Recipes / Expansion signals on existing customers

Expansion signals on existing customers

HubSpot + PredictLeads

Who it's for

CSMs and account managers who want to spot growing customers before the renewal call, from hiring, funding and technology changes. HubSpot can be swapped for Salesforce, Attio or Pipedrive, and PredictLeads for ZoomInfo or Crunchbase.

What you get

Customers ranked by expansion signal, with a next action each (read only). Sample output: every company and person in it is fictional.

30 customers checked. 5 show expansion signals from the last 60 days. Nothing was written to HubSpot.

1. Quarry Lane Analytics (fictional), 40 seats, renews Feb 12
   - 9 open roles in sales and RevOps, up from 2 last quarter (jobs, Oct 3)
   Next action: ask Sam Whitlock in your next check-in which new teams will need seats.

2. Larkspur Freightworks (fictional), 15 seats, renews Jan 30
   - Series C financing (Sep 24)
   Next action: flag to Alex Rivera (AE) for an expansion conversation before renewal.

3. Fernhill Software (fictional), 25 seats, renews Mar 5
   - Started using a competing tool in the same category (technologies, first seen Sep 18)
   Next action: this is a risk, not an expansion. Ask Priya Raman how the teams are splitting work.

...2 more

Similar companies to Quarry Lane not in HubSpot: Tidewell Metrics (fictional), Oxbow Ledger (fictional).
PredictLeads requests used: 90, plus 5 similar-company results.

Time and cost

Estimate About 15–20 minutes for 30 customers, most of it reading the ranked list.

Assumptions: One run over 30 customer accounts, with a similar-companies lookup for the top account only.

• HubSpot: Reads only: customer companies, owners, open deals and any renewal date property you use. Credit use not confirmed by HubSpot.
• PredictLeads: About one request per signal type per customer (jobs, financing, technologies), so roughly 90 for 30 accounts. Similar Companies is billed per returned company. Every call uses credits. Each MCP request counts toward the plan's monthly request limit. Once the limit is reached, requests return HTTP 402. PredictLeads pricing: 1 API credit per API request; discovery endpoints (e.g. Similar Companies) are billed per returned company. Source: docs.predictleads.com, checked 2026-10-16

Exact credit costs depend on your plan with each vendor. Check the tool pages before a large run.

MCP servers

Every server is confirmed in: ClaudeChatGPT

• HubSpot Up · auth
  ClaudeChatGPTRead + writeCredits not confirmed
  https://mcp.hubspot.com
• PredictLeads Not probed
  ClaudeChatGPTClaude CodeAccess not confirmedEvery call uses credits

Set up in a chat app

Every server in this recipe is confirmed for Claude and ChatGPT. Connect each one, then paste the prompts below into one conversation.

• Connect HubSpot without code (Claude, ChatGPT)
• Connect PredictLeads without code (Claude, ChatGPT)

Developer setup

{
  "mcpServers": {
    "hubspot": {
      "url": "https://mcp.hubspot.com"
    },
    "predictleads": {
      "url": "None"
    }
  }
}

Step-by-step prompts

Copy all prompts

1. Find my current customers in HubSpot: companies with lifecycle stage Customer {owned by me / in segment}. Show a table with company, domain, owner, open deals and renewal date if we track one. Don't change anything.
2. Tell me how many PredictLeads requests it will take to check these customers for job openings, financing and new technologies over the last {60} days, and run it only after I approve. Show each signal with its date and source.
3. For the top account, check PredictLeads for similar companies only after I approve it, and list up to 5 that aren't already in HubSpot.
4. Rank the customers by expansion signal and suggest one next action for each, such as a question for the champion or a flag to the AE. Label any churn-risk signals as risks. Keep it as a list for me, and don't create tasks or notes in HubSpot.

Run it on a schedule (optional)

• Run it monthly, or a quarter before each batch of renewals.
• A scheduled run should end with the ranked list sent to you. It makes no changes anywhere.
• Cap the number of customers in the prompt. Every PredictLeads request counts toward the plan's monthly limit, and once it's reached requests return HTTP 402.

Where this goes wrong

• Replacement hiring read as growth. Open roles can mean people left, not that the team is growing. Ask for the count compared with the previous period, and only call it growth if roles in that team went up.
• A risk scored as an expansion. A newly detected technology can be a competitor to your product. List new technologies by category, and flag any that overlap with what you sell as a risk.
• Wrong domain for the customer. Customers with a parent company or a rebrand may be on a different domain than the one in HubSpot. Have it show the domain it checked, and spot-check the top 5 by hand.
• Pushing expansion on an unhappy customer. Signals say nothing about how the account feels about you. Before acting, check HubSpot for open tickets on the top accounts and skip any with an open escalation.

Gotchas

• PredictLeads doesn't publish its MCP endpoint URL. Connect it from the Claude connectors directory or the ChatGPT plugins directory. On ChatGPT Enterprise or Edu, an admin may need to enable it first.
• We couldn't confirm whether PredictLeads' MCP is read-only, and this recipe never asks it to write.
• Similar Companies is billed per returned company, so ask for a small number.
• HubSpot non-admin users need a Super Admin to approve the connector first. The HubSpot connector can also edit records, so keep the read-only guardrail in the prompt.

Download

Get the prompts, guardrails and expected output as one Markdown file. Paste it into a Claude Project or ChatGPT project's instructions, or save it as CLAUDE.md for Claude Code.

Download expansion-signals.md

Vendor sites: HubSpot. If we join an affiliate program, these links will be labeled as affiliate links.