top of page

Stop Buying Insights. Start Buying Action.

Spotlight.ai VP of Growth & GTM Ops Lolita Trachtengerts on RevOps Unboxed with Tana Jackson


RevOps teams are drowning in insights. Every tool in the stack summarizes a call, scores a sentiment, flags a risk, and pushes another notification. None of it tells you what to do next. You get the dashboard. You still have to do the work.


On a recent episode of RevOps Unboxed, hosted by Tana Jackson, Lolita Trachtengerts sat down to make a blunt case: an insight you have to act on yourself isn't help — it's homework. The conversation runs from her path out of industrial engineering and into RevOps, through the real difference between copilot and autopilot, how to orchestrate agents without building a Frankenstack, and the three moves any RevOps leader can make right now to put AI to work.


The through line is simple: if a tool hands you data and walks away, it's adding to your job, not taking from it.


From Pipes to Pipeline

Lolita is a chemical engineer by training who spent roughly a decade in industrial water treatment — project management, operations, leading teams — before moving into tech. She joined Spotlight.ai through one of the co-founders she'd worked with in the water industry, and she's candid that RevOps was never the plan. It was, in her words, a happy accident.


What carried over is the thing that matters. "I'm an operator at heart," she says. Operations is operations, whether the system you're optimizing is a treatment facility or a revenue funnel. Her line for it: she used to do this on pipes, now she does it on pipeline. The instinct is the same — find the inefficiency, fix the flow, make the whole thing run better.


That operator's lens is why the insights problem bothers her so much. An operator doesn't want a readout. An operator wants the machine to run.


The Insights Trap

Asked what's overestimated in AI right now, Lolita doesn't hesitate: summaries and insights. There's a tool for summarizing your email, your Slack, your meetings, your calls — and the market has convinced itself that's the prize. She thinks it's overhyped. What's underestimated is AI that actually takes action. From day one, that's been the bet at Spotlight: not just surfacing data, but moving something on your behalf.


The reason it matters is the experience on the receiving end. When a tool throws insights at you with no guidance, it creates work and pressure instead of removing them.


You're left asking what the analysis even means and what you're supposed to do with it. As she puts it: forget the pretty dashboards — tell me what it means, and ideally, just do something about it. An overloaded operator with a hundred things on their plate doesn't need a hundred-and-first notification.


So her advice to anyone auditing their stack is direct. If a tool only nudges people to do their own work, kill it. The UI being nice or having a contact you like inside the vendor isn't a reason to keep paying for something that doesn't move the needle.


From Copilot to Autopilot

Everyone in B2B has been saying "copilot versus autopilot" for a while, and most people's eyes glaze over. Lolita landed on an analogy that actually makes it click: think about how you worked with ChatGPT a few months ago versus how you work with Claude now.


With the copilot, you did the work. You drafted the email, you asked for the summary, you built the deck — and you were still the one executing every step. The autopilot is different. With tools like Claude Code and agentic workflows, you can build the flow itself: given this data, here are the actions to take, and here are the actions I'm comfortable handing off entirely. The shift isn't a better assistant. It's deciding what you no longer need to touch.


That, she notes, is exactly what Spotlight has been saying all along — stop asking what the AI tells you, start asking what the AI does for you. In the world of MCP and connected tooling, people are finally on the same page. The blank stares are gone. It's clicking.


One Brain, Many Hands

Tana raised the obvious risk: if every function spins up its own agents, haven't you just built a new Frankenstack — the same tool sprawl, now with AI on top?


Her answer is that more agents are coming, and that's fine — as long as they answer to one brain. Picture a unified brain with specialized hands: a set of agents executing the work, governed by something that knows whether they're doing the right thing with the right data. Without that orchestration layer, agents run around doing whatever, and the hallucination risk climbs fast. Build an agent on a general model and assume it understands how your company runs, and you'll have a painful day when it confidently invents something that was never true.


For Spotlight, that brain is the Knowledge Graph — built over four to five years to hold the enterprise sales knowledge, industry context, and playbooks that actually work, so the agents acting on top of it stay inside the guardrails. In a small company, she notes, coordination is easy because everyone knows everything. In an enterprise with sprawling tools and data, the orchestrating brain isn't optional — it's the only thing keeping the agents honest.


(One aside worth keeping: Lolita admitted to building a Claude skill that replicates a growth leader she admires but has never met, so she can run ideas past a stand-in version of him. A small, very on-brand example of bending these tools to a real workflow instead of chasing the hype.)


Three Ways RevOps Leaders Can Operationalize AI Right Now

For leaders just putting AI initiatives on the table, Lolita offered three concrete moves:

  1. Kill anything that doesn't act. Start by ripping out tools that only nudge people to do their own work. They aren't helping.

  2. Automate one workflow you already know. Ignore the LinkedIn theater of people claiming they built a hundred agents in a weekend. Pick a single workflow you understand and run today, automate that, and let the confidence compound from there.

