Zeta Graded Its Own Customers on AI Adoption. The Gap Is Four to One.

Zeta Global reported Q2 yesterday. Revenue hit $443 million, up 44 percent. Adjusted EBITDA margin was 20.7 percent. Free cash flow rose 73 percent. The company posted positive GAAP net income for the quarter, $8 million, and raised full-year guidance to roughly $1.8 billion.

That is a good quarter. It is not the interesting part.

Buried in the release is a line about introducing an initial framework to measure adoption and monetization of Zeta's AI. In plain English: Zeta started grading its own customers on whether they actually use the product they bought.

Then the CFO published the grades.

Four to One Is Not a Rounding Error

On the call, Chris Greiner laid out the split. Roughly 20 percent of Zeta's customers have comprehensively adopted its AI tools. Those customers account for about 70 percent of revenue.

They grew four times faster than the other 80 percent still early in adoption.

Their net revenue retention, which is just a measure of whether existing customers spend more or less over time, runs 400 basis points above Zeta's overall rate and more than 20 percentage points above the customers still ramping up.

Same platform. Same contracts. Same sales team. Four to one on growth.

The variable is not the technology. Every one of those customers has access to identical software. The variable is whether anybody inside the building did the work to put it into production.

The Vendors Figured Out What the Buyers Haven't

Zeta is a public company. It does not invent a new disclosure category for fun. You report a metric when it explains your results better than the metrics you already report.

Think about what that means. Zeta looked at its own book of business and concluded that customer implementation depth predicts revenue better than customer size does. Among its largest accounts, the super-scaled ones averaging $1.8 million a year, half have comprehensively adopted the AI and that half drives 75 percent of the segment's revenue.

Being big does not protect you. Half of the biggest customers are still leaving most of the value on the table.

So the vendor now tracks your adoption, forecasts against it, and knows before you do whether your renewal is going to expand or shrink.

You Are Already Being Graded

Here is the uncomfortable read for anyone running technology at a company that bought an AI platform in the last two years.

Your vendor has this data on you. They know your seat utilization, your API call volume, which modules you turned on, which ones you never configured. They are building the same framework Zeta just published, whether or not they say so out loud.

If you are in the 80 percent, they know. Your account team knows. It is in a slide somewhere with your logo on it.

That is not a reason to be defensive. It is a reason to go look at the same numbers yourself, because the vendor is measuring the exact thing that determines whether your investment returns anything.

What Separates the 20 From the 80

It is never access. It is never budget approval. Both groups cleared those hurdles when they signed.

Somebody owned it. Not a committee, not a steering group, one person whose performance review says the thing works. Zeta's comprehensive adopters have relationships averaging 56 months, up from 48. That is not a honeymoon. That is years of somebody grinding on configuration.

The workflow got rewritten. Buying AI and bolting it onto a process designed for humans doing it manually gets you a slower version of what you had. The 20 percent changed how the work runs. That is unglamorous, political, and slow, which is exactly why most companies skip it.

The data was ready before the tool showed up. Clean records, defined fields, an actual definition of a customer. If your data is a mess, the AI will be confidently wrong at scale, and your team will stop trusting it in about six weeks.

Somebody trained the people. Not a launch webinar. Sustained, boring, repeated training until the new way is the normal way.

None of that is AI work. All of it is the work that decides whether the AI pays.

What I'd Do This Week

Ask your vendor for your adoption numbers. Use those words. Seat utilization, feature adoption, call volume against license. If they hesitate, that tells you something. If they hand it over fast, you just got a free audit.

Pick one workflow and finish it. Not five pilots at 40 percent. One process, fully converted, measured against what it replaced. Prioritize and execute.

Name an owner with authority. Someone who can change the process, not just report on it. Empowerment without authority is just a title.

Put a number on done. If you cannot say what success looks like in a metric, you are not implementing, you are experimenting. Experiments are fine. Just do not budget for them like they are transformation.

The Part That Matters

The gap between the companies winning with AI and the companies stuck with it was never about who had the better technology.

It was always about who did the unglamorous work after the contract was signed.

Zeta just put a number on it. Four to one.

Your vendor already knows which side you are on. Go find out yourself.