The Core Challenge: Reducing Coordination Costs

Go-to-market (GTM) teams often suffer from a bottleneck where great ideas (like targeted outbound campaigns) take months to execute due to fragmented data, operational busy work, and the difficulty of aligning cross-functional teams. The goal of GTM orchestration is to shift from manual execution to describing intent, allowing agents to handle audience segmentation, content generation, and multi-channel distribution.

Building a Unified Data Foundation

Effective orchestration requires a consistent data substrate. Ramp built an internal customer data platform (CDP) that integrates:

  • Transactional Data: CRM, product usage, and buying signals stored in Postgres to maintain referential integrity.
  • Unstructured Data: Call transcripts, emails, and notes are embedded in a vector database to allow agents to search for context without loading entire raw corpora into their prompt windows.
  • Real-time Events: Kafka topics handle live signals, which are then piped into the CDP to ensure agents act on the most current information.

Solving Vertically, Scaling Horizontally

Rather than building a monolithic architecture, the strategy is to solve a specific team's problem first, then generalize the patterns.

  • Example (Pre-meeting Briefs): Account managers need to synthesize usage data, open tickets, and agendas before calls. By building a durable agent that maps attendee emails to accounts (a non-trivial task due to email-to-business overlaps), the team created a repeatable, high-value workflow.
  • Durable Execution: Using Temporal ensures that agent workflows are durable; if a worker fails, the process resumes from the last activity rather than restarting, which is critical for complex, multi-step GTM tasks.
  • Skill Libraries: Allowing users to define their own brief formats and instructions within the agent framework drives adoption. By exposing these tools via an internal Model Context Protocol (MCP), employees can "vibe code" their own automations, which the engineering team then identifies and productionizes for the wider organization.

Advice for Smaller Teams

Do not spend a year designing a perfect architecture. Start by identifying narrow, high-impact automations—such as personalized outbound sequences—and build the infrastructure iteratively. Solving real, immediate problems provides the necessary understanding of how to extend those patterns to other teams and channels later.