Lead Enrichment and Routing with a Claude Agent

Catch the submission and hand it to the agent instantly

Wire your form, whatever it is, a HubSpot form, a website form, Typeform, to fire a webhook on submission that hands the payload to your Claude agent. Instant is the whole idea: the agent starts working the moment someone hits submit, not on an overnight batch, because the intent you are racing decays in minutes. This thin webhook is the only glue left; everything after it is Claude.

Pass whatever the form has plus the email. Even a bare email is enough to proceed, because the agent fills in the rest from the domain. Do not gate the pipeline on a long form; long forms suppress conversion, and the enrichment makes them unnecessary anyway.

Grab the page they submitted from and any UTM or campaign parameters too. A demo request off the pricing page from a high-intent campaign is a different animal than a blog download, and the rep wants that in the alert. It is free to capture at submission and impossible to reconstruct later.

Enrich the lead with web research

The agent resolves the company from the email domain, then researches it: size, industry, the person's role and seniority, and the buying signals that matter to fit. For a work email, Claude's web research covers this in one pass, which is what makes second-level speed-to-lead real instead of aspirational. No separate enrichment platform sits in the middle for the everyday case.

Before it scores anything, have the agent check HubSpot over the MCP connector for an existing account or open deal. This is the single most important field in the whole flow, and it is a read the agent does itself, not a separate tool you wire in. Flag personal-email domains (gmail, outlook) in their own lane too, because there is no company domain to resolve and they usually carry lower B2B intent.

For very high volume, or for the deep firmographic and waterfall coverage a bulk cold list needs, you can still hand enrichment to Clay via its API and let the agent read the result. For everyday inbound, one work email and web research is plenty, and a lead that genuinely cannot be enriched should go to a human fallback, never get auto-filed as nurture on thin data.

• Company resolved from the email domain
• Company size and industry
• Person role and seniority
• Existing HubSpot record / open-deal check (read over MCP, the most important field)
• Buying signal: source page, message, campaign
• Personal-vs-corporate email-domain flag

Score the lead against your ICP

The same agent scores the enriched lead against your ICP and assigns a routing tier. Reuse the rubric from your account-scoring workflow so a sourced account and an inbound lead are judged by the same standard; that is what keeps the whole funnel coherent. The output is a simple, machine-actionable tier plus a one-line reason, because the agent branches on the tier and the rep reads the reason.

Keep it fast and decisive: hot, warm, or nurture, a clear reason, and an explicit existing-customer flag. This is a routing call made in seconds, not a deep account teardown, so the instruction should be lean.

Handle the existing-customer case right here in the score. If the HubSpot check shows a current customer or an open deal, the score says so and sends the lead to the account owner, never to a new-business rep. Cold-pitching your own customer is the most embarrassing routing failure there is, and it belongs handled before anything else can fire.

Score this inbound lead for routing. Use ONLY the enriched data.

Lead: {{FULL_NAME}}, {{TITLE}} at {{COMPANY}}
Company size: {{EMPLOYEES}}, Industry: {{INDUSTRY}}
Email type: {{EMAIL_TYPE}} (work/personal)
Existing HubSpot record: {{CRM_MATCH}}
Source page / message: {{SOURCE_PAGE}}, {{MESSAGE}}

Return JSON only:
{"tier": "hot|warm|nurture", "reason": "<one sentence>", "is_existing": true|false}

Rules:
- If the HubSpot match shows an existing customer or open deal: set is_existing true, and reason must say to route to the account owner (NOT new business).
- tier 'hot' if: ICP fit (industry in {{ICP}}, size in {{RANGE}}, decision-maker title) AND a work email AND a buying-intent signal (demo request, pricing page, 'pricing'/'demo' in message).
- tier 'nurture' if: clearly non-ICP, a student or job seeker, or a competitor.
- otherwise: tier 'warm'.
- Treat missing data as neutral. Do not penalize a blank field; a lead that failed to enrich goes to a human, not to nurture.

Write the record and route by rule

The agent writes the lead to HubSpot over MCP with all the enriched fields, the score, the tier, and the reason, then sets the owner per your rules: round-robin within a territory, by company-size segment, or to the existing account owner when it is a current customer. Your routing policy stops being a wiki page nobody reads and becomes something the agent runs identically every single time.

Handle the existing-customer case as the first branch, always. If is_existing is true, route to the account owner or CS and stop, never falling through to a new-business assignment. This one rule prevents the most damaging error there is, so it goes at the top where it cannot be skipped.

Give the agent a fallback owner for anything that fails the rules: the personal-email leads it could not enrich, the edge cases, the weird ones. A real human owns that bucket and works it, so a lead that matches no rule lands on a desk instead of disappearing into a gap in the logic. Build this before the clever rules, not after.

Alert the rep in Slack with context

The moment the record is written, the agent posts to the assigned rep in Slack, over the same MCP connector, no separate Slack app to wire up. The whole chain from submission to alert should finish in well under a minute; that speed is the entire value proposition, so measure it and protect it.

Make the alert genuinely actionable: who they are, where they work, the source page and intent, the tier and reason, and one-click links to the record and their LinkedIn. The Slack ping with context is what drives fast action. An assignment buried in the CRM gets noticed hours later, by which point speed-to-lead is already gone.

Start the speed-to-lead clock on assignment and put it in the message. A rep who reads 'this came in 90 seconds ago, reach out now' moves differently than one who has no idea how fresh the lead is.

Watch speed-to-lead and tune

Track two things: time from submission to first touch, and routing accuracy. If reps flag mis-routed or low-quality 'hot' leads, tighten the rubric, because a hot tier that is often wrong trains reps to ignore the tier entirely. If enrichment is thin for a meaningful share of leads, send the misses to the human fallback rather than letting a blank field drive a bad auto-classification.

Keep half an eye on model and research usage at volume. Cache what you already know: skip re-enriching a domain you enriched last week, and do not research the same company on every fill from it. Inbound spikes unpredictably, and an unbounded per-lead cost is a budget surprise waiting to happen.

Audit the nurture bucket for false negatives on a schedule, the real buyer who got mis-scored as nurture and quietly buried. That cost is invisible, you never hear about the deal you did not get, so you have to go looking for it. Once tuned, the agent runs itself, and you have faster speed-to-lead and reps focused only on qualified inbound.