CRM data decay audit: bounce risk and job changes in a small contact sample

LeadMagic + HubSpot

IntermediateRevOps

Who it's for

RevOps teams who want to know how stale their contact data is, segment by segment, before a big campaign or a data-vendor renewal, without changing any records. LeadMagic can be swapped for ZeroBounce, and HubSpot for Salesforce, Attio or Pipedrive.

What you get

Decay rates by segment from a 60-contact sample. Sample output: every company and person in it is fictional.

Sample: 60 HubSpot contacts, 20 per segment, picked at random from contacts not updated in 12 months.
LeadMagic credits: balance checked before and after. 15 credits for 60 email checks, plus the people lookups you approved.

| Segment | Checked | Valid | Invalid | Couldn't confirm | Changed company |
|---|---|---|---|---|---|
| Enterprise (1,000+) | 20 | 15 | 3 (15%) | 2 | 4 (20%) |
| Mid-market | 20 | 17 | 2 (10%) | 1 | 3 (15%) |
| SMB | 20 | 13 | 6 (30%) | 1 | 5 (25%) |
| Total | 60 | 45 | 11 (18%) | 4 | 12 (20%) |

Examples:
- Dana Okafor, [email protected], Larkspur Freightworks (fictional): email invalid. Now VP Operations at Brindle Cargo (fictional).
- Sam Whitlock, [email protected], Quarry Lane Analytics (fictional): email valid, same company, new title. Not a job change.
- Alex Rivera, [email protected], Mossgate Labs (fictional): email couldn't be confirmed. Two people with this name found. Marked 'not sure'.

Rough read: about 1 in 5 old contacts in this sample would bounce or has moved on, and SMB decays fastest. With 20 per segment, one contact moves a rate by 5 points, so treat these as rough rates.

No HubSpot records were changed.

Time and cost

Estimate About 20–30 minutes, most of it picking the sample and reading the results.

Assumptions: One run over a sample of 60 contacts, 20 in each of three segments.

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

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.

Developer setup

Auth setup and mcp.json for Claude Code and Cursor
  • LeadMagic: Sign in with OAuth in your client (run /mcp in Claude Code). Do it once interactively, and the client stores and refreshes the token.
  • HubSpot: Sign in with OAuth in your client (run /mcp in Claude Code). Do it once interactively, and the client stores and refreshes the token.

Put this in .mcp.json (Claude Code, project root) or ~/.cursor/mcp.json (Cursor). Swap the example keys for your own, and don't commit real ones. Remove the // comment lines if your client requires strict JSON.

{
  "mcpServers": {
    "leadmagic": {
      "url": "https://mcp.leadmagic.io/mcp"
    },
    "hubspot": {
      "url": "https://mcp.hubspot.com"
    }
  }
}

Step-by-step prompts

  1. Count my HubSpot contacts by {segment_field} and show me a table. Then pick a random sample of {n_per_segment} contacts per segment from contacts not updated in {months} months: name, email, company, company domain, job title and segment. Don't change any records.
  2. Check my LeadMagic credit balance. Tell me how many email checks and people lookups this sample needs, and wait for my approval.
  3. After I approve it, check each work email in the sample with LeadMagic's validate_work_email. Show me a table per segment with valid, invalid and couldn't-confirm counts and rates.
  4. Only after I approve the lookups, use LeadMagic people search to find each sampled person by name and old company domain. Flag anyone now at a different company, with the new company and title, and mark weak matches 'not sure'. Then check my credit balance again.
  5. Give me a short summary by segment: invalid rate, job-change rate, the combined decay rate, and three example records. Keep it here as text. Don't write anything to HubSpot or LeadMagic.

Run it on a schedule (optional)

Where this goes wrong

Gotchas

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 data-decay-audit.md

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