HomePlaybooksSales & Leads

Sales & Leads

Two-Minute Inbound Lead Router

Score each inbound lead against a written rubric, draft the reply, and route it to an owner. A human sends.

Updated 30 Oct 2026

Difficulty
Starter
Time to implement
Half a day to install, then about 2 minutes per lead
Monetization potential
High

Goal

Score every inbound form lead against rules you wrote down, enrich the company in one pass, and draft a reply a human can send in under two minutes — routed to a named owner with a reason.

Tools required

Direct is the vendor's public site. Affiliate appears only when ShipPlays is enrolled in that program.

  • HubSpot

    Direct

    System of record for the contact, the owner, and the reply that was actually sent. Another CRM you already pay for is fine.

  • Clay

    Direct

    Firmographics from the work-email domain. Skip enrichment on personal domains.

  • Zapier

    Direct

    Form submit creates an unscored contact and posts a draft to Slack. It must not create a deal.

  • Slack

    Direct

    The queue where a human accepts or edits before anyone hits send.

Step-by-step setup

  1. Write a one-page rubric with four buckets

    The buckets are: book a call, nurture, disqualify, needs a human. Each bucket needs observable rules: employee band, country, role of the person who submitted, words they used, existing customer domain. If a rule needs a feeling, it is not a rule. Put owner logic in the same page: Nordics to Alex, otherwise Sam.

  2. Cut the form down to fields the rubric uses

    Required: work email, company, what they want, timeline. Hidden: page URL. Do not ask for a phone number you will not call. Say in the privacy line what the form is for.

  3. Create the contact as unscored

    On submit, Zapier creates the CRM contact in a stage named New — unscored. It sends the payload plus the rubric to the model. The model returns JSON only. It does not get a token that can send mail or edit deal amount.

  4. Post the draft where a human decides

    Slack gets the bucket, the score, two evidence bullets, missing facts, the draft reply, and the suggested owner. The human sends from the CRM so the thread lives on the record. Personal email domains (the big free providers) always land in needs a human, even when the note sounds like a buyer.

  5. Send the no

    Disqualified leads get the short honest note the draft already wrote. Silence trains people to try a second inbox. A clear no is part of the system.

  6. Edit the rubric on Friday, not the prompt

    Read ten scored leads. Any rule the model keeps missing gets added to the rubric in plain language. If the prompt is growing and the rubric is not, you are hiding policy inside a paragraph nobody else will maintain.

Exact prompts

Score and draft

On every form submit, after enrichment. Paste the rubric in full every time. Do not rely on memory from a previous lead.

You score one inbound lead. You do not send email, create deals, or change amounts. Return JSON only.

Rubric (the only policy):
{RUBRIC}

Lead:
{FORM_FIELDS}

Enrichment (may be empty):
{ENRICHMENT_OR_NONE}

Rules:
- Personal email domains go to "needs a human" unless the rubric explicitly says otherwise.
- If a required fact is missing, say so. Do not guess employee count or budget.
- Evidence bullets must quote or paraphrase the form or the enrichment. No "they seem like a great fit."
- The draft reply may not claim you read their website unless enrichment included a page.
- Disqualify drafts are short and specific. They do not upsell.

JSON shape:
{
  "bucket": "book a call | nurture | disqualify | needs a human",
  "score": 0,
  "evidence": ["", ""],
  "missing": ["", ""],
  "owner": "",
  "draft_reply": "",
  "why_this_owner": ""
}

Friday rubric edit

Once a week, on ten leads a human already corrected.

You maintain a lead rubric. You do not score a new lead in this pass.

Current rubric:
{RUBRIC}

Ten leads with the model's bucket and the human's corrected bucket:
{CORRECTIONS}

Return:
- Rules to add, in the same plain style as the rubric. Each rule names the bucket.
- Rules to delete because humans overrode them every time.
- Rules to leave alone.
- One sentence on the failure mode that showed up most.

Do not make the rubric longer for its own sake. Prefer replacing a vague line over adding a fifth exception.

Expected results

  • Every lead has a bucket, evidence, and a suggested owner during work hours.
  • Reps edit or send. They do not start from an empty box.
  • Disqualified leads get a clear no.
  • The rubric changes on Friday. The prompt stays a wrapper around it.

Guardrails

  • The agent never sends email and never edits opportunity amount.
  • Free-mail domains go to a human until you have a written exception.
  • The CRM is the record. Do not keep a second copy of form payloads in a random doc.
  • Do not claim research you did not run.
  • Match the privacy notice on the form. Do not add hidden enrichment you did not disclose if your region requires it.

Common mistakes

A 40-line prompt and no rubric.

Write the one-page rubric first. The prompt tells the model to follow the rubric and return JSON.

Auto-creating deals from the form.

Deals are created only after a human accepts a book-a-call lead.

Routing from a territory sheet the model has never seen.

Put owner rules in the rubric in words: which countries, which owner, what happens if the country is blank.

Related