GTM.ai
Sign up
Docs menu

How to ask for ZoomInfo data in an agent team step in GTM One

Learn how to word a step's instructions so it gets the ZoomInfo data you want.

Srivatsa MarthiSrivatsa MarthiPrincipal Product Manager

Learn how to word a step's instructions so it gets the ZoomInfo data you want.

Overview

An AI agent step in an agent team can look up ZoomInfo GTM data in the middle of a run: companies and people that match criteria, a deep dive on one named company, account, opportunity or person, your own open pipeline and book of business, your meetings and emails, and what was said on your recorded calls. The step asks for the data in plain words, and what comes back is passed to the step's own work and to later steps.

The lookup only reads. It never creates or changes a record, and it cannot create an opportunity or log a call note. Use a step for the CRM, such as a Salesforce step, to write.

Which data tool a step has depends on how its tools are set up, so you may not see a ZoomInfo data tool on every team. If the step does not look anything up, contact your ZoomInfo admin.

Before you start

  • Permission to edit the team. The steps below change a step's instructions. If the builder's controls are turned off, you can view the team but not change it. Ask your ZoomInfo admin for edit access.
  • Your ZoomInfo sign-in. The lookup runs as the person the team runs as and uses that person's ZoomInfo access. A triggered or scheduled run uses the team owner's access, not the access of whoever opens the run.
  • A connected CRM for your own pipeline. Opportunities, accounts and CRM contacts come from the CRM your organization connected. Public company and person data does not need it.

Write instructions that ask for ZoomInfo data

  1. In the left navigation panel, click Agent Teams, open the agent team you want to change, and click the Steps tab.
  2. Click the step that needs the data, or click Add action to add a new one.
  3. In Instructions, say what data the step should look up, in one self-contained request. Name the company, domain, ID, person name or email, owner and date range the request depends on, because the lookup does not see the rest of the instructions. For example: "Find fintech companies in California with 500+ employees", "Show the open pipeline owned by Alex Johnson closing this quarter" or "Get the contact profile for Derek Osgood at ZoomInfo".
  4. For a list, say how many records the step needs. A list request can return up to 100 records. Say "all" only when the step truly needs every match: the lookup then returns up to 500 rows in one go.
  5. For a count, rate or "how many" question, ask for the count. Counts are exact over everything that matches, while a list is capped, so a count worked out from a list is not reliable.
  6. When the request is about a colleague's records, give their name or email. The lookup finds the person in your CRM first and then scopes the request to them.
  7. In Input source, select the earlier step whose output this step should use, when the request depends on it.
  8. Click Save changes, then run the team. See How to run an agent team manually in GTM One.

What to expect in the result

  • A different shape than you asked for. The result may be summarized or counted instead of a raw list, for example when the data is large. This is intended, and the step should use it rather than ask again.
  • Coverage notes. When a filter was applied after the fact, a value matched loosely or the result is partial, the result says so. The step should pass that caveat on. Check the notes before you rely on the answer.
  • Order. Opportunity lists come back by close date, earliest first, and account lists by account name. A page is not a ranking, so ask for a count by group when you want "the biggest" or "the worst".
  • No matches. A request that finds nothing is a valid answer ("none"), not a failure.
  • Sources. Facts from your own calls and CRM are labeled by where they came from, such as a CRM account or a recorded meeting.

If you see an error

Resources