Better targeting, faster follow-up: Inside Cockroach Labs’ AI-powered GTM playbook

What happens when a GTM Ops team decides to build its own AI-powered prospecting system instead of waiting for a vendor to deliver one?

Cockroach Labs built MICA to bring the account research and outreach tools reps need into one place. It helps them prioritize accounts, find the right contacts, and tailor their messaging. As the team puts the system into practice, it's learning which account signals are actually useful, how to get reps to act on AI recommendations, and how to balance the cost of AI with the value it brings.

Those questions were at the center of our recent AI IRL: Inside Cockroach Labs' AI-powered GTM stack webinar with RevOps Co-op. Joining us were:

• Jackson Mattox, Manager of Marketing Operations at Cockroach Labs
• Dominic Freschi Jr., Director of GTM Operations at Openprise
• Matthew Volm, founder of RevOps Co-op, moderating

The session explored the practical side of AI for GTM, from composite account scores and inbound sequencing to scoring fatigue, AI costs, and data orchestration.

Here are five lessons from Cockroach Labs' experience.

1. Build AI around how sellers actually work

When reps know which accounts to prioritize and have the context to engage them, they can spend more time on conversations that create pipeline. That's the goal behind MICA, Cockroach Labs' in-house account prioritization and outbound system.

Jackson described MICA as a shared instance of Claude Cowork with custom Python scripts on top, connected to roughly two dozen apps across the business.

Reps see their accounts ranked by a composite score that combines marketing engagement, product activity, external signals, and ideal customer profile (ICP) fit. From there, they can:

• See why an account ranks where it does
• Identify engaged contacts and spot gaps in buyer committee coverage
• Generate cold emails and call scripts using messaging plays from Product Marketing
• Add contacts to outbound sequences or map contacts for an SDR to follow up with

MICA went live about two weeks before the session, following eight to ten months of iteration on earlier outbound workflows. The app now has a product owner to guide its ongoing development. As Jackson joked, the team is “creating work for ourselves.”

Cockroach Labs also automates time-sensitive inbound follow-up. When a lead reaches the marketing-qualified threshold, a workflow adds it to an SDR sequence in under 40 seconds. If no SDR is assigned, the workflow uses regional round-robin routing.

Together, these workflows show two ways to bring AI into the seller's day: MICA helps reps prioritize and personalize outbound, while automation keeps qualified inbound leads moving without waiting for manual intervention.

2. Make every score earn its place

AI makes it easier to bring more signals into account prioritization. The challenge is knowing which ones actually help reps make better decisions.

Jackson called out a familiar risk: score fatigue.

“It's never been easier to use AI to create scores.”

– Jackson Mattox, Cockroach Labs

Cockroach Labs combines signals from several sources, including external signals, new job postings at named accounts, and custom Claude queries that check for events such as outages. Funding rounds and product launches also contribute to the picture.

The team is still learning which signals deserve the most attention. Reps can inspect the score breakdown, decide what's relevant, and provide feedback when a signal isn't useful. The next step is to distinguish signals by urgency, so a high-priority event could trigger a Slack alert rather than waiting for a rep to log in.

The team is also working to make outbound messaging more measurable. Product Marketing and sales leadership have defined relevant events and how reps should respond. Jackson plans to tag outbound sequences by play and compare results, while exploring how AI could help classify prospect replies and identify which messaging approaches work best. He's also exploring a product-fit score to help identify which Cockroach Labs solution may be best suited to an account.

A score is only useful if it changes what a rep does next. Making the underlying signals visible and connecting them to specific plays gives the team a way to test whether its prioritization is working, then refine it based on results.

3. Give AI a trusted source of data and clear boundaries

Account prioritization and personalized outreach depend on reliable information about accounts, contacts, engagement, and product activity. That makes the data foundation behind an AI application just as important as the model itself.

When asked what advice he'd give other teams building similar systems, Jackson pointed to a simple starting point:

“If there's one piece of advice I would say for everyone, it's figure out what system is gonna be your repository for all of your data. What is your source of truth?”

– Jackson Mattox, Cockroach Labs

For Cockroach Labs, Snowflake serves as a central data repository, while Salesforce remains the system of record. External, product, and engagement signals live in Snowflake tables that Claude can query, giving MICA broader account context. The data team controls which product signals are made available to the application.

Governance also shapes what happens after AI generates an answer. During the discussion, Dominic explained how a guided interface can pair AI-generated recommendations with defined workflows that control what happens when a user takes action:

“It's this merger of probabilistic, so it's the presentation and how it's generating, with the deterministic of what are the actual flows that are gonna happen when they click that button.”

