Hi Reader,

If you run Google Ads with smart bidding, the accuracy of your conversion count will make or break your campaigns. Think about it... you give Google a target cost per lead, which is the most you’re willing to pay, on average, to win one lead. Google then raises or lowers every bid to hit that target. It judges its progress by counting your conversions, meaning the leads that it believes your ads produced. Google never questions that count. That means a wrong conversion count will, by default, steer every bid that Google places, even when your ads themselves are fine.

I recently saw a wrong conversion count at work on a live video call with an enterprise AI client. While we talked, I used Synter to evaluate conversion tracking of my client’s Google Ads account. Synter is an agent platform for paid media that Joel Horwitz built after he ran growth for Amp at Sourcegraph.

I asked Synter to audit how the account’s conversion tracking had been set up across the last 2 years of campaign data.

About 3 minutes later, Synter gave me its answer. My client’s ads were fine. Unfortunately, the conversion count that Google had been optimizing toward for 13 months was categorically wrong.

Google Ads had recorded 755 conversions. My client got zero leads.

Over 95% of those recorded conversions came from a single lead conversion action that nobody had ever checked against a real lead. That action was even set to count repeat form submissions from the same click as brand new leads. With that setting in place, my client’s strongest campaign showed a cost of around $12 per lead. Synter put it plainly: “the number proving it’s good is the least trustworthy number in the account.”

The account spent ~$15k over those 13 months. Every dollar of it went toward winning more conversions that never turned into a single lead.

I've generated $7M+ in attributed revenue across product launches for tech startups of all shapes and sizes. Fact of the matter is that attributed revenue holds up only when each lead is counted once, where attribution traces every sale back to the lead that produced it. My client's $12 CPA failed that test. The conversion action behind it counted repeat submissions as new leads. That error carried into every number that was built on top of that conversion count.

Here are the numbers that Synter's read-only audit pulled from my client's account.

Joel built guardrails into Synter to catch that kind of problem much earlier. When speaking with Joel about a month back, I asked about things like how he built the guardrails that sit around his agent (those guardrails have already cost him at least one deal). I also asked him how that deal fell apart…

Read my full interview with Joel Horwitz on auditing the conversion count behind your ad spend

READ THE FULL INTERVIEW

What We Discussed

• What Synter's 3-minute, read-only audit found in 2 years of my client's Google Ads data.
• How Joel built Synter's prebuilt tools, and why he counts the LLM bill as part of your CAC.
• The six guardrails that Synter wraps around your campaigns, and the e-commerce deal that they cost Joel.
• How Joel decides which work to hand to an agent first on every new account.
• How Joel rebuilt Synter's pricing after a percentage-of-spend model cost him deals.
• The local ad channels that let a small business run its own campaigns without an agency.

Referenced

• Amp, the Sourcegraph product that Joel ran growth for.
• Google Ads, the platform that ran my client's campaigns.
• GA4, the source of the imported conversion action that inflated the count.
• MCP, the protocol that lets any agent that you already use reach Synter's tools through Synter's server.
• StackAdapt, an ad platform where Synter can provide the advertiser seat (the advertiser account that buys the ad inventory) for you.
• Nextdoor, Yelp and Apple Maps, the local ad channels that Joel has added.
• Agent Shield, the skill file scanner that Joel has open-sourced on GitHub.

Where To Find Joel Horwitz

• Joel Horwitz on LinkedIn
• Synter

Biggest Takeaways

1. A strong CPA tells you that the bidding algorithm hit its target. It doesn't tell you whether the conversions behind that target were real new leads, because the CPA is calculated from whatever the account counts as a conversion. In my client's account, the best-looking CPA sat on the least trustworthy conversion action in the account. As long as that action counted repeat submissions as new leads, every budget decision for 13 months was made against a number that overstated how well the campaigns were performing.
2. Synter flagged one of its findings as unverified: a possible double count between two form actions. Synter couldn't confirm that finding from inside the ad account, because confirming it required someone to inspect the tags on the website itself. An agent's audit becomes safe for you to act on when it separates the findings that it verified from the hypotheses that it couldn't verify. That distinction carries real weight once an agent has access to your ad spend. Without it, a hypothesis can turn into a budget change before anyone has checked whether it's true.
3. Joel counts the LLM bill as part of your CAC. He's pointed about it: "So people are like, oh, I launched this ad campaign and I got a CAC of like $45. I'm like, "but how much did you spend on AI to do it?" A CAC that leaves out the model bill makes an agent look cheaper than it really is. Unfortunately, that understated CAC is the number that you'll use to decide whether the agent stays in your stack. Your decision then starts from the wrong number.
4. My audit found the broken conversion count after the account had already spent 13 months of budget against it. Catching a broken conversion count before you execute ad spend costs you far less than catching it 13 months later, after every dollar spent has been optimized toward the wrong signal. Joel built guardrails into Synter to catch that kind of problem much earlier. Those guardrails have already cost him at least one deal.
5. If you pay someone to manage your ads, you can't yet say how much of that monthly fee pays for routine work that an agent could take over. Every month that you can't answer that question, you keep paying management rates for work that an agent could do. Joel showed me where he finds the answer on every new account that he connects.
6. If you're pricing an agent product of your own, Joel's path is worth studying. A fee that's set as a percentage of your customer's ad spend can lose you deals at both ends of your customer base. A percentage of spend tracks how much your customer spends on ads, while your own costs track how much work the agent does for that customer. Joel lost deals at both ends of his own customer base before he found a pricing model that fit.

In the full interview, I’m sharing the read-only audit that will return results on whether the conversion action that your bidding algorithm optimizes toward is counting each lead only once.

That’s vital information as every bid that the algorithm places depends on that answer. In my client's account, Synter answered that question in about 3 minutes, after 13 months of spend had already gone out the door.

I'm pointing Synter at the short-term rental homes that I manage next. Low season is when those homes need bookings most. Before my first low-season campaign goes live, I want Synter to audit the conversion tracking on my direct-booking site.

If you run the audit on your own account, hit reply and tell me what your conversion count turned up. I'd love to compare notes.

READ THE FULL INTERVIEW

All the best,

Lillian Pierson

Fractional CMO & GTM Engineer