  3. Pick one that's already broken. Choose a workflow that doesn't work now, because there's no downside. Fix it and you've made something better. Don't, and you're no worse off than before.


Underneath all three is the failure mode she sees most: starting with the hype instead of the problem. Companies launch AI projects because AI is cool, not because they've named what's actually broken. That's a failure before the work even begins.


The Source of Truth Is Moving

If the CRM has always been the system of record, Lolita argues it's quietly stopped being the source of truth. The problem is timing — you have to work to fill the CRM and work to update it before you can get anything out of it, which means the data you need rarely arrives when you need it. And it doesn't hold the thing that actually moves a deal: the voice of the customer, both in your conversations and in everything they signal publicly.


She reads Salesforce's Headless 360 — announced at TrailblazerDX in April 2026, exposing the platform through APIs and MCP tools rather than a UI — as a tacit admission of exactly this. It's the CRM acknowledging it isn't the single source of truth and reaching to connect into the tools where the work now happens. Plug the CRM into Claude or whatever you're running, pull in the voice of the customer alongside it, and the truth assembles across systems instead of living in one form nobody wants to fill out.


Why This Is RevOps' Moment

For most of her career watching RevOps from the sales side, Lolita thought of the function the way many leaders did: the people who build the dashboards, prep the pretty slides, and clean up the CRM the AEs won't touch. That picture is gone.


RevOps now sits at the very front of everything happening with AI. The role is becoming more strategic than almost anything else in the go-to-market org — working in workflows and automation, operationalizing things that software limits used to make impossible, and serving as the testers at the edge of what these tools can do. There's no better time to be in the seat.


What Leaders Should Take Away

Four things worth keeping from this conversation:

  1. An insight you have to act on yourself is homework, not help. Evaluate every AI tool by whether it takes an action off your plate or just adds one more thing to read.

  2. Copilot makes you faster at your work. Autopilot does the work. Know which one you're buying, and be deliberate about which actions you're ready to hand off.

  3. More agents need one brain. Specialized hands are fine; ungoverned hands are a Frankenstack. The orchestrating layer — the knowledge graph that holds your context — is what keeps autonomous agents accurate.

  4. Start with the broken thing, not the trendy thing. Name the problem first, automate one workflow you already understand, and let trust build from a real win.


And the line she'd put on the back of the tour T-shirt? Your pipeline is a lie. She has yet to meet an enterprise sales org whose leaders fully believe their own pipeline — and until the data assembles itself from the voice of the customer instead of manual CRM entry, she isn't expecting to.


About the Show

RevOps Unboxed is the Revenue Operations Alliance podcast for RevOps practitioners navigating the realities of modern GTM. This episode was hosted by Tana Jackson, a revenue operations leader who has built and run global RevOps functions across complex, multi-system organizations. Her work centers on taming tech-stack and data complexity and translating it into processes that actually move revenue.



agentic AI for RevOps


Listen to the Full Conversation

🎙️ Lolita Trachtengerts on RevOps Unboxed with Tana Jackson



Q&A

Q: What's the single most overrated category in AI for sales right now?

A: Insights and summaries. There's a tool to summarize every call, email, and meeting, and the market treats that as the destination. It isn't. An insight you still have to interpret and act on yourself is just more work. The underrated category is AI that takes action — that actually moves something on your behalf instead of handing you a dashboard.


Q: How do you explain copilot versus autopilot so people get it?

A: Compare how you used ChatGPT a few months ago to how you use Claude now. With the copilot, you did the work — you drafted, you asked, you built, and you executed every step. With the autopilot, you build the workflow itself and decide which actions to hand off entirely. The difference isn't a smarter assistant. It's the work you no longer have to touch.


Q: If every team builds its own agents, how do you avoid a new Frankenstack?

A: You give the agents one brain. More agents are coming, and that's fine, but they have to answer to a unified orchestration layer that knows your context — a knowledge graph holding your sales knowledge, industry context, and playbooks. Without it, agents run on general web knowledge, hallucinate, and act on things that were never true about your business.


Q: Where should a RevOps leader actually start with AI?

A: With a workflow that's already broken. The downside is zero — fix it and you've improved something, don't and you're no worse off. Then automate one workflow you genuinely understand and run today, and let the confidence build from there. The most common failure is the opposite: launching AI projects because AI is trendy, without ever naming the problem you're solving.


Q: Is the CRM still the source of truth?

A: It's the system of record, but the timing is broken — you work to fill it and work to update it, so the data rarely arrives when you need it, and it doesn't hold the voice of the customer. Salesforce's own Headless 360 move, opening the platform up through APIs and MCP tools, reads as an admission of that. The truth increasingly assembles across connected systems, not inside one form nobody wants to complete.


Q: What's the first thing you'd rip out of a tech stack?

A: Any insights tool that doesn't take action. You're being handed more data than you can process, with no guidance on what to do with it, so it becomes next week's problem and then nobody's problem. If it doesn't act, it goes.


Comments

bottom of page