Companies: Kelley Blue Book and AutoTrader
The short version: Lead volume grew 15x, quality went up instead of down, and we eventually had to install caps because the system produced too many leads. That last part is a problem most marketers would love to have. It’s still a problem.
The deal, and the catch
In 2007, Kelley Blue Book partnered with AutoTrader. The logic was clean: KBB helps people figure out what their car is worth, AutoTrader helps them sell it. We agreed to send up to 100,000 prospective sellers into AutoTrader’s used-vehicle marketplace.
It worked well enough that when the agreement came up for renewal in 2008, AutoTrader asked for a million leads. Ten times the volume. With one catch: the new leads had to close at a higher rate than the old ones.
So we weren’t being asked to open the firehose. We were being asked to open the firehose and make the water cleaner. That’s a different job entirely.
The question we didn’t ask
The obvious question was “where can we cram in more AutoTrader links?” We skipped it. More traffic was easy. More of the right traffic required understanding which behaviors on KBB.com signaled someone genuinely ready to sell, and what would move that person naturally to the next step.
Someone skimming general car content is not the same opportunity as someone pricing a specific vehicle or working out a trade-in value. One is browsing. The other is doing math. We needed to find the ones doing math.
Finding the signals
We started with a deep dive into KBB.com traffic, looking not at aggregate pageviews but at how people actually moved: what they researched, where they engaged, where they converted, where they left, and where they seemed ready for a next step.
That map showed us the moments where an AutoTrader offer would feel useful instead of intrusive. Context was everything.
Every placement was a bet
Instead of one generic message everywhere, we built different calls to action for different audiences, pages, and stages of the journey, and treated each combination as a hypothesis. Does an inventory-focused offer beat a general shopping prompt? At what point in the research journey does someone actually continue on to AutoTrader? Which placements get clicks from people who never do anything afterward?
Every CTA got a unique tracking code tied to its message and location. That instrumentation is unglamorous, and it’s the whole ballgame. Without it, we’d have known people clicked. With it, we knew which experience prompted the click and where that person was in their journey when it happened.
Testing as a habit, not an event
We ran A/B and multivariate tests across messages, designs, offers, page locations, and landing experiences. Continuously — not as a one-time optimization sprint.
Here’s the trap we were trying to avoid: the most-clicked CTA is not automatically the best one. A placement can pull huge traffic from people who will never list a car. Optimize for click-through rate alone and you end up scaling exactly the wrong experiences, with great-looking dashboards to show for it.
To know what was actually working, we had to see what happened after people left our site.
Closing the loop
AutoTrader shared conversion data, which let us connect KBB referrals to actual downstream results. This changed everything. We were no longer measuring impressions, clicks, and form fills. We could see which messages, placements, and audiences led to vehicles getting listed.
The loop looked like this: observe behavior on KBB, spot a moment of intent, present a relevant offer, measure the response, compare it against AutoTrader’s sales outcomes, refine, and scale what worked. KBB knew how the customer researched. AutoTrader knew whether they transacted. Neither dataset alone could tell you what a qualified customer looked like. Together, they did.
Pruning while growing
As results came in, we cut low-performing links and CTAs, including some that generated impressive-looking activity, and concentrated traffic on the paths with the strongest mix of relevance, engagement, and purchase behavior. The program got more efficient as it grew, because the system kept trimming its own dead weight.
The results
Within two months, we were on pace to deliver roughly 2 million customers to AutoTrader. Twenty times the original agreement. Double what AutoTrader had asked for in the expansion.
We renegotiated the partnership around a commitment of 1.5 million leads and hit it while clearing a higher quality bar. The number people remember is the 15x growth. The number that mattered was the closing rate, which went up instead of getting sacrificed to scale.
Then we installed brakes
Once we knew what the system could do, the next job was restraint. Maximum volume wasn’t in anyone’s interest. It could overwhelm operational capacity, blow past contractual commitments, and drag lower-quality customers into the pool.
So we set daily and monthly lead caps. Steady, predictable delivery of qualified customers beat sporadic floods of everyone. Optimization isn’t always about producing more. Sometimes it’s understanding the system well enough to know when more starts producing less.
What this taught us
A few principles from The Adaptive CMO, as they played out here.
Optimize for the business outcome, not the proxy. Clicks and referrals were indicators. Listed vehicles were the point. Connecting marketing activity to closing data kept the intermediate metrics from quietly becoming the goal.
Intent is contextual. Demographics didn’t tell us who was ready to sell. Behavior did: what they were reading, what they were pricing, where they were in the journey.
Instrument before you optimize. The tracking structure turned a pile of links into a measurable system. Skip that step and you generate more activity without understanding what caused it.
Treat execution as a portfolio of hypotheses. Every CTA and landing experience was a testable belief about customer behavior. Some survived. Others didn’t. The market votes; you count.
Close the learning loop. KBB’s data explained the before. AutoTrader’s explained the after. The insight lived in the combination — feedback that crossed company lines, not just department lines.
Scale evidence, not assumptions. The program didn’t start with a 1.5-million-lead commitment. It started with behavioral analysis and controlled tests, and scale followed the evidence.
More is not always better. When the system exceeded its target, we capped it rather than chasing unlimited volume. The job is balancing the variables that create the greatest total business value, not running one metric to the moon.
The lasting lesson
This started as a request for more leads. It became a shared intelligence system connecting customer research, message relevance, experimentation, referral behavior, and final sales — one that grew volume fifteenfold while improving the quality of every customer we sent.
That’s the difference between running a bigger campaign and building an adaptive growth engine. A campaign produces an outcome. An adaptive system explains why the outcome happened, improves on what it learns, and makes the next decision better.
Build Marketing That Learns Across the Entire Customer Journey
The AutoTrader partnership is an example of the operating principles explored in my book, The Adaptive CMO: The Secret to Revenue, Relevance, and Results.
The book provides a practical framework for recognizing customer signals, designing measurable experiments, connecting marketing activity to revenue, and building programs that get smarter while they run.
[Explore The Adaptive CMO]