Most RevOps teams are asking the same question right now: when it comes to AI, do we build our own solutions, buy from vendors, or string together some combination of both? The answer, as three practitioners made clear in a recent RevOps Co-op webinar, is almost always "both" — but the framework you use to decide which is which matters enormously.
Matthew Volm, CEO and Co-Founder of RevOps Co-op, moderated a panel with Ross Rich, CEO and Co-Founder of Accord; Matt Flotard, VP of RevOps at Gong; and Matt DeLauro, President of Go-to-Market at SEON. Together, they unpacked the decision frameworks operators are actually using, called out where AI is delivering real productivity gains, and issued a clear warning about the gap that's quietly undermining most AI initiatives.
The panel wasted little time getting to the core tension. Ross Rich framed it clearly: the question isn't really build or buy, it's about knowing which part of the value chain you should own and which you should outsource.
"The best operators are thinking about how they consolidate all of the product usage data, the conversational intelligence, the understanding of their customers — the why they buy, their unique value props for the right segments, GEOs, and products — and using that to power the workflows of their outbound messaging, their pitches, their deal reviews." — Ross Rich, CEO and Co-Founder at Accord
The implication: own your intelligence layer, leverage third parties for workflow execution. Building your own version of Gong Engage or Outreach is a losing proposition — not because it can't be done, but because of what it costs to maintain.
Matt Flotard added a practical three-part filter his team uses at Gong: How difficult would this be to build and maintain to a standard that won't make the InfoSec team lose their minds? How much ongoing resource does it require? And — unique to Gong's position — does this sit on the product roadmap anyway, making internal experimentation a form of customer-zero feedback?
Matt DeLauro took a data-quality lens to the same problem. His rule of thumb: if a vendor has high-quality first-party data with clear provenance — data that comes from direct customer interactions rather than rep field updates — that's a strong signal to buy.
"Going into it, the quality of the data that you're gonna use to train whatever outcome is super important. It's just garbage in, garbage out." — Matt DeLauro, President of Go-to-Market at SEON
His team at SEON learned this the hard way when an early agentic deployment for market research ran into CRM data quality issues that undermined the whole initiative. The lesson stuck: data provenance matters as much as the AI itself.
This aligns with broader conversations in the RevOps community about thinking data-first before reaching for AI tooling.
When the conversation turned to what's working in practice, DeLauro introduced what might be the most useful mental model of the session: the horseshoe graph.
AI initiatives produce the best outcomes — and carry the lowest risk — at the two extremes of the customer lifecycle. Very early in the funnel (target account strategy, enrichment, market research) and very late in the customer journey (product usage analytics, churn signals, automated QBR narratives) are where teams have the highest-quality data and the most tolerance for iteration.
The middle of the funnel — active deal execution — is a different story. The data is messier, the stakes are higher, and the cost of a bad AI intervention is real. That's exactly where DeLauro's team leans on focused vendors with deep workflow expertise, including tools like Accord for deal execution and Gong for conversation intelligence.
"In the middle of the funnel, we're really looking to our vendors who are super focused on specific aspects of the customer engagement lifecycle to make our best people more efficient." — Matt DeLauro
Flotard shared a concrete example of AI working well in the middle of the funnel when the implementation is precise: Gong's AI sales stage feature. Rather than replacing the forecast, it adds a parallel AI-assessed stage that surfaces deal health signals from unstructured call data — changing the conversation in forecast reviews from "where do you think this is?" to "here's what the data suggests, and here's why."
"It changes the framework of the call completely," Flotard explained. "You have this third-party input suggesting the deal is at a later stage based on thousands of deals and unstructured data."
For RevOps teams thinking about where to start, this horseshoe model is a useful prioritization guide. For more on how AI is reshaping territory and account coverage decisions — another area where early-funnel data quality matters — see How AI and Automation Are Changing Territory Design.
The panel spent significant time on the seismic consolidation happening across the go-to-market technology landscape. The merger of SalesLoft and Clary, PE-backed rollups in content management, founder departures at once-prominent tools — the message is that the technology map operators built their stacks on is being redrawn in real time.
DeLauro's framing was direct: "You're either an orchestrator or you're getting orchestrated." Vendors that add proprietary first-party data or drive meaningful workflow orchestration have a defensible position. Vendors that sit in between — pulling data from a CRM and running statistical calculations to produce an output — are in trouble.
"If you're not adding proprietary first-party data or orchestrating a workflow of some kind, it's gonna be really tough for the next couple of years." — Matt DeLauro
Rich extended this into a broader product philosophy argument. The old model — walled gardens, everything must happen in our system, we need to own every interaction — is a losing strategy. The vendors winning right now are the ones playing well with others: open APIs, MCP compatibility, genuine data-sharing with adjacent tools.
"I think every call I'm having with more sophisticated operators — you don't need to be everything anymore. You can add a lot of value, you can still be sticky. But there's this old product philosophy that you need to take up more time in-app. I think that is a losing strategy going into the future." — Ross Rich
He pointed specifically to Salesforce's handling of Slack data as a cautionary example of a walled-garden instinct that operators find increasingly frustrating.
