We optimized for phone calls. We got phone calls.

A nationwide auto transport brokerage was running Google Ads. The campaign was a Performance Max campaign, which means Google decides where the ads go: search, display, YouTube, Gmail, wherever it thinks it can find a conversion. You give it a goal and a budget. You don't see much of what it does with either.

The goal was phone calls.

That's a reasonable goal for a business where deals close on the phone. It's also the goal Google makes easiest to pick. Calls were coming in, cost per call looked fine, and the dashboard was green. What nobody could answer was a simpler question: who was calling?

The first thing we did was listen

Before changing anything, we added call tracking, so every call from the ads got a distinct number and a recording. Then we listened.

The campaign was doing exactly what it had been told. It was finding people likely to call. A lot of those people were:

• existing customers checking on a shipment already booked
• people with a complaint or a service question
• customers calling back from a different phone, which slipped past the "first-time caller" filter
• people who wanted a service the company doesn't offer

Each of those was a conversion. None of them was a new deal. The platform wasn't wrong. The instruction was. Ask a bidding algorithm for calls and it will find you people who call.

This is Goodhart's law with a budget attached: when a measure becomes the target, it stops being a good measure.

Then we asked for leads, and got a different kind of junk

The next move was a form. A lead form gives you fields, and fields give you something to filter on that a phone call doesn't.

Leads came in. Many of them were requests for international shipping. The company only ships domestically. A large share of the audience was Spanish-speaking, and the ads were reaching people whose need, quite reasonably, was getting a car to or from another country. The platform had found a real audience with a real problem. It just wasn't a problem this business could solve.

So we were one rung up the ladder and still paying for conversions that couldn't become revenue.

The ladder

Each rung is harder to track and closer to money.

1. Calls. Easiest to count. Weakest signal. You know someone dialed.
2. Leads. Someone gave you information. You can filter, but you still don't know what a lead is worth.
3. Qualified leads. Someone, or some rule, looked at the lead and said "real."
4. Lead value. A dollar estimate attached to each lead the moment it arrives, so the platform can chase value instead of volume.

You can't skip rungs. You can climb them.

What we told the machine to chase instead

Every lead had an origin zip code and a destination zip code. For an auto transport broker, those two numbers predict most of what a deal is worth: the distance, the route, the likely price, and the margin left after the carrier takes their share.

So we built a simple calculation: given origin and destination, estimate what this lead would be worth if it closed. International and out-of-scope routes got no value. Short local hauls, which the company had been chasing by default, got a small one. Long domestic routes got a large one.

That estimate was passed back to Google as the conversion value. From that point on, the campaign was no longer optimizing for "someone called" or "someone filled out a form." It was optimizing for "someone we'd want to talk to."

What happened

It took almost two months for the campaign to learn from the new signal. That's the part nobody likes hearing. Value-based bidding needs enough valued conversions to find the pattern, and for a small account that takes weeks, not days.

Then the shape of the business changed:

• Total leads went down.
• Lead quality went up.
• The average value per closed deal went up, because the campaign stopped chasing low-margin local moves and started finding long-haul domestic ones.
• Total revenue went up, even with fewer leads.

Fewer leads and more revenue is the outcome that a "calls" or "leads" dashboard will never show you, because both of those dashboards would have reported this as a decline.

The caveat

Value-based bidding needs conversions to learn from. For a tiny account, the first move is not lead value. The first move is getting off "calls" and onto a form that asks one qualifying question. The second move is listening to your own calls. Only after that does a value model have anything to work with.

And the value model doesn't need to be clever. Ours was two zip codes and a lookup. What mattered was that it encoded the one thing the business actually knew: which deals were worth having.

Methods and assumptions

How lead value was estimated. Origin and destination zip codes were mapped to an estimated deal value using the company's own pricing and margin by route length and region. Routes outside the service area were assigned zero value. The estimate was passed to Google Ads as a conversion value on form submission.

What it deliberately ignored. Vehicle type, timing, and customer history were left out of the first version. They likely predict value too, but adding them would have delayed the switch, and the two-zip model already separated the deals that mattered from the ones that didn't.

What would change the conclusion. If lead count had dropped without a matching rise in closed revenue, the value model would have been wrong about what predicts a deal. Revenue was the check, not lead quality on its own.

What this is not. This is one account over a few months, not a controlled experiment. The comparison is before and after, with the same business, the same season change, and the same competitors moving underneath. It is strong evidence that the goal mattered. It is not a measurement of how much.