SaaS VP of RevOps Says AI tool Sprawl Is The Hidden Cost That Nobody Is Measuring
Find out why Mollie Bodensteiner says AI succeeds in RevOps only when strong data, disciplined processes, and human judgment come first.
Mollie Bodensteiner
mollie-bodensteiner
AI Impact: AI challenges traditional CRM reliance, emphasizing real-time engagement over conventional data logging.
Deal Assessment: AI analyzes diverse signals, enhancing deal health scoring beyond traditional CRM field entries.
Productivity Shift: Reducing tool sprawl boosts productivity, focusing AI integration in primary workspaces.
Data Foundations: Strong data foundations are critical before implementing AI to ensure accurate outputs.
Human Judgment: AI offers insights, but humans must retain decision-making roles for context-specific outcomes.
Mollie Bodensteiner is VP of Revenue Operations at ZoomInfo, a public SaaS company serving enterprise clients globally. She’s also the creator of The RevTech Review.
We sat down with Mollie to understand how she’s changing deal assessment and forecasting workflows with AI. Here’s what she told us.
AI tests whether your data and process foundations can hold
I lead Revenue Operations at a public, enterprise-scale SaaS company serving mid-market and enterprise customers globally. Our federated model spans pipeline ops, sales ops, CX ops, partnership ops, and GTM systems, supporting a complex, multi-motion go-to-market organization.
Create a free account to continue reading this article.
Want more from the CRO Club? Sign up to continue reading and join modern revenue leaders using proven playbooks, peer insights, and AI-driven strategies to build predictable pipeline and scalable growth.
Have an account? Log In
We run a subscription, usage-influenced SaaS model with a multi-motion go-to-market: new business, expansion, and renewals across self-serve, mid-market, and enterprise segments. My RevOps team is around 20 people supporting the full GTM organization.
My background is over 15 years working across go-to-market: RevOps, systems, compensation, automation, territory and quota planning, data architecture, and engineering. I've built and scaled these functions inside multiple hyper-growth unicorns, the kind of companies where the business is moving faster than the systems meant to support it.
That constant rebuilding is what shaped how I think. Every time a company doubled in size, the systems that worked yesterday stopped working today. I learned to design for scale before scale demanded it.
That's the journey that brought me to this moment in revenue leadership: not chasing AI as a trend, but seeing it as the next test of whether your foundation can hold.
How AI-powered workflows enhance deal assessment and forecasting
RevOps has built entire careers around the assumption that the CRM is the system of record. If it's not in Salesforce, it didn't happen, and the forecast is only as good as what was typed in.
AI broke that for me. The CRM was never the source of truth. It was always what a rep had time to document, filtered by what they remembered, what they thought mattered, and what they didn't want to admit. The real signal was always somewhere else: in the call, in the email thread, in how a buyer responded versus how the rep wrote it up.
So, we changed what we asked reps to do. We stopped asking them to manually log notes and update fields that an agent could now infer from the call itself. Their job shifted from data entry to selling, and from defending their stage and close date to actually engaging with what the data was telling them.
The forecast process changed, too. Instead of starting with what a rep typed in and working backward to validate it, we start with the full signal set and check the CRM as one input against everything else.
Once you can pull all of that into a single context layer, the CRM stops being the system of record and becomes one input into it. The system of record is the full picture now, not the field a rep filled in on their way to their next call.
A real-world deal assessment agent
We built an agent that reviews opportunities as they progress through the pipeline, pulling both first-party signals — CRM activity, call notes, and email — and third-party signals — intent data, buying committee research, and market context.
It scores deal health based on actual sentiment and engagement, not just methodology field completion. A deal can check every MEDDPICC box and still be cooling. A deal can be missing a field and still be hot.
The model looks past filled fields to what's happening in the conversation. Are buyers showing up to calls, or is the rep still only talking to a champion who can't sign? What kind of resistance or questions come up? Are they tactical and expected, or are they signs the buyer is reconsidering? Is the tone getting warmer or cooler over time? Is the close date realistic, or is it a date the rep wants to be true versus one the buyer confirmed?
That last one matters more than people think. There's a real gap between what's said on a call and what a rep hears and writes down, and it represents one of our strongest risk signals. The model catches the gap, comparing call transcripts and notes against what reps documented in the CRM.
And the model also tracks responsiveness over time, not as a snapshot. A buyer who replied fast a month ago and has gone quiet for two weeks reads very differently from a deal that has been steady throughout.
All of that rolls up into a risk category that weighs the deal in the forecast. A deal with the right buyer engaged, real urgency, and a defensible close date carries real weight. A deal where the rep is the only one optimistic about the timeline is discounted, regardless of its stage.
That's the shift: pipeline weighting used to reflect a deal's stage in the process. Now, it reflects what is actually happening.
A deal with the right buyer engaged, real urgency, and a defensible close date carries real weight. A deal where the rep is the only one optimistic about the timeline is discounted, regardless of its stage.
How AI reliance creates blind spots
When an agent scores a deal, flags a risk, or recommends a next step, reps easily follow it without asking if it's right for that specific buyer.
