Most revenue teams chasing AI-readiness are solving the wrong problem first. They invest in agentic workflows, automation tools, and sophisticated orchestration layers — only to discover that the output is only as good as the data underneath it. When the CRM is incomplete, stale, or unevenly enriched, even the most carefully designed AI agent will produce results that erode rather than build trust with the sales and customer success teams it was built to serve.
In a recent RevOps Co-op webinar, Matthew Volm, CEO and co-founder of RevOps Co-op, sat down with Spencer Hardey, who leads Business Operations and RevOps at HG Insights, and Hari Kumar, a product team member at Clay. The session walked through why cleanup projects don't stick, how to prioritize enrichment without blowing your budget, and what a genuinely AI-ready revenue stack looks like in practice — illustrated with real workflows Spencer has built and deployed at HG Insights over the past six months.
An opening poll set the scene. Asked about their biggest CRM data problem, attendees split across three primary concerns: incomplete or missing fields (33%), an inability to identify which accounts actually matter, and data going stale faster than teams can fix it. All three problems, as the session made clear, are connected — and solving one without addressing the others just moves the bottleneck.
The pattern is familiar to most RevOps practitioners. A data cleanup initiative is scoped, approved, and executed. Fields get enriched. Coverage metrics climb. And then, a quarter or two later, the data is stale again, engagement rates are poor, and the team is back at the beginning.
Spencer described exactly this cycle playing out at HG Insights. Early AI agent deployments — including automated AE-to-implementation handoff briefs and account research tools — produced wildly inconsistent results across the sales team. Some reps raved about the quality of the briefs. Others disengaged entirely.
"Even though the skill or the agent that we had designed was consistent across different accounts and different reps and scenarios, the underlying data that we had in our CRM completely drove the results." — Spencer Hardey
The agents weren't broken. The data was. Where account information was comprehensive — intent signals, past interaction history, enriched firmographics — the outputs were excellent. Where coverage was thin or stale, the agents amplified those gaps into confidently wrong outputs.
This is the core problem with treating CRM hygiene as a one-time cleanup project: data decays continuously, enrichment without an activation plan just creates maintenance debt, and AI has no tolerance for the kind of "close enough" data quality that human judgment can work around. As Hari put it plainly, chasing fill rates is not the same as having useful data.
"Don't waste your dollars. I see so many people enrich and do nothing with the data in the end because they're just chasing fill rates." — Hari Kumar
This connects directly to a broader point about how RevOps teams should think about data management — the goal is not maximum coverage, it's actionable coverage on the accounts that matter.
Spencer introduced a maturity framework that reorders how most teams approach the problem. Most organizations default to the enrichment step first — they identify gaps, find a vendor, and fill fields. The framework flips that sequence.
Stage one is access. Before any enrichment happens, teams need a clear picture of what data they actually have, what fields matter for the actions their teams will take, and where the most critical gaps exist. The emphasis is deliberate: not every field on every account needs to be complete. The goal is coverage on the fields that enable execution, for the accounts that will actually be worked.
Stage two is prioritization. Once the critical fields are identified, the question shifts to which accounts require that data most urgently. This means scoring and tiering the addressable market before enrichment begins — so the budget flows to the accounts where it will generate the most return, not the accounts that happen to be easiest to enrich.
Stage three is enrichment. With a clear picture of what matters and for whom, teams can then use tools like Clay and data vendors like HG Insights to fill those specific gaps — and maintain them on an ongoing basis rather than in periodic bulk runs that go stale before they're activated.
Hari reinforced the framework from the product side, framing it as three critical questions: What data do we already have? Which accounts does it actually matter for? And, critically, what's actually working — are the enrichment investments translating into pipeline?
One of the more practically valuable segments of the session was Spencer's account on how HG Insights rebuilt its total addressable market (TAM) definition — and why traditional industry classifications failed them.
The problem is one many GTM teams face: standard industry codes (SIC, NAICS, and similar taxonomies) are too blunt to accurately identify best-fit accounts. At HG Insights, reps were spending time on companies that fell within the right industry category but were structurally wrong fits — fitness technology and aerospace technology companies appearing in the same "technology" classification as their ideal customers. Meanwhile, genuinely good-fit accounts in adjacent categories were being missed entirely.
