Short answer
You can build a no-code enrichment pipeline in Zapier that: triggers on new HubSpot leads, collects company & activity data (HubSpot or third‑party enrichers), calls ChatGPT to synthesize company details + intent signals and return a numeric intent score, then writes the enriched fields back to HubSpot.

Recommendation (one-liner)
Use Zapier to orchestrate: HubSpot trigger → lookups (HubSpot/third‑party) → ChatGPT prompt (strict JSON output + examples) → parse JSON → Update HubSpot contact. Keep prompts deterministic and include a short scoring rubric.

Why this works
Zapier handles triggers and field mapping without custom servers; ChatGPT (via Zapier’s OpenAI/ChatGPT action) can convert messy inputs into structured outputs and produce a reproducible intent score if you provide a consistent schema and examples.

Decision criteria (pick what matters most)
- Budget: ChatGPT/OpenAI calls + third‑party enrichers (Clearbit, etc.) cost per call. If budget is tight, enrich only new MQLs or use cheaper lookups.
- Skill level: This is no‑code but needs prompt design and Zapier field mapping knowledge.
- Workflow stage/team size: High volume favors batch or sampled enrichment to control costs. Small teams can run full real‑time enrichments.
- Output quality: Higher quality needs multiple data sources (company APIs, web signals) and richer prompt examples.

Practical checklist (step-by-step)
1) Trigger: New Contact in HubSpot (Zapier HubSpot trigger).
2) Gather context: Use HubSpot “Find Company” or a Zapier Clearbit/Crunchbase app to fetch company name, domain, industry, employee count. Also capture contact role, lifecycle stage, last touch dates, page views, and message text.
3) Normalize: Use Formatter (Zapier) to normalize domain, dates, and combine text fields into a single “context” variable.
4) ChatGPT call: Send a prompt that:
- Specifies output schema (JSON object with fields: company_enriched{industry,revenue_estimate,headcount_bucket}, intent_score:0–100, intent_reasons:[...], recommended_next_action).
- Gives 2–3 few‑shot examples (input → expected JSON).
- Asks for a short explanation only inside a “reasons” array.
5) Parse JSON: Have ChatGPT return strict JSON. Use Zapier’s Formatter/Utilities to parse the JSON into fields.
6) Write back: Update HubSpot contact/company properties via Zapier action.
7) Error handling: Add conditional steps (Paths) for missing domain or failed parse; route to a Slack alert or Tasks queue.
8) Test with 20 varied leads and tune prompt/weights.

Scoring tips (what to include in prompt)
- Signals: company fit (industry, size), behavioral signals (page views, demo request, pricing page), recency, contact role, message intent.
- Weighting: specify rough weights (e.g., behavior 50%, fit 30%, role 20%) or ask ChatGPT to output component scores and a combined 0–100.

Best‑for and avoid‑if
- Best for: midvolume real‑time enrichment, teams that want fast setup without code, workflows where automated intent scoring speeds SDR prioritization.
- Avoid if: you need guaranteed legal/PII-safe processing without vendor review, extremely high volume (costly), or fully auditable deterministic scoring (ML model preferred).

Final notes
Keep prompts strict and versioned (store prompt text in a Zap note or internal doc). Start with lower frequency (e.g., enrich only Marketing Qualified Leads) to validate cost and quality, then expand. If you want, start the Zap with a sample template prompt I use for scoring — I can paste it next.

Compare Zapier and Make