Run an always-on research agent across a segment of accounts
This agent keeps researching a whole group of accounts, such as lost deals, and records things like the real reason a deal was lost. Reps stop redoing the same research every time they revisit an account.
Paste this into ChatGPT, Claude or your agent tool. See the prompt
- Prospecting / Outbound
- intermediate
- RevOps
- AE
- Sales leader
- Category
- Prospecting / Outbound
- Difficulty
- Intermediate
- Est. cost
- Variable Clay credits plus one action per record (sourced from vendor docs)
Problem
Account context such as the real closed-lost reason or an engagement summary is scattered across CRM records, call transcripts and email, so reps re-research the same accounts.
Setup
In Clay, connect CRM, warehouse and Gong data to Audiences, define a segment (for example a closed-lost cohort), and build an Account Research Agent by describing the job to Clay's copilot. The agent runs across every account in the segment, maintains agent-managed fields such as closed-lost reason or engagement summary with its reasoning, and can write back to the CRM with human approval. Runs follow the segment and keep memory per account.
Prompt
Starter prompt by The GTM Agent Weekly, written from the setup described on this page. It is not the source's own prompt; see the source link below for their write-up.
# ROLE
You are an account research agent that runs across a segment of accounts (for example, every
closed-lost account from the last year) and keeps a few fields current on each one.
# INPUTS PER ACCOUNT
- {{CRM_RECORD}}: account, opportunities (with close reason and date), contacts.
- {{CALLS}}: call summaries or transcripts for the account (e.g. Gong).
- {{WAREHOUSE}}: product usage or other data, if any.
- {{WEB_SEARCH}}: public news and job posts.
- {{PREVIOUS_FINDINGS}}: what you recorded on this account last run.
# FIELDS TO MAINTAIN
1. True loss reason: what the calls and notes show, not just the CRM picklist. Quote evidence.
2. What changed since the loss: new leaders, funding, hiring, product launches, competitor
contract timing. Each with date and link.
3. Re-engage now? yes, not yet or no, with a one-line reason.
4. Suggested angle for the first message back.
# TASK
Compare against {{PREVIOUS_FINDINGS}}. Update a field only if something new supports it,
and keep your reasoning next to the value.
# GUARDRAILS
- Propose CRM write-backs; a person approves them. Never overwrite rep-entered fields.
- No evidence, no claim. "No change" is a valid result. Replace every {{PLACEHOLDER}} before running. Works with any capable LLM and runner.
Tools
Steps
- Connect at least one data source (CRM, warehouse or Gong) in Audiences.
- Create a segment for the accounts the agent should cover, such as closed-lost deals.
- Describe the agent's job to the builder and define the output fields it should maintain.
- Test on a few sample records and review the cost estimate before running on the whole segment.
- Inspect per-account outputs and reasoning, then sync approved fields to the CRM or warehouse.
Results
The documentation describes setup and outputs but reports no outcome data.
What goes wrong
Gotchas
- Credit costs scale with segment size; the docs recommend testing on samples and checking the cost estimate first.
- Outputs are only as good as the connected data; missing call or CRM coverage leaves fields thin. (editor note)
- The feature was in open beta at the time of the docs, so behavior and pricing may change. (editor note)
Get 5 like this every week, free.
Real agent setups with sources and failure modes, in one short weekly email.
Free. One email a week. Unsubscribe anytime.
Source and credit
Clay University docs · vendor
Source dated 2026-10-11 · last reviewed 2026-10-11. Summarized in our own words; see the source for details.