DeLauro added a buyer-versus-builder distinction that captures where the market is heading: "A lot of people built tools for buyers. The big change is now everybody's becoming a builder. You have to be cross-compatible, friendly — open APIs — because you have to be open to fitting into somebody's strategy."
For teams navigating the complexity of rebuilding or rationalizing a RevOps tech stack, Episode 48: Rebuilding a RevOps Tech Stack and the post on fixing tech bloat without making enemies are worth revisiting.
Perhaps the most important insight of the session came from an exchange that reframed the entire AI productivity conversation. Rich pushed back on the idea that faster reports and automated QBR decks represent real impact — if no one was actually using those reports to make decisions before, you've just made a zero-impact process faster.
The real limiting factor, he argued, isn't data or tooling. It's whether organizations have the people and processes in place to act on what the technology surfaces.
"I almost feel like the easy stuff is the technology, is the data, is the insight. And the culture and the people and the process stuff is really hard and is kind of being left behind." — Ross Rich
His framework was sharp: estimated impact equals insights and data multiplied by the ability of your people to actually use that information. Most organizations are investing heavily in the numerator and neglecting the denominator.
DeLauro illustrated the problem through the lens of process definition. The best reps don't need a system to tell them the 30 micro-behaviors that move a deal from stage to stage — it's pattern recognition built through experience. But training an LLM to replicate that requires explicit documentation of things that most sales processes have never bothered to write down.
"If I were to train a model, our level of process definition at this point is probably not sufficient to do the 30 things that a rep needs to do on every deal to actually be successful. Like those things aren't in the process." — Matt DeLauro
This echoes a challenge RevOps leaders have faced long before AI entered the picture. As Rich put it, "That's been a problem since well before AI — software buyers thinking that buying a software tool is gonna fix a broken process or one that's not well documented." The AI era has compounded the complexity, but it hasn't changed the underlying dynamic.
Episode 8: The One Constant Issue Spanning Maturity: Change Management explores exactly this phenomenon across company stages.
With the cautionary notes on the table, the panel also offered concrete examples of AI deployments that are delivering measurable results.
At SEON, DeLauro's team is generating automated customer health narratives — quantitative, personalized summaries of how each customer is performing with the product — at a scale that was previously impossible. What once took a customer success manager several days to assemble for a single account now runs across hundreds of customers continuously. The key: using high-quality first-party product usage data with clear provenance, not messy rep-logged CRM fields.
At Gong, Flotard described using AI to validate territory coverage hypotheses — specifically, whether "buying power" was actually being discussed on calls with accounts the team had flagged as high-priority. By triangulating structured data (firmographics, enrichment from data vendors) against unstructured call data, the team could confirm or refute coverage assumptions with a level of confidence that wasn't previously achievable.
Both use cases share a common structure: they start with a clear, measurable question; they use high-quality data; and they produce an output that a human can verify and act on. That last piece — the human-in-the-loop — came up repeatedly as essential, especially in early deployments.
Accord's approach to buyer engagement and mutual action plans reflects this same principle: the value of the tool comes from surfacing the right information at the moment a seller is engaging with a customer, not from generating dashboards no one reads.
The session closed on the question that's generated the most LinkedIn heat recently: is the CRM dying? Could something like Claude Code or Clay eventually replace Salesforce or HubSpot entirely?
The panel's consensus: probably not as a system of record, but the percentage of seller time spent inside a CRM is going to decline significantly — and that's a feature, not a bug.
"Will less people be spending less time in core systems over the next five years? Absolutely. But it doesn't mean that it's replaceable in that sense." — Matt DeLauro
The CRM's moat, the panel agreed, isn't the UI or the field structure. It's the accumulated business logic — the integrations with ERP, configure-price-quote (CPQ), and billing systems; the permissioning structures; the financial reporting dependencies. That infrastructure took decades to build and can't be replicated in a weekend sprint.
The more interesting question is who owns the layer between seller and CRM. Rich noted a meaningful shift he's observed since founding Accord: operators who used to argue for the CRM as a system of action are now much more comfortable saying the system of action should live elsewhere and sync back. The question of what owns seller workflows — and what keeps that data structured enough to be useful — is genuinely open.
Flotard put it concisely: "Do we want reps necessarily spending time in a CRM? We want them speaking to customers. You don't want them necessarily going into the CRM updating notes and things like that."
For more on the evolving CRM debate, see Episode 69: Should You Blow Up Your CRM? Here's the Truth and It's Time to Stop Thinking of a CRM as a Selling Tool.
The message from this panel is clear: the question isn't whether to adopt AI in RevOps. The question is whether your people and processes are actually ready to use what AI gives you. For most organizations, that gap — not the technology gap — is where the real work is.
Check out our blog, join our community and subscribe to our YouTube Channel for more insights.
And be sure to check out the Accord blog for more resources on buyer engagement, deal execution, and go-to-market strategy.
.png)