– Dominic Freschi Jr., Openprise

AI can generate recommendations and personalized content, but governed workflows determine how those recommendations translate into actions in business systems.

The architecture matters on both sides of the model: Snowflake gives MICA access to curated account context, and defined workflows help ensure that acting on an AI recommendation follows the intended process.

4. Design AI workflows with cost and ROI in mind

AI can help GTM teams research accounts and personalize engagement at scale, but teams need to understand what those workflows cost and whether they deliver results.

Cockroach Labs is weighing those tradeoffs as it develops MICA. Running Claude research on named accounts more frequently could surface fresher signals, for example, but also increase costs. Until the team understands how much those signals influence pipeline, it needs to be deliberate about how often it runs queries.

The team is also comparing how it accesses AI models. Cockroach Labs routes requests through Agent Hub, an internal gateway for multiple AI services. Jackson shared one comparison in which the same query was roughly 50% cheaper through Agent Hub than by calling the model API through Clay. While that result applies to this specific comparison, it shows how architecture can affect AI costs.

“You'll be surprised if you actually run some analysis on two different methods of calling a particular LLM, what the price difference can actually look like.”

– Jackson Mattox, Cockroach Labs

Dominic highlighted three ways GTM teams can make AI spend go further:

• Match the model to the task. Reserve more powerful models for work that requires deeper reasoning.
• Prepare the data first. Give models consistent, relevant inputs to reduce unnecessary processing.
• Build efficient workflows. Define how AI interacts with data and business systems so each request serves a clear purpose.

Cockroach Labs can already track overall AI usage through IT and plans to combine usage data with outreach performance and seller feedback. The goal is to understand whether account research improves prioritization, personalized plays drive engagement, and faster follow-up improves conversion.

Connecting usage to outcomes will help the team decide where further AI investment is worth it.

For more on managing usage and making AI spend accountable, read our AI last mile webinar recap.

5. Match the orchestration layer to the job

Different GTM workflows have different requirements. Some need to respond immediately to an event; others need to process thousands of records consistently. Choosing the right orchestration approach helps teams balance speed, scale, and cost.

Cockroach Labs uses a mix of CRM, data warehouse, enrichment, workflow automation, and orchestration tools. Jackson acknowledged that the team may have more orchestration tools than it needs, but the right choice ultimately depends on the job.

As covered in lesson 1, event-driven workflows can move qualified inbound leads into the right sales sequence quickly. Cockroach Labs uses Openprise for bulk data processing, deduplication, and lead-to-account matching.

The approaches can also work together. Cockroach Labs' contact sales flow needs to process a lead immediately, so it performs an initial lead-to-account match. Occasionally, that can create a duplicate account, for example when a subdomain isn't associated with the existing record. Openprise can then identify and resolve the duplicate through a follow-on process.

Jackson described Openprise's role this way:

“It is like that supportive layer so that every other AI function can execute the way that we want it to.”

– Jackson Mattox, Cockroach Labs

Dominic also noted that Openprise's MCP connectivity was in preview at the time of the session. MCP, or Model Context Protocol, provides a standard way for AI applications to connect with other software.

For simpler, lower-volume needs, native CRM or marketing automation workflows may be sufficient. As GTM teams add more data sources, AI applications, and real-time processes, a dedicated data orchestration layer can help maintain consistent information and business rules across workflows.

The key is giving each layer a defined role: respond quickly when timing matters, process data consistently at scale, and keep information flowing reliably between systems.

What an AI-powered GTM stack looks like in production

Cockroach Labs is building AI into the way its GTM team prioritizes accounts, researches prospects, and follows up with qualified leads. The team is still testing which signals matter most, which outbound plays perform, and where additional AI investment will pay off.

Its experience offers a practical starting point for other GTM teams: choose a workflow where better prioritization or faster follow-up could influence pipeline, connect AI to trusted data, and give people a clear path to act on its recommendations. Then measure the results before expanding.

Openprise helps GTM teams prepare, match, and manage the data behind these workflows, while orchestrating consistent processes across their GTM stack. That foundation helps teams put AI to work with greater confidence and turn more insights into meaningful action.

Trusted data. Defined workflows. AI that drives GTM execution.

Watch the full AI IRL: Inside Cockroach Labs' AI-powered GTM stack replay to see Jackson's walkthrough of MICA and hear the discussion about account prioritization, governance, AI costs, and orchestration.

Want to make your AI workflows more reliable and actionable? Read our enterprise AI ROI guide for Ops teams to learn how Openprise helps GTM teams connect AI with the data and processes that drive execution.