--- title: "What stays human when RevOps goes AI-native | In conversation with Anthony Enrico" description: "Anthony Enrico of Leanscale ran a company-wide hackathon, rebuilt how his team works, and arrived at a conclusion that surprised him: the agents were never the hard part. Here's what actually determines whether AI in RevOps delivers or disappoints." canonical: "https://vasco.app/blog/ai-native-GTM-anthony-enrico" date: "2026-09-03T21:55:00.000Z" authors: - Anthony Enrico jobTitle: CEO and Founder readingTimeMinutes: 8 contentType: video intent: strategy-insights pillar: "RevOps maturity & scaling journeys" audiences: - cros - founders - fractional --- # What stays human when RevOps goes AI-native | In conversation with Anthony Enrico Anthony Enrico has spent five years building Leanscale, a RevOps agency for VC and PE-backed companies. He ran a company-wide AI hackathon in December 2024, watched his assumptions about agents get inverted, and spent the next nine months rebuilding how his team operates. We sat down with him to find out what actually changed, what the mistakes look like from inside, and what RevOps leaders still can't hand to a model. Watch the full replay: [The data model is 90% of AI-native RevOps | Anthony Enrico, LeanScale](https://www.youtube.com/watch?v=WcPjQ_bQcvs&t=2918s) ## Where most RevOps teams are right now In December 2024, Anthony flew the entire Leanscale team to Arizona for an off-site hackathon. He came back awestruck at what they built. Claude Max accounts went out the following week. Their CTO trained every person on how to use them. The instruction: stop planning and start building. Most companies are in that same window right now, without an off-site to inspire the shift. The poll we ran at the start of this webinar told the story plainly: - **33%** hadn't started yet - **33%** had connected AI to their tools but weren't seeing results - **33%** were actively building - **0%** said they were fully AI-native and iterating One-third of practitioners have built the wiring and gotten nothing. A foundation gap explains every one of them. Let’s get into it. ## Why agents fail without a data model Every stuck AI deployment runs the same sequence: 1. The board issues a mandate. 2. The company distributes licenses. 3. Reps experiment. 4. Workflows appear in Slack. It looks like momentum until someone asks the agent a question and gets a confident, wrong answer. Ask any team: how much pipeline did we build last week? You'll get ten different answers depending on how each person defines pipeline. Put an agent on that ambiguity and you don't get ten uncertain answers — you get one certain wrong one. The models aren't the issue. Anthony's analogy: an F1 car with no map. The most common mistake is starting with agents before building the data model they'll run on. ## How to build a semantic layer That 33% who connected AI and got nothing back all ran into the same wall. They plugged agents into their data without first defining what the data means. The agent looked at a field called "pipeline," made its best guess, and returned something different from what the CRO would say — and different again from what finance would say. The fix is a semantic layer, or a set of written definitions that tell your agents what your metrics actually mean. This means any definitions your business uses to measure success. "It's really just writing down definitions," Anthony said. "It's not anything more fancy or intellectual than that." The exercise runs three steps: 1. **Identify your contested terms.** SQL, pipeline, churn, customer, ICP: whatever your go-to-market motion depends on. If two people in a room would define it differently, it goes on the list. 2. **Write the definition your business actually uses.** Your version: "a customer is anyone with a Salesforce record marked account type: customer and an active billing record in QuickBooks." 3. **Connect the definition to the data.** The agent reads the definition first, pulls the data that matches it, then answers. Ask five times, get the same number. This is what Vasco does at onboarding: walk through every go-to-market definition before any agent runs on top of it. Anthony's team starts every new engagement the same way: semantic layer first, data model second, agents third. This is also why RevOps owns this problem and not data engineering. Data engineers know the schema, while RevOps knows what the metrics mean to the business. Those are different things. ## How the context graph connects your data The semantic layer tells an agent what things mean. The context graph tells it how they connect. For a single customer account, Leanscale ingests Slack data, call transcripts, CRM records, project management activity, and external GTM performance signals, all tied back to the same account. Every signal routes to the right place, so agents draw from a complete, accurate picture of each relationship. "The semantic layer has the definitions," Anthony