This quarter, Weisi Kang built an AI agent to do a job her whole team kept getting wrong. The job sounds trivial until it isn't. People forget to associate a contact. They forget to log a meeting. At scale, those small misses mean opportunities never get tied to the marketing programs that actually drove them, and the pipeline story quietly goes wrong. So she built what she calls a first-touch ops detective. It reads across her systems, decides whether an opportunity really belongs to a program, fixes the record, and flags the ones that look off. The other week it let her clean up ten opportunities in twenty minutes.

That work used to mean digging through contact records, account history, and several systems by hand. It is also a small window into a bigger claim: the next generation of CMOs will come from ops.

Weisi has spent more than a decade in marketing operations. Medallia, Rubrik, Autodesk, Lattice, and now Glean, where she leads marketing ops. Glean builds AI for the enterprise, which puts her in a rare seat. She markets AI for a living, and she uses AI to do the marketing. She is also a RevSure customer.

The pitch is seductive. End-to-end automation. Agentic workflows that complete your work while you sleep. And in certain narrow use cases, it genuinely delivers.

But Weisi has seen where it falls apart, and it is always the same place.

The demo works because someone already did the invisible work off-screen. The data was clean. The metrics were defined. The processes were documented. The relationships between systems were mapped. Strip that away and the same agent produces confident nonsense.

Few people want to hear it. The model was never the hard part. What takes the work is the context layer underneath, assembled one definition and one documented process at a time.

For years, documentation was the work you did when you had nothing better to do. Define the metrics. Write down the process. Keep the taxonomy clean. Necessary, unglamorous, easy to postpone.

Weisi flipped that for me.

Every LLM you point at your marketing data is only as good as the context it can reach. Feed it a pile of numbers with no definitions and it guesses. Feed it your documented processes, your metric logic, and the language your team actually uses, and it reasons.

The teams drowning in AI slop are not using worse models. They are sitting on a thin context layer and blaming the tool.

Weisi's practical move is worth stealing. In an age where everything gets recorded and transcribed, you rarely have to document from scratch anymore. The raw context already lives in your call recordings, your notes, your Slack. The job is capturing it once and making it findable, not repeating yourself into a wiki nobody reads.

The most boring job in marketing quietly became the one that decides whether your AI works at all.

AI democratized access to data. Every recording, every note, every transcript is now searchable and available to the whole team.

That sounds like pure upside. Weisi is more honest about it.

More data is only a gift if someone turns it into a decision. Otherwise it is just a bigger pile to get lost in.

Where she sees the real value is in what happens after the stitching.

Then the human takes over. The business owner, who knows why the program ran and what the team was trying to do, looks at that judgment and decides whether it holds. The agent assembles the read, and the person makes the call.

The payoff is not just speed. It changes what her marketers spend their day on. Instead of waiting on someone in ops to explain what the numbers mean, they get a synthesized read and move straight to the decision. The ask-and-answer loop that used to eat days collapses into minutes. But it only works if leadership already understands the business context behind the metrics. That trust has to come first.

I asked Weisi whether AI is rewriting the old idea of peacetime versus wartime marketing organizations.

The work itself is not new. What changed is the compression.

Picture the most common decision in marketing. A channel looks like it is underperforming. Paid social, a brand campaign, an out-of-home experiment. The numbers on a straight-line attribution report are weak. So you cut it and move the budget somewhere that measures cleanly.

Then this happens.

The channel was not underperforming. It was doing invisible work holding up everything around it. Your last-touch report just could not see it, because the contribution never showed up as a clean touch.

This is why one measurement model is a trap. Multi-touch attribution might tell you direct mail contributes almost nothing. Marketing mix modeling might tell you direct mail is quietly lifting your paid search and your SDR motion. Incrementality testing tells you what actually would not have happened without the spend. Each method sees a different slice of the truth. Judge a channel on one of them and you will eventually cut the thing propping up your whole mix.

Weisi's team looks at every major program through several angles and assembles one story from all of them. And because they can see the cross-channel lift early, they avoid the painful version of this lesson, the one where you cut, watch everything sag, and scramble to move the money back. She caught the connection in the data instead of learning it the hard way.

The same discipline shows up in how she handles finance. B2B payback runs longer than B2C, so a straight-line ROI story sells the CFO short and sets everyone up to kill good programs early. Her move is to educate finance on the models beyond one-to-one attribution before she needs the budget, set an honest expectation on payback, and then actually check back at the horizon she promised. The credibility compounds.

I opened with this one because it is the clearest picture of how she thinks about AI. She is skeptical of the hype and precise about where it earns its place, and the ops detective is where it earns it.

In the Glean's system, people forget to make associations and log meetings. The agent does not invent anything. It reads across the places that already hold the answer, her CRM, her team's Slack, the account journey, and works out whether an opportunity really belongs to a program. When it does, it fixes the record. Every week it also sweeps the new opportunities, flags the ones that look mislabeled, explains why, and waits for her to approve.

That is what useful AI looks like in practice. It did not replace her judgment. It handled the research grind across four systems so she could make the call. The version that actually works takes the busywork and leaves the judgment with her.

Marketing ops leaders rarely take the CMO stage. I think that is about to change, and Weisi is a good argument for why.

The thing that makes her valuable is not that she knows the answers. It is that she knows where to look for them. She understands what a multi-touch model tells her versus a marketing mix model versus an incrementality test, so she knows which instrument to pick up for which question. In a world where the tools keep democratizing, that signal-detection instinct is the rare skill.

Her build-versus-buy philosophy comes from the same place. In the era of the vibe coder, anyone can prompt something into existence. But the last mile, the QA, the security review, the checklist that takes a prototype to production, is exactly what most creators forget. So she does not treat it as build or buy. She treats it as leverage the best pieces of everything, using RevSure's data associations as an input into their own warehouse rather than picking a side.

And she sits in the one seat that sees the entire motion. Demand gen, field, events, the whole engine and how each part performs on the back end. That vantage point is exactly what a CMO needs.

Key takeaways - if you lead marketing or marketing ops, carry these out of this conversation:

This conversation reminded me why I started Moneyball CMO. The leaders worth learning from are not the ones with the flashiest AI stack. They are the ones who did the patient, unglamorous work underneath it, so that when the tools finally caught up, they were ready to use them for real.

The technology will keep getting more capable. The foundation still gets built by hand. That is the part nobody can automate for you.

So when you go looking for your next marketing leader, do not just reward the ones who talk about AI the best. Find the ones quietly building it, on top of groundwork they laid themselves. They are probably sitting in ops right now.

Watch the full conversation with Weisi Kang here:

Moneyball CMO is a series featuring marketing leaders who challenge conventional B2B wisdom with data, candour, and real experience. Know a marketing or ops leader who should be on the show? Reply to this email.