How to automate sales follow-ups with Zapier + ChatGPT

Short answer / recommendation
Use a single Zap: HubSpot (New Form Submission or Contact property change) → Formatter/Filter → OpenAI/ChatGPT action to draft the email → Gmail send → HubSpot Create Engagement (log the email). Add a Delay/Storage step and retry paths for rate‑limit handling. Prefer gpt-3.5 for high volume and gpt-4 only for high‑value prospects.

Decision criteria
- Volume vs cost: high volume (hundreds/day) → gpt-3.5 (cheaper, faster). Low volume or high‑value demos → gpt-4 for better tone.
- Team skill: no devs → keep everything in Zapier using its OpenAI action. Devs available → consider sending HubSpot webhook to your backend to manage queuing, retries, and richer logging.
- Workflow stage: if you need immediate send, do inline generation. If you can tolerate small delays, queue generation and sending to smooth rate limits.

Field mappings (practical example)
- Trigger: HubSpot New Form Submission (or New Contact with demo_requested=true)
- Map these tokens into the prompt: {{firstname}}, {{lastname}}, {{email}}, {{company}}, {{jobtitle}}, {{hubspot_owner_email}}, {{form_name}}, {{source}}.
- Subject template: "Quick follow-up re: {{company}} — demo next steps"
- Prompt template (short): "Write a 3-paragraph personalized sales follow-up email to {{firstname}} {{lastname}} at {{company}}. Mention they requested a demo via {{form_name}} and reference {{source}}. Include a clear CTA to book a 30-min demo with {{hubspot_owner_email}}. Tone: friendly, concise, 120–160 words. Use their job title: {{jobtitle}}."

Zap step-by-step and logging
1) Trigger: HubSpot form submission (include form ID).
2) Filter: only process demo forms or contacts with demo_requested=true.
3) Formatter (optional): normalize names, remove extra whitespace.
4) Delay/Rate limiter: Delay For or Delay After Queue (e.g., 1–3s) to avoid bursts.
5) OpenAI/ChatGPT action: pass the prompt and mapped tokens.
6) Gmail action: To={{email}}, Subject={{subject template}}, Body={{OpenAI_output}}. Send as the right HubSpot owner.
7) HubSpot Create Engagement: set engagement type = EMAIL, body = generated email, associate contact and owner, set outcome/sent property.
8) Failure path: on error, add row to Google Sheet + notify Slack and mark contact with a custom property like followup_failed=true.

Rate limits & patterns (practical)
- OpenAI: assume limited throughput. Practical rule: gpt-4 ~1 request/sec/account; gpt-3.5 ~3–5 reqs/sec. If uncertain, conservatively space to 1–2s per request.
- Gmail: respect Google Workspace sending limits (often daily caps). Batch or throttle outbound sends if you hit limits.
- HubSpot API: avoid large bursts — pace writes (updates/engagements) and use conditional checks to skip duplicates.

Error handling checklist (implement these in Zap paths)
- Detect 429/5xx: exponential backoff (2^n seconds), up to 5 retries.
- On permanent failure: write minimum info to a “failed followups” Google Sheet and notify Slack/owner.
- Dedupe: set and check a custom HubSpot property (followup_generated=true) before creating another followup.
- Audit: write generated content + model metadata to HubSpot engagement and an internal log (sheet or DB).

Best-for / Avoid-if
- Best for: small-to-medium teams with no backend who need quick personalization at scale.
- Avoid if: you have massive volume, strict SLAs, or need advanced prompt chaining—build a backend queue instead.

Quick practical checklist to implement now
- [ ] Create HubSpot trigger for the demo form or contact property.
- [ ] Build prompt with placeholders and test 10 variations.
- [ ] Add Delay step (1–3s) before OpenAI to smooth traffic.
- [ ] Map output to Gmail send and HubSpot Create Engagement.
- [ ] Add error path: retries, failed-row sheet, Slack notification, set followup_failed flag.
- [ ] Monitor Zap task history; tune delay and model choice based on cost/latency.

If you want, I can draft the exact Zapier action field mappings and a ready-to-paste prompt for ChatGPT.

Compare Zapier and Make