most accounts ignore the distinction entirely. A secondary conversion is observed but not optimized against. That makes it the right home for actions worth watching but not worth buying: newsletter signups on an ecommerce account, pageview-based micro-conversions, anything upstream of the action that pays you. The primary and secondary split is not bookkeeping. It is the difference between what the algorithm chases and what it merely records.

The four ways conversion setups go wrong

Nearly every broken conversion layer I audit fails in one of four ways, and most fail in two or three at once.

Everything is marked primary. Purchase, add to cart, phone click, form view, scroll depth, all feeding the bidding algorithm as equals. The system optimizes for the cheapest action in the set, which is never the one that pays you.

The same action is counted twice. A GA4 imported conversion running alongside a native Google Ads tag for the same purchase is the most common version. Reported conversions inflate, cost per conversion looks better than reality, and budget decisions get made on a number that is quietly doubled.

The counting setting contradicts the action. A purchase should count every time; a lead should usually count once per click. Get those backwards and a single enthusiastic form-filler registers as five leads, or repeat purchases from one customer vanish from the record.

There is no value data. Without conversion values, the platform treats every conversion as identical, so a $50 order and a $5,000 order pull bids with exactly the same weight. Value-based bidding is not available to you, and neither is any honest answer about return on spend.

When the algorithm optimizes for the wrong thing

A B2C lead-gen client came to me frustrated that their paid media swung wildly from month to month. The swings pushed them into reactive decisions, pulling and reallocating budget on a few weeks of data, then reversing when the next few weeks looked different.

The audit found the cause fast. They were tracking plenty of useful activity, which is the right instinct. The problem was that every one of those events counted as a conversion. The account had no hierarchy of value, so the campaigns had nothing meaningful to optimize toward and the reporting could not separate a real lead from a soft engagement.

The fix was not more tracking. It was less, used better. I moved most of those events to secondary conversions, where they still inform observation and audience building but no longer drive bidding, and refocused the campaigns on the primary actions that actually tied to business outcomes. The wild monthly swings settled into a trend the client could read, the campaigns started optimizing toward the actions that mattered, and the reactive reallocation stopped.

That was the first move in a larger measurement rebuild. Once the conversion signal was trustworthy, the harder problems were attribution and lifetime value, and solving those is what eventually let them scale acquisition at a stable cost. That is a longer story for its own piece.

How to audit your own setup in an afternoon

You do not need a consultant to find most of this. Open the conversions summary in Google Ads and put it next to a plain-language list of the actions your business actually gets paid for. Every primary conversion should map to something on that list. Anything that does not is a candidate for secondary status or deletion.

Then check for duplicates. For each real-world action, there should be exactly one conversion source counting it. If a purchase appears as both a native tag and a GA4 import, one of them goes.

Third, read the counting and attribution settings on each surviving action. Every-versus-one should match the economics of the action. Conversion windows should be sane for your sales cycle; a 30-day click window on a same-session impulse purchase tells the algorithm to take credit for sales it did not cause.

Last, confirm values are flowing. Ecommerce accounts should see real order values, not a static placeholder. Lead gen accounts should at least assign differentiated values by lead type, even rough ones, because a rough ranking beats treating everything as equal.

If that afternoon turns up problems you are not sure how to sequence, that is the point where the work stops being a checklist. Which actions deserve primary status, how to restructure without wiping bidding history, and what the reporting should look like afterward are judgment calls, and getting them wrong resets the algorithm's learning for weeks.

What good looks like

A healthy conversion layer is small, deliberate, and boring. A handful of primary actions that map one to one to revenue, secondaries for context, one source of truth per action, values flowing on all of them. When the setup looks like that, the platform's optimization and your reporting finally describe the same business.

The takeaway is easy to state and easy to get wrong. Counting everything as a conversion is close to counting nothing. The work is deciding which actions represent real value, marking only those as primary, and keeping the rest as secondary signals you can learn from without letting them steer your budget.

When your bidding and your reporting disagree about what is working, the conversion layer is usually the first place I look.