Prospecting agent for mid-size B2B SaaS sales teams running Salesforce

What The Prospector does

The Prospector works for the sales team at a 51 to 200 person B2B software company that runs Salesforce as its system of record. It owns the top of the outbound funnel up to the first human conversation: keeping the ICP, personas and buying-signal definitions current, scanning for accounts that match, researching and enriching the right contacts, and drafting outreach built on real reasons to reach out. It writes everything back to Salesforce as account research notes, contact records and draft messages assigned to the owning rep. It does not send unapproved messages, negotiate, or handle replies. Once a prospect responds or books, the rep takes over. Hacking Demand builds it, tunes it and keeps it running, so the sales team gets the output without maintaining the system.

Custom build

This team runs Salesforce, so the right build is a custom Claude agent connected to Salesforce through its API, not HubSpot Breeze. The Breeze Prospecting Agent and its plays and selling profiles only work inside HubSpot, and adding a second CRM just to run one agent would split data and create sync problems. We keep the same proven structure (written plays, selling profiles, and AI context drawn from research on the company) but express it as guidelines for Claude, which reads Salesforce directly, uses a browser for account research, and runs on a schedule. That keeps Salesforce as the single source of truth, avoids middleware, and lets us adapt the research depth to this company's market rather than a fixed template.

What starts it

• A scheduled daily sweep each morning that scans target accounts for new buying signals and refreshes each rep's prospect queue
• A known target account shows high-intent website activity identified by Warmly (pricing, integrations or comparison pages, repeat visits)
• A new account is added to a rep's territory or a target account list in Salesforce
• A closed-lost opportunity or stalled account passes its re-engagement date set in Salesforce
• A champion from a current or past customer changes jobs to a company that fits the ICP

What it reads

• Salesforce (accounts, contacts, opportunities, closed-won and closed-lost history, activity) read through its API
• Warmly website visitor and intent signals
• FullEnrich and Apollo for waterfall contact enrichment and firmographics
• Gong call recordings and transcripts from won deals, used to extract the pains, language and objections that actually close
• Public web research by Claude with a browser (company news, hiring pages, funding announcements, product launches, leadership posts)
• The company's own positioning, case study library and messaging documents

When a person takes over

• Any reply, positive or negative, goes straight to the owning rep. The agent never responds to a prospect.
• Accounts with an open opportunity, an active customer relationship, or an owner conflict are routed to the sales manager to decide who engages, instead of the agent picking.
• When the research is thin or contradictory (unclear fit, merged companies, uncertain role), the account is flagged for rep judgment rather than drafted with assumptions.
• Changes to the ICP, personas or approved claims require sign-off from the sales leader before the agent uses them.

Rules it follows

• No message is sent without a rep approving it. The agent drafts, people send.
• Every personalization point must cite a verifiable source in the research brief. No invented metrics, customer names, quotes or results.
• Respect do-not-contact flags, opt-outs, existing customers and regional privacy rules stored in Salesforce, and skip any contact that fails email verification.
• Read access to Salesforce by default, with write access limited to research notes, contact enrichment fields and draft tasks. No changes to opportunities, stages or ownership.
• Messaging stays within the approved positioning document and tone guide, and the agent caps daily drafts per rep so queues stay workable.

How you know it works

• Meetings booked from agent-sourced accounts per rep per week
• Pipeline created from agent-sourced accounts
• Reply rate and positive reply rate by signal type and persona
• Share of agent drafts sent with light or no edits
• Rep hours spent on research per week, before and after rollout

From pilot to live

1. Pilot (two to three reps, one segment)Hacking Demand writes the ICP, personas and signal definitions, connects Claude to Salesforce, Warmly and the enrichment tools, and runs the daily sweep for a single segment. Reps review every brief and draft, and we tune scoring and messaging weekly based on their edits.
2. Expand (full team, more signals)Roll out to all reps and territories, add job-change and closed-lost re-engagement triggers, and turn on weekly outcome reviews. Owner-conflict routing and Chili Piper booking links go live.
3. Run and optimize (ongoing)Hacking Demand operates the agent month to month: maintaining connectors, refreshing personas as the product and market shift, retiring signals that do not convert, and reporting meetings and pipeline back to the sales leader.