AI Guardrails for GTM Teams:
Lessons From Kore.ai’s CMO and VP of GTM Strategy
With Peter Mullen, CMO, Kore.ai + Raksha Koneru, VP of GTM Strategy, Kore.ai
Episode Summary
Every marketing playbook Peter Mullen built over 20-plus years, including the ones that worked at Netflix, is going out of date in real time. On this episode, Kore.ai’s CMO and his VP of Go-to-Market Strategy, Raksha Koneru, open up about what happens when AI hands every seller the power to build their own decks, pull their own prospect lists, and write in 90 seconds what used to take three weeks, while buyers themselves show up already informed, moving in paths Mullen calls “squiggly lines.”
They walk through the real fallout: fractured prospect lists, credibility risks buried inside AI-scraped content, and a governance structure Mullen now owns company-wide. Koneru shares how she measures whether a new AI tool actually earns its place, and the weekly ritual that keeps her and Mullen honest with each other’s numbers. Along the way, Mullen makes the case for why today’s CMO has to get back into the role of an operator again. It’s a candid look at marketing leadership when the ground refuses to stop moving.
What to Learn
Trust between a CMO and their data lead matters. Knowing why a number changed matters more than just seeing the number. That kind of trust helps a CMO make smarter calls about where to spend time and money.
A tool working isn’t the same as a tool succeeding. A tool can run perfectly and still fail if nobody uses it. Real success means tracking whether people actually use it and asking why early on if they don’t.
Buyer journeys aren’t straight lines anymore. Buyers don’t follow one fixed path from start to finish. They move in unpredictable, “squiggly” ways, so marketers need to keep adjusting instead of sticking to one plan.
CMOs need to get hands-on again. Marketing leaders today need to check the data and the CRM themselves, not just watch from a distance. They still keep doing public things too, like speaking on stage or building the brand’s voice.
AI speeds things up, but it needs guardrails. Giving teams AI tools to build decks, research prospects, and write content gets work done faster. Without some ground rules, though, it can create messy, conflicting data and hurt the brand’s credibility.
What is this Episode's Playbook?
- Define the Outcome
Define the result before you touch the tool: cut cost, save time, fix one exact breakdown. “Use AI” isn’t a result. - Rapidly Deploy
Let teams create without routing every asset through marketing first. That’s the opportunity, not the risk. - Keep Judgment in the Loop
Decide in advance what nobody gets to improvise: a disgruntled ex-employee’s Reddit post shouldn’t end up in a Fortune 500 pitch. Define your sources, your claims, your line for human sign-off. - Fast Feedback Loop
Ask why a number moved this week, not once a quarter, the way Kore.ai’s own weekly “living the data” call does. - Tune and release
A tool that works and nobody uses hasn’t succeeded. Waiting three months to find out is how you end up cleaning up a mess instead of catching it early.
Episode Highlights
When Everyone Can Build Their Own Deck
Kore.ai’s sales team doesn’t wait on marketing to build a pitch deck anymore. Using tools the company built in-house, any seller can generate their own presentation, decide what to show a Fortune 500 buyer, and have it ready in minutes. Peter Mullen talks about that shift as a fact to manage. The tools work, and they work well enough that a seller can press a button and walk away with something genuinely usable. But “usable” and “accurate” aren’t automatically the same thing, and a deck built at that speed can just as easily carry a claim nobody checked as one that’s rock solid. That gap, between what a tool can produce and what a marketer would actually sign off on, is the starting point for almost everything else Mullen and Raksha Koneru talk about on this episode. It’s already happening, every day, across a global sales team.
We have the ability now for our incredible sales team around the world to use tools we have built to develop their own decks, to decide what do they want to present tomorrow to a Fortune 500 company. And they can press a button and they can deliver some pretty good work, but they can also go a little bit sideways.
Loosening the Grip Without Losing Credibility
For years, Kore.ai’s marketing team controlled what sellers said and how they said it through a message library built in tools like Seismic and Highspot: approved language, consistent cadence, a single voice. AI tools built for speed don’t wait for that kind of review, and Mullen is candid that the company has had to loosen its grip on a system it spent years building. The checkpoint stays in place regardless. Mullen’s line is specific: whatever a tool produces, a human still has to look at it before it reaches a buyer. He points to a concrete failure mode to explain why. A tool scraping the open internet for talking points doesn’t know the difference between a verified fact and a disgruntled former employee’s post on Reddit. Without a person checking the sourcing, either one can end up in a deck presented to a Fortune 500 company. That’s the actual shape of a guardrail at Kore.ai: someone checks the sourcing before anything goes out the door.
We have built, whether it’s in Seismic or Highspot or something else, a very strong library that we’re proud of: what should be said, how we say it, what is the cadence. Part of what has changed is that we’ve got to loosen up our grip on a lot of these things. However, you still have to have that human in the loop to come back and make sure what is being produced is credible.
