Stop Chasing the Trend. Start with the Bottleneck.
- Lolita Trachtengerts
- Mar 20
- 7 min read
Spotlight.ai CEO Roi Carmel on AI Powered RevOps with Sandy Robinson
RevOps teams right now are getting hit from every direction. New AI tools every week. Every vendor promising transformation. Every conference track dedicated to the stack you're supposedly missing.
And somewhere in all of that noise, the actual work of running revenue operations is getting harder to prioritize.
In a recent episode of AI Powered RevOps, hosted by Sandy Robinson, Roi Carmel sat down to talk about what's going wrong with how organizations approach AI — and what a more grounded path looks like. The conversation covers evaluation frameworks, the division of labor between humans and machines, the gradual handoff of control, and where the productivity model in enterprise sales is actually headed.
The through line is simple: most teams are looking outward when they should be looking inward first.
The AI Hype Problem Is a Prioritization Problem
RevOps teams right now are getting hit from every direction. New AI tools every week. Every vendor promising transformation. Every conference track dedicated to the stack you're supposedly missing.
Most teams go looking for what's popular — conversational intelligence, CRM automation, agents that fill fields — and then try to map those capabilities back to their operation. That's backward. The right starting point is an honest look inward: where are we missing our forecast? Where is pipeline converting poorly? Is it an early-stage qualification problem, a mid-cycle execution problem, or something that's falling apart at the close?
Identify that bottleneck first. Then go looking for what solves it.
Roi walks through a real customer example: a mature company with strong execution but persistent forecast surprises. They were consistently finishing the number, but not the way they thought they would. Deal outcomes didn't match deal signals. They had good people and a lot of activity. What they lacked was structured qualification data and a way to surface risk before it became a miss.
The risk of skipping that diagnosis: you automate the wrong thing. Agentic AI applied to a broken process just runs the broken process faster.
Humans Are Bad at Analytical Work. That's the Design Principle.
One of the clearest frameworks Roi lays out in this conversation is a taxonomy of what salespeople actually do — and where they fall apart.
Enterprise reps are good at building trust and human connection. That's what they're wired for. That's what gets them in the room. The problem is that on top of that, they're also expected to map deals to a qualification framework, identify risks in real time, update CRM fields, and generate assets for their next meeting. They either hate doing these things or they're not trained to do them well. Usually both.
This isn't a talent problem. It's a design problem. Those activities require different parts of the brain, different skills, and a different feedback loop. The analytical layer — finding patterns across calls and emails, identifying risks from historical loss data, structuring evidence into a CRM — is exactly where AI will always be superior.
So the design principle isn't "give reps better tools for the things they hate." It's "take them out of the things AI will always do better, and free their attention for the things only humans can do."
Roi articulates three principles Spotlight.ai applies when evaluating where to deploy AI with a customer: let humans excel at human interactions; let AI take over where it will always be superior through the Listen → Understand → Decide → Act framework; and don't give teams more tools — integrate AI into the workflows they already run and remove them from the loop.
That last one matters more than it sounds. There's a difference between giving a rep a tool that makes CRM updates easier and taking CRM updates off their plate entirely. The second approach is the only one that changes behavior at scale.
The Driving Analogy Is Still the Right One
Roi returns to the autonomous vehicle parallel, and it holds up. The journey from side-mirror blinkers to Tesla's full self-driving didn't happen overnight. It happened because each step gave drivers more confidence, and that confidence made the next step feel less risky.
The same thing is happening in enterprise sales now. Gong came out and nobody wanted to be recorded. Then reps realized they didn't have to take notes anymore, and suddenly they wanted the bot in the room. That's one small transfer of control. It changed behavior without requiring anyone to change their beliefs about AI.
The adoption curve for agentic AI in sales follows the same logic. Internal workflows with no direct customer contact are the parking garage — low risk, low stakes, good place to prove the system out. Once teams see that AI qualifies deals more accurately and structures data more consistently than reps ever did, the case for going further writes itself.
What needs to happen between now and full autonomy is the right controls — guardrails that match the risk of the workflow. The answer to "I'm not ready to give AI that much control" is almost never "then don't." It's "then let's figure out what the first step looks like and build trust from there."