The solution was to use LLM-based classification against HG Insights' own company description data — a dataset covering seventy million companies — rather than relying on industry codes. The model analyzed what companies actually sell and how they describe themselves, producing a TAM definition grounded in business reality rather than taxonomic convention.
"The biggest insight for us on whether or not they were a good fit for our products and services — we took what is currently in the HG dataset, there's seventy million companies. We pulled the descriptions, not the industry codes. We ran that model and that gave us our total addressable market." — Spencer Hardey
The process was iterative — Spencer noted they ran and refined the model more than ten times before arriving at a definition they trusted — and is now maintained on a quarterly refresh cycle. It's a useful reminder that defining what a good-fit account actually looks like requires more than copying the parameters that were set when the company first built its ICP.
Once the addressable market was defined, HG Insights moved to account scoring — but not with a single composite score. Spencer described scoring each account against thirteen distinct use cases that the sales team leads with, producing an A-through-F tier for each use case rather than a single overall tier.
This distinction matters operationally. A company that is a strong fit for one use case may be a weak fit for another. A single composite score obscures that nuance and leads to misaligned outreach — reps leading with the wrong angle for the wrong account. With per-use-case scoring, the enrichment that follows is also targeted: Clay is used to identify the specific buyer personas and influencers for each use case, and contact-level enrichment is executed only for tier A accounts within the use cases they score highest on.
The workflow is then connected to a specific use case campaign — enrichment feeds activation, not just coverage metrics. This is the core of the framework's logic: enrichment that isn't tied to a specific action is budget that will go stale.
Spencer also flagged two data points that most teams underutilize in scoring: spend category data (how much a company spends in the category you serve, which indicates actual budget availability) and AI and data maturity scores. For a business selling data and intelligence products, understanding where a target account sits on the AI maturity spectrum meaningfully changes both the angle and the urgency of outreach.
Hari walked through what a complete account data stack actually looks like — and how most teams are working with only the bottom layer of it.
The foundation is the company and people card: basic firmographics, contact information, email addresses, previous experience. This is the data most teams have and most enrichment projects stop at. But it's the layers above this that drive differentiated execution.
The next layer is technographics and signals. HG Insights brings technographic data — what tools an account is currently running — which enables use cases like competitive displacement outreach tailored to the specific CRM or platform a prospect is using. Intent data adds another dimension: accounts showing buying intent can be routed automatically into ad campaigns, email sequences, or direct rep outreach, depending on their fit tier.
First-party data sits above that: the engagement history, product usage signals, and customer lifecycle data that lives in the data warehouse and CRM. Hari's observation was that the teams seeing the most success are those who have built systems that keep this layer continuously fresh and connected to their orchestration tools — so that an agent acting on account data is always working with current information, not a snapshot from three months ago.
At the top of the stack is AI as a fallback: research agents that can connect the dots between technographic signals and intent when the relationship isn't immediately clear, or that can surface new data points that no vendor directly provides.
"You can't just think about one piece of the data. You can't just think about my fill rate for company domains. You actually need to think about in concert for the play that I'm trying to run, what is all the data I need, and how can I scalably refresh, check, and make sure that I don't have stale data in my CRM?" — Hari Kumar
The practical architecture Hari described is one where feedback loops are built in by design. If an agent running a workflow discovers it's missing a critical technographic data point, it can call out to HG Insights via an MCP connector, fill the gap, update the CRM, and continue its task. The data infrastructure is self-healing rather than dependent on periodic manual enrichment runs. This is a meaningful shift from how most RevOps teams currently operate — and it's directly relevant to the broader conversation about what AI-ready revenue systems actually look like.
One of the most concrete illustrations in the session was Spencer's description of the automated account brief that HG Insights now produces for its sales team — and how impossible that would have been eighteen months ago.
The brief is triggered through the HG Claude MCP connection. A rep can simply ask for an account brief on a named account, and the system returns a structured document covering firmographic fundamentals, surging intent topics, AI and data maturity scores, tech stack details (including competitive intelligence and integration partner flags), and — critically — which of the thirteen use cases the account scores highest on, along with specific angles and suggested openers tied to that use case.
