Recommendation (short):
Build a Zap that triggers on each new form submission, normalizes the fields, sends a structured prompt to ChatGPT to return a JSON qualification object (score + tags + short reason), then use a Zapier Filter to push only qualified leads into your CRM. Log failures and low-confidence results to a Google Sheet and notify the team for manual review.

Trigger setup (Zapier)
1. Trigger: New Form Submission (Typeform/Gravity Forms/Google Forms/HubSpot form). Include a unique lead ID/email in the payload.
2. Action: Formatter — clean phone/email, normalize country/timezone fields, concatenate multi-field answers into one “lead_text” field.
3. Action: ChatGPT (Zapier OpenAI integration) — send a system + user prompt that returns machine-readable JSON (see templates below).
4. Action: Filter — pass only when qualification.score >= 70 (adjustable).
5. Action: CRM create/update (Salesforce/HubSpot/Pipedrive). Map fields including score, tags, and short reason.
6. Paths: low-score path -> append row to Google Sheet + Slack/email notify. Failure path -> create task in helpdesk or Google Sheet.

Prompt templates (use verbatim and require JSON output)
System message:
"You are an expert B2B lead scorer. Return ONLY valid JSON with keys: score (0-100), qualified (yes/no), top_tags (array of short strings), reason (1-2 sentences), confidence (low/medium/high). No extra text."
User message (example):
"Lead data: {lead_text}
Our ICP: company size 50-250, industry: SaaS, ARR > $250k, decision-maker role only. Produce JSON output. Score higher for company fit, budget signals, intent language, and decision-making authority."
Example expected output (must be returned exactly as JSON):
{"score":82,"qualified":"yes","top_tags":["mid-market","decision-maker","product-fit"],"reason":"Matches industry and role; mentions current spend and implementation timeline.","confidence":"high"}

Decision criteria (what to choose and why)
- Model: gpt-4 for higher accuracy/context parsing (recommended if budget allows); gpt-3.5 for lower cost and higher volume.
- Threshold: start at 70, monitor conversion rates for 2–4 weeks and adjust up/down.
- Output format: enforce JSON to avoid parsing errors.

Retry / error handling recommendations
- Use Zapier’s built-in retry for transient failures (enable Zap task retry). Set webhook timeout higher if using OpenAI API.
- On parse errors or non-JSON responses: route to an error path that records the raw input+response to Google Sheets or S3 and notifies Slack for manual triage.
- Implement idempotency: use the lead ID/email as a dedupe key in CRM create step to avoid duplicates.
- Rate limiting: monitor token usage; back off to gpt-3.5 if limit/budget hit.

Practical checklist (ready to follow)
- [ ] Configure trigger and include lead ID
- [ ] Add Formatter step to normalize inputs
- [ ] Add ChatGPT step with system + user messages, require JSON
- [ ] Add Zapier Filter for score threshold
- [ ] Map qualified leads to CRM with score & tags
- [ ] Add Google Sheet + Slack error/low-score path
- [ ] Monitor results for 2–4 weeks and tune threshold/model

Best-for and avoid-if
- Best for: small-to-mid sales teams wanting fast lead triage without dev resources.
- Avoid if: you must guarantee 100% accuracy (use manual gating), or you lack budget for GPT-4 at scale.

Notes on budget/skill/team: If you have tight budget and high volume, start with gpt-3.5; if accuracy and nuance matter and team size is >3 SDRs, invest in gpt-4. This workflow is execution-stage ready for no-code builders using Zapier and ChatGPT.

Compare Zapier and Make