Good Data Is the Real Bottleneck
Raksha Koneru built an account research agent to cut down the hours Kore.ai’s sellers spent prepping before a first meeting with a prospect. The hard part came after the idea: making sure the agent had something reliable to work from. An agent is only as good as what feeds it. Koneru’s point is that a genuinely good idea for an AI tool can still fail if the underlying data is messy, incomplete, or untrustworthy, because nobody adopts a tool that hands back bad answers with confidence. Getting the data clean enough to trust came before anything else in building the tool sellers would actually use. That’s a different question than most teams ask first: is our data good enough for this agent to be worth using at all, before anyone gets to what the agent can do.
All ideas are really, really good as long as your data is good, and how do you get your data to be good enough for this agent to be effective and for people to adopt it?
The ZoomInfo-Yourself Problem
A few years ago, Kore.ai had one process for building a prospect list, and one list came out of it. Today, any rep can build their own list solo, using a personal AI tool, in an evening. Mullen’s example is almost funny until you think through what it means: a seller with a personal AI tool and a glass of wine in the evening can produce solo what used to take a defined process and a shared system to create. The result is a set of lists that no longer overlap. Sales ends up with a list. Customer success has its own. Marketing has a third. None of them match, and nobody’s fully sure which one reflects reality. Mullen frames this as what happens when individual capability quietly outpaces organizational coordination. Fixing it means rebuilding a shared source of truth that absorbs what any one person builds on their own into a single system everyone shares.
Today, anybody can go out and capture. You can ZoomInfo yourself just by using one of your own personal AI tools with a glass of wine in the evening. You can get your own list. And this is actually a real problem, because this empowerment is leading to the fracturing, the lack of governance, the lack of control.
Adoption Is the Only Metric That Counts
Koneru’s account research agent worked, technically, from the day it launched. What she actually measured was whether sellers opened it, used it, and kept coming back to it after the initial novelty wore off. That standard comes directly from how Kore.ai sells to its own customers. The company’s pitch to buyers centers on ROI and outcomes actually showing up, well past whatever a feature demo can prove. Koneru holds her own internal tools to that same bar: a solution earns success through real usage, regardless of how well it was built technically. That framing changes what counts as QA. Koneru’s real question is whether people are getting value from a tool, and if they aren’t, why not.
What really matters is adoption. Once you actually release a solution, are people using it? Are they finding value out of it?… We sell ROI. We sell outcomes.
The CMO Has to Get Back in the CRM
Mullen has had the kind of career that could let him stay at a comfortable distance from daily execution: 20-plus years building marketing playbooks, stage time, a strong external voice for the brand. He argues that distance is exactly what a modern CMO can no longer afford. His own habit makes the case concretely. He keeps a running tracker of his top priorities, and it used to run three pages for most of his career. Recently, it hit eight, a direct measure of how much more a CMO has to track now that AI touches every part of the job at once. Mullen is careful to frame this as additive to the public-facing parts of the role: the stage appearances, the LinkedIn presence, the company’s external voice all still matter. What’s changed is that none of that is enough on its own anymore. A CMO also has to be willing to sit down with the person running go-to-market strategy and look directly at what the CRM is actually showing.
I think a CMO today needs to be an operator. I think a CMO needs to roll up their sleeves and be able to look at the AEO results… and needs to sit down with Raksha and say, ‘Let’s open up the CRM, and I want to see what is happening inside there.’
Trusting the Operator’s Read on the Data
Every week, Mullen and Koneru sit down for what they call a “living the data” call. It’s not a status update. Koneru’s rule for that meeting is that a number by itself doesn’t actually tell you anything useful. A conversion rate being low or an MQL number hitting target only matters once someone’s dug into why it moved. That’s a harder standard to meet than it sounds, because it requires trust running in both directions. Koneru has to trust that when she brings Mullen a read on the data, he’ll actually act on it. Mullen has to trust that what she’s bringing him is accurate enough to make a real budget call on, like pulling spend from an initiative that isn’t working and moving it somewhere that is. It plays out closer to a standing agreement that the “why” behind a number gets real weight every week, on the weeks nothing’s visibly wrong too.
I think every CMO really needs to trust their operator, and they need to empower their operator to bring the insights to them the way they see it.
Related Resources
Episode FAQs
At Kore.ai, it’s the tension between letting sellers self-serve decks with AI tools and keeping the brand’s content credible. Mullen’s own example: a tool scraping the internet for talking points can surface an ex-employee’s Reddit post as easily as a verified fact, and only a human check catches the difference before it reaches a Fortune 500 buyer.
At Kore.ai, it means one accountable owner, in this case the CMO, across the whole company’s AI use. Mullen describes his role in that as “connecting the dots and connecting people across teams.”
Because any rep can now build a list solo with a personal AI tool in an evening, bypassing the process that used to produce one shared list. Mullen’s read: sales, customer success, and marketing each end up with their own version, aligned to none of the same goals.
Track usage starting the week it launches, and ask people directly why they’ve stopped using something. That’s Koneru’s approach with her account research agent.
Kore.ai’s clearest example is Koneru’s account research agent, built to cut the hours sellers spent prepping before a first meeting and measured by whether sellers actually opened it before that meeting.
The standing question behind Kore.ai’s weekly living-the-data call is why a number moved. That’s what lets Mullen make an actual budget call.