Building Reliable Tools for GTM Agent Teams

Neha worked on the tool layer behind ZoomInfo's GTM agent teams, where agents use systems such as Salesforce, HubSpot, Dynamics, and Microsoft Teams to execute lead-to-revenue workflows. The product needed clear distinctions between success, recoverable failure, and terminal failure.

She first addressed false success in CRM record creation by tightening tool prompts and defining explicit guardrails. When users omitted the CRM object, she added a UI control that asked them to specify the intended write target and prevented unsupported assumptions.

Further research showed that users would still skip inputs. Neha moved the design from blocking ambiguity to supplying context, passing the workflow trigger type into the tool so an account trigger, contact trigger, or company record could guide the next action.

She also developed fallback logic around how sales teams work inside a CRM. For ownership matching, the agent could inspect recent open opportunities, activity, dates, account ownership, and contact or lead records, while still honoring an explicit user instruction when one existed.

The same reliability work extended to structured outcomes and retry policies across CRM lookups, team messaging, search, email discovery, and GTM content reads. Recoverable errors could retry or pivot, while unresolved failures remained visible and actionable for the user.

Neha grounded these decisions in conversations with ZoomInfo's sales organization and beta admins who managed CRM environments. The product reached general availability recently; measurable post-launch reliability and adoption outcomes were not yet supplied.