said. "The context graph is connecting all the data points." The practical result: a go-to-market diagnostic that used to take weeks now takes 48 hours. Not because the analysis got faster — because the foundation makes the analysis possible without the manual download, pivot, and rebuild cycle. ![semantic layer and context graph](https://cdn.sanity.io/images/ys8gstp8/production/8aa7e3b1b890920322f559d71ef1f63fc7b0a1f9-2400x928.png?w=1600&fit=max&auto=format) **Related resources:** - [Context graphs for GTM teams](https://vasco.app/blog/context-graphs-for-gtm-teams): Build one without a data team. - [Revenue data provenance: A forecast accuracy guide for CROs](https://vasco.app/blog/data-provenance-forecast-accuracy): Trace numbers before your next QBR. - [Centralized vs. decentralized AI in GTM: In conversation with Kyle Norton](https://vasco.app/blog/build-or-buy-gtm-context-layer): Kyle Norton on context that compounds. ## What AI can handle today Once the data model and context graph are in place, a specific set of tasks moves to AI. These tasks were always better suited to a machine — teams did them by hand because they had no alternative. ADD TABLE Data assembly bottlenecked every task in this list. The thinking was fast. Getting the data into a usable shape was slow. Once the context layer exists, that work disappears. ## What stays human The human list is shorter than it sounds, but the things on it are irreducible: ### GTM strategy AI can surface research, run scenarios, and stress-test assumptions. Owning the call is different. Understanding the context, the trade-offs, and the stakes still requires someone accountable to the outcome — someone who can explain not just what to do, but why. ### Business process design Building a go-to-market motion that fits a specific company — their ICP, their team structure, their stage — takes lived experience and judgment. An agent can help document the process once it exists. It can't invent it from scratch. ### Accountability This one is underrated. Customers still want someone they can trust before making a big decision. These are high-stakes investments. If Leanscale makes a bad call for a customer, they can be held to it. Nobody's calling Anthropic. ### Stakeholder alignment Getting the board, CRO, and cross-functional team pointed in the same direction requires political context, relationship history, and timing judgment no model has access to. Judgment and accountability are the human work, while execution and synthesis are getting faster and cheaper every quarter. ![RevOps humans VS. AI](https://cdn.sanity.io/images/ys8gstp8/production/05ffd33a1135d6adab765cd6369563162c920acb-2400x1008.png?w=1600&fit=max&auto=format) ## The performance-to-plan blind spot Most companies build the annual plan in November and stop looking at the assumptions by February. Teams reverse-engineer the ARR target into a bookings plan, a pipeline plan, a churn plan, conversion assumptions, and velocity targets. Then the year starts, and those assumptions sit in a spreadsheet while live data lives in the CRM. The gap between them stays open all year. For AI to measure plan variance, the context layer needs to know what the plan assumed: which conversion rate, which sales cycle length, which velocity targets. A pipeline number shows what's in the funnel. Knowing whether that's enough requires the plan's assumptions to live in the same layer as the live data. Define those targets in the context layer and two numbers become visible in real time: - **Conversion rate.** A 1-point improvement on a $20M plan is $1M. A 1-point drop is the same loss, and it won't surface in the aggregate pipeline number until Q4. - **Sales cycle length.** One fewer day in a 90-day cycle adds roughly $250K. Nine days off plan is a Q4 miss already sitting in your velocity data. With those targets defined, an agent reports that your conversion rate is 3.2 points below assumption and shows exactly why the current pipeline won't close the year — rather than a graph that looks fine until someone checks the Y-axis ## Why RevOps teams win with AI Nine months before this webinar, Anthony thought LeanScale might not be relevant in a year. He assumed the future belonged to agent workflow builders, and that RevOps agencies would have to become software shops to survive. The opposite happened. "I can spin up an agent in 10 minutes. I can build a very, very good forecasting agent and workflow, if I have the data model. My value has been: the data model is 90% of it. This agent-building part? That's 10%. And anybody can build this." The bottleneck in every AI deployment is the same thing RevOps has always owned: context. Metric definitions. ICP logic. Attribution rules. Lifecycle stages. The business understanding that makes any AI answer accurate instead of just plausible. The webinar audience confirmed it. When we asked what their biggest concern was about going AI-native, 50% named the data foundation. Technology, agents, and