Skip to content
GTM Agent Weekly

Rewrite Salesloft cadence steps from real call history with Claude

An AI assistant reads what was actually said on past calls with an account and helps rewrite the follow-up steps to match. Managers can also ask which committed deals have gone quiet.

Paste this into ChatGPT, Claude or your agent tool. See the prompt

  • Prospecting / Outbound
  • intermediate
  • SDR manager
  • Sales leader
  • RevOps
Category
Prospecting / Outbound
Difficulty
Intermediate
Est. cost
Not stated by source

Works with

Problem

Cadence steps get written from a persona template while the actual call history for the account sits unread. Managers also want a Monday view of committed deals with no activity.

Setup

Claude connects to the official Salesloft MCP server, which exposes cadence and activity data, call intelligence and forecast data with gated write-back. Reps ask questions that join datasets, such as stalled cadences where budget came up on a call, and Claude drafts the next cadence step from that call. Recurring jobs like a Monday pipeline note move to Claude Code or the Agent SDK on the Salesloft v2 API.

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.

Starter prompt · 22 lines
# ROLE
You help reps rewrite the next cadence step using what was actually said on calls.

# INPUTS
- {{CADENCES}}: people in cadences with their current step and activity history.
- {{CALLS}}: call recordings or summaries for the same people or accounts.
- {{QUESTION}}: the rep's request, e.g. "stalled cadences where budget came up on a call".

# TASK
1. Find the people or accounts that match {{QUESTION}} by joining cadence and call data.
2. For each, quote the relevant call moment with date.
3. Draft the next cadence step (email or call script) that picks up from that moment:
   acknowledge it, add one useful point, ask one question.
4. Show the draft next to the current generic step.

# OUTPUT
Per person: current step | call moment (quote, date) | rewritten step.

# GUARDRAILS
- Draft only. Changes to cadences are written back only after the rep approves.
- Don't quote the buyer's words back to them verbatim if it would feel like surveillance; paraphrase.
- No pricing or commitments that were not on the call.

Replace every {{PLACEHOLDER}} before running. Works with any capable LLM and runner.

Tools

Steps

  1. Confirm with your Salesloft admin that the org has the add-on that includes MCP access.
  2. Connect Salesloft from the Claude connector directory with your Salesloft login.
  3. Start with a read-only question that joins two datasets and check the answer in Salesloft.
  4. Draft one cadence step from a real call transcript and compare it to what a rep would have sent.
  5. Move recurring digests to the API and keep every write-back behind review.

Results

Practitioner guide; no outcome data.

What goes wrong

Gotchas

  • MCP access is gated by entitlement; a connector that returns nothing useful is usually missing the add-on.
  • Transcripts mangle names and product terms, so give Claude a glossary.
  • Adding a batch of people to a cadence from a bad filter is tedious to undo, so gate writes.

Source and credit

www.getwickedgrowth.com ›

Vikash Koushik, Get Wicked Growth · newsletter

Source dated 2026-09-22 · last reviewed 2026-10-11. Summarized in our own words; see the source for details.

Related.

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.

Works with ClaySalesforceHubSpot +3

  • Prospecting / Outbound
  • intermediate