The risk isn't that the AI is wrong sometimes. It's that the reps stop checking. They outsource the thinking, not just the task, and you don’t notice until a deal everyone trusts because of the score falls apart for a reason the model would never catch.
That's the part most leaders underestimate. AI is supposed to sharpen judgment, not replace it. If your reps can't explain why a deal is healthy beyond "the model said so," you've traded one blind spot for another.
How AI benefits revenue processes
Here are the results I’m seeing overall.
The good: Forecast variance decreased, which improved pipeline coverage. We moved past basic analytics into insights and intelligence that tell reps not just what happened but also what to do next. We also developed stronger messaging and engagement materials, powered by a context graph that provides the right information at the right moment for every interaction instead of generic talk tracks.
The bad: AI adoption created sprawl before it created leverage. More tools meant more places for reps to look, which, for a while, hurt productivity instead of helping it. We also learned that an agent built with bad data doesn't fail quietly. It fails confidently, and confidently wrong is worse than not knowing.
The wins were real, but they only showed up after we fixed the foundation. Every result that looked like an AI win was a data and process win wearing an AI label.
Why the cost of tool sprawl matters more than the tools themselves
Let’s talk more about sprawl. Initially, adding more AI tools and functionality created more surface area, not less. Reps had more places to look, more systems generating signals, more dashboards claiming to tell them what to do next. The tools that were meant to create leverage fragmented attention and slowed execution.
So, we deliberately narrowed the surface area where reps do their work. Instead of adding another tool every time we solved a problem, we asked where the rep should be living day to day, and pushed the AI and the insights into that single surface instead of building a new one.
As a result, reps spent less time figuring out where to look and more time executing. Productivity went up not because we added more AI, but because we made the AI that we had usable. Sprawl was the hidden cost nobody was measuring, and fixing it mattered more than any individual tool we deployed.
Why AI should only make recommendations
AI can recommend what to do. It cannot be accountable for the outcome. When a decision carries real consequences for a person or a relationship, accountability must sit with a human, not a model.
So, AI informs decisions that benefit from large-scale pattern recognition:
- Pipeline scoring
- Deal health
- Forecast roll-ups
- Churn risk signals
AI can process more signals than any human could track, surfacing what deserves attention.
Human judgment remains for calls involving trade-offs that data cannot fully capture. This includes:
- Decisions about how to handle a struggling rep versus a struggling territory
- Pricing exceptions that weigh relationships, not just the math
- POMP plan design where incentives shape behavior in ways models won't predict
- People decisions — who is ready for more scope, who needs support, and who is a flight risk versus someone who is just having a hard quarter
Why revenue leaders must ask if “more” is actually needed
AI-generated mass outbound has been a disappointment. The promise was scale: more emails, more personalization at volume, more pipeline. It delivered noise. Buyers can tell when a message was generated for everyone and personalized for no one, and response rates prove this.
The broader lesson is that AI excels at scaling what already works. It struggles to manufacture relevance from nothing. Volume was never the constraint in outbound. Relevance was. Relevance still requires a human decision about who matters and why, before AI touches the message.
That's the pattern I watch for now. When we use AI to do more of something, I ask whether “more” was the actual problem. Usually it wasn't.
Why data foundations matter most in AI transformations
AI doesn't fix a broken data foundation; it exposes it. Every CRO I talk to wants to start with the exciting part: the agent, the copilot, the tool that promises to change rep behavior overnight. But if the data underneath is fragmented or untrustworthy, AI just automates the dysfunction faster and with more confidence.
In the end, the CROs who get burned aren't the ones who moved too slowly. They're the ones who bought the tool before they fixed the foundation, then lost credibility with their team when the output was wrong, and nobody trusted it again.
That trust is hard to rebuild once it's gone.
How to prevent perfectionism from derailing AI implementation
Trying to be perfect was the wrong instinct. I spent too much time early on trying to map every edge case before anything went live. We’re talking weeks spent building things nobody needed, and a much longer wait before reps actually trusted them.
Now, we pick one narrow problem, ship it to a handful of reps, and let usage tell us what to build next.
We track actual usage, not satisfaction surveys. Did reps open it? And, did they act on what it told them? Did they come back? This tells us more in one week than any planning session could.
If reps don't use something, we remove it — fast. Even if we’ve already built it. Sunk cost isn't a reason to keep it.
And we expand in narrow steps. Once the smallest version works and reps trust it, we widen the rollout and add the next piece, instead of launching the full vision at once.
In short, move fast, learn fast, and scale what works.
Advice for revenue leaders
The job isn’t to automate decisions away from humans. It’s to give humans better information faster, so the decisions they still have to make are sharper.
Here’s my advice to revenue leaders:
- Fix your data before you buy another tool.
- Resist the pressure to adopt everything at once. The right AI, in the right place, that reps actually use — that’s the goal.
- Hold onto judgment. The job isn't to automate decisions away from humans. It's to give humans better information faster, so the decisions they still have to make are sharper. That distinction is the whole job right now.
Follow along
You can follow Mollie Bodensteiner on LinkedIn. And check out The RevTech Review.
More expert interviews to come on The CRO Club!