budget barely registered. That's the RevOps problem. Companies that understand this are already leaning on their RevOps leaders to run AI strategy at the executive level. Anthony has watched heads of RevOps and VPs of RevOps move into COO and CRO roles as a direct result of this shift. "If you're sitting in RevOps right now, just soak it up. Take the opportunity. Raise your hand. Lead these strategies. It'll open up doors." ![Humans vs AI with a context graph](https://cdn.sanity.io/images/ys8gstp8/production/7a0276932ce3b03afe6e3149a1854c20e09b1fee-2400x1096.png?w=1600&fit=max&auto=format) ## Two companies, one year from now Two RevOps leaders are reading this today. **Company A** spends the next six weeks doing the unglamorous work: writing down definitions, auditing the data model, connecting CRM to Slack to billing in a context graph, and baking plan targets into the layer before a single agent runs on top of it. By Q1, their forecast takes minutes. Their CRO self-serves answers instead of waiting three days. They catch conversion rate variance against plan the week it starts drifting — weeks before it becomes a miss. **Company B** buys licenses, lets reps experiment, and waits to see what sticks. By Q1, they have a dozen disconnected workflows and answers no one quite trusts. The board asks about AI strategy. The response is a Slack thread. Six weeks of unglamorous work separates them. The data model is either your foundation or your bottleneck — and right now, it's one or the other. If you want to hear Anthony walk through how Leanscale actually made this shift — the hackathon, the data model work, the specific workflows they rebuilt — watch the full webinar. ## FAQ ### What is the semantic layer in RevOps? The semantic layer is a set of written definitions that tell your AI agents what your data actually means — what "pipeline" is, what "a customer" is, what a SQL or a churned account looks like in your business specifically. Without it, agents infer — and the "confidently wrong" answers you get back are the result. With it, agents retrieve the exact numbers behind a specific definition and return a deterministic, auditable output. ### How do you teach an AI agent to answer "why" questions about revenue? Start by zeroing in on the root cause metric. If sales velocity dropped, break it into its three components — conversion rate, average contract value, and sales cycle length — and identify which one moved. Then let the agent correlate against everything that changed in that window: call transcripts, competitive mentions, rep activity, external signals. The "why" emerges from the intersection of a precise metric definition and a broad data set. ### What AI tasks can RevOps actually delegate today? The ones that work reliably: forecast rollups, proposal and SOW generation, customer health scoring across multiple data sources, go-to-market diagnostics, and performance-to-plan gap analysis. These work because they run on structured data with clear definitions. Tasks that still need a human: strategy decisions, business process design, stakeholder alignment, and anything where accountability matters. ### What stays human when RevOps goes AI native? Three things reliably: strategy (what to do and why), business process design (what the motion should look like), and accountability (someone who can be held responsible when a big call goes wrong). AI can surface options, synthesize data, and run the analysis. It can't own the decision or take the call from the customer when the recommendation didn't work. ### How has the RevOps function changed with AI? It's become more strategic, not less. The execution work — data analysis, reporting, diagnostics — is getting faster and cheaper. That's freeing up RevOps leaders to do the thing agents can't: understand the business deeply enough to make the data definitions accurate in the first place, and own the context layer that every agent draws from. Companies are already leaning on RevOps leaders to run AI strategy at the executive level. ### How does Vasco support an AI-native go-to-market motion? asco is the revenue data layer purpose-built for AI agents. It resolves identities across your revenue systems, holds your metric definitions in one place, connects plan targets to actuals, and grounds every agent you deploy in a shared context graph of 750+ sources. You don't have to build the data infrastructure before you can get accurate answers — Vasco is that infrastructure. ### What is a context graph and why does RevOps need one? A context graph connects data from every revenue-related system — CRM, call transcripts, project management, billing, Slack — and links it to the account it belongs to. Without one, agents query systems in isolation and miss context that would change their answer. With one, every agent draws from the same grounded, connected version of reality. The semantic layer says what things mean; the context graph shows how they relate.