What the Next 18 Months Actually Looks Like
The practical vision Roi describes isn't abstract. It's a specific change in the productivity model.
Right now, an enterprise rep manages a finite number of opportunities because each one demands a certain amount of manual effort: updating fields, running analysis, building materials. When AI handles that effort, the same rep can manage more opportunities and go deeper into the accounts that matter. The bottleneck shifts from time to judgment — which is exactly where you want a human in the loop.
The same logic applies to first-line managers. Today's pipeline reviews are expensive. A manager spends thirty minutes with a rep piecing together what happened last week. In an AI-native org, both people walk into that conversation with the same picture already. The conversation starts at the coaching question, not the status update. That changes how many people one manager can effectively run — not seven or eight, but closer to fifteen or twenty.
That's not a small efficiency gain. It's a different business model for scaling revenue teams.
And it starts with data structure, not data volume. Roi is clear on this point. AI agents are only as good as the structure of the data they run on. A two-hour call transcript thrown at an LLM will produce errors. The same information mapped into atomic signals inside a knowledge graph — that's where accurate decisions come from.
What Leaders Should Take Away
Four things worth keeping from this conversation:
Start with your bottleneck. Not with the trend. Figure out what metric needs to change for your biggest problem to go away, and evaluate AI against that. Everything else is a distraction.
There are three ways to deploy AI in a workflow: replace a human entirely, work alongside the human, or do something the team couldn't do at all before. Each requires a different level of readiness and a different set of guardrails. Know which mode you're in before you buy.
Good CRM hygiene is downstream of data structure. The reason most RevOps data isn't AI-ready isn't dirty fields — it's that conversations never got turned into structured signals in the first place. Structured data is what agents actually run on.
The productivity model changes when reps move from executor to orchestrator. That's the destination. Every AI deployment should be evaluated by whether it moves the team closer to that state or just makes the current state marginally faster.
About the Show
AI Powered RevOps is hosted by Sandy Robinson, VP of Revenue Operations at Quavo. Sandy has spent over a decade leading RevOps and sales operations functions across high-growth companies, including NYMBUS, Supplyframe, and Patra Corporation. She also hosts RevOps Unboxed in partnership with Revenue Operations Alliance, a podcast built for RevOps practitioners navigating the realities of modern GTM operations. Her work centers on the intersection of people, process, and technology — and what it actually takes to build revenue functions that scale.
Listen to the Full Conversation
🎙️ Roi Carmel on AI Powered RevOps with Sandy Robinson
🎧 Available on Spotify and Apple Podcasts.
Q&A
Q: Why do most RevOps teams evaluate AI the wrong way?
A: They start with what's popular. Conversational intelligence is everywhere, so they look for conversational intelligence. Agents for CRM are trending, so they look for agents. The right approach is the inverse — understand your biggest operational bottleneck first, then go find what solves it. You'll often land on the same tools, but you'll buy them for the right reason and deploy them where they'll actually move a metric.
Q: What's the right way to think about "AI readiness"?
A: It starts with your data structure, not your data volume. Agents running on raw transcripts will make mistakes. Agents running on structured signals mapped to a knowledge graph will be precise. Before asking whether your data is clean, ask whether your conversations are being turned into structured information at all. Most aren't.
Q: How do you measure ROI from an AI deployment?
A: Start with the problem you were trying to solve. What metric needed to change for that problem to go away? That metric is your ROI measurement. Don't let a vendor define success for you by picking the metric where their tool looks best. Define it yourself before the deployment starts.
Q: Should RevOps be afraid of AI replacing the function?
A: No. But the role changes. RevOps leaders who understand AI and deploy it well will have more leverage, more visibility, and more scope than they do today. Those who don't will get outpaced by the ones who do. The question isn't replacement — it's whether you're the person building the new operating model or inheriting someone else's version of it.
Q: What's the right entry point for AI in a sales org?
A: Internal workflows with no direct customer contact. The risk is lower, the feedback loop is faster, and you build the organizational trust needed to go further. Get AI running qualification evidence into CRM with no rep involvement. When your managers see cleaner data and better deal visibility, the conversation about what comes next gets a lot easier.
Comments