What the brief deliberately does not do is write the messaging for high-priority accounts. Spencer was explicit about this: for tier A or B accounts where hyper-personalization is required, the rep needs to craft the outreach themselves. The brief saves them the research time, surfaces the angles, and provides the context. The engagement strategy remains a human responsibility.
"The reps are gonna wanna craft the messaging themselves, and they should, as that's their expertise. But providing and saving them time on all that additional research is where we were able to drive the most value in terms of leveraging AI and automations." — Spencer Hardey
Hari added that QBR decks are another workflow seeing rapid adoption — with first-party data connected to an orchestration tool, AE prep documents can be auto-generated ahead of quarterly calls, with direct quotes from customers in the platform noting time savings of three hours per AE per QBR cycle.
Pre-AI, Spencer noted that HG Insights could only produce a single composite account score and tier. Now they run thirteen. Meeting type tagging — previously dependent on manual rep input and never reliably complete — is now handled automatically via AI, giving the operations team full visibility into meeting mix and post-meeting outcomes. The shift is less about individual features and more about what becomes operationally tractable when the underlying data is clean and structured enough for AI to act on reliably. This matches what practitioners are finding across the industry, as explored in Episode 94: The Boring Work Behind Great AI.
Volm pushed both speakers on the other side of the ledger — what hasn't worked, or hasn't worked yet — and Spencer's answer was worth noting.
Reporting is the area he flagged most directly. The pattern of using Claude or ChatGPT to generate one-off reporting artifacts — weekly reviews, pipeline summaries, analytical dashboards — creates a consistency problem. Each artifact comes out slightly differently. The metrics shift. The people receiving those reports lose confidence in what they're reading and stop acting on it.
"Reporting is an area where you still want to have a really clear framework of what KPIs matter, how they're calculated and trending historical data against those KPIs. And it's a consistent report that people are familiar with that drives actual action." — Spencer Hardey
The practical application he's found: use AI as a sounding board for identifying what should be in a report and testing whether the data is there to support it — then build the actual report as a static, consistent output in your existing business intelligence tool. AI accelerates the design process. It doesn't replace the discipline of structured reporting. This is a useful corrective to the tendency to reach for AI wherever a task feels tedious, rather than asking whether AI is actually the right fit for that task. The compounding costs of deferred governance apply here — inconsistent reporting erodes executive confidence in the data over time.
Spencer also noted the cost dimension of AI workflows as an area where HG Insights had to course-correct. Asking whether a workflow genuinely needs to run daily versus weekly — and what the incremental value of that frequency is relative to the token cost — is a discipline that doesn't come naturally when teams are excited about what automation can do.
Synthesizing the session's core recommendations, Hari and Spencer each offered a single piece of guidance for teams trying to move from dirty CRM data to AI-ready account intelligence.
Hari's framing was two questions that should precede any enrichment decision: Which accounts actually matter? And what are the data points that will make the difference for the specific play I'm trying to run? Domains and email addresses might matter less than technographic displacement signals. Buying intent might matter more than job title seniority. Getting clear on the play first makes both questions answerable.
Spencer's framing was a warning against the mass-enrichment trap. Don't let a business stakeholder or an executive push for broad coverage of all accounts in the CRM — coverage that will go stale before it's activated. Scope each enrichment initiative to a specific play, a specific set of accounts, and a specific timeline. Then prioritize which play comes next. The mental model is a sequence of defined, executable campaigns — not a database project.
Both observations point toward the same underlying principle: the value of CRM data is realized at activation, not at enrichment. A field that gets filled but never used is a cost, not an asset.
For teams working through the related challenge of building and maintaining the approval workflows, lead routing, and territory assignments that sit underneath this kind of account intelligence infrastructure, the tactical guide to smarter lead routing and the principles around building approval processes that actually work offer useful adjacent context.
The underlying current across every section of this session is the same: AI readiness is a data readiness problem before it is anything else. The teams building durable agentic workflows are not the teams that moved fastest to deploy agents — they're the teams that built the data foundation first, then designed agents to run on top of it.
Learn more about how HG Insights helps revenue operations teams access technographic intelligence, score and tier their addressable market, and build the account-level context that makes AI-driven automation worth deploying.
Check out our blog, join our community and subscribe to our YouTube Channel for more insights.
