Prospect: ICP in, ranked leads out

Most lead lists start from a static database. This one starts from a live buying signal: a company that posted a job this week for the role your product serves. The pipeline then checks whether that company fits, whether it has money, and who to talk to.

Read references/conventions.md first. The spend and fail-closed rules there apply to every step.

Step 1. ICP

Call get_icp. If nothing is saved, run the define-icp skill. Do not invent an ICP from the user's company name.

Step 2. Estimate, then source

Call estimate_cost with the limits below and state the worst case. Then:

source_companies(max_companies=25, max_jobs_per_keyword=40)

Job boards match loosely: a search for "sales development representative" also returns "Inside Sales Representative - Flooring". The server ignores postings whose title does not contain every word of one of the ICP's hiring keywords and reports them as off_title_jobs_ignored. Expect to keep a third or less, so ask for about 4 jobs per company you want. If almost everything was ignored, the keyword is phrased differently on job boards; go back to define-icp and add the variant as its own keyword. Read the dropped list (agencies, exclusions) to the user in one line.

Sourced accounts usually have no industry, size or website yet. Scores mean little until Step 3.

Step 3. Enrich the top accounts

enrich_companies(top=10)

This runs three lookups in order: the Indeed employer page (cheap, and usually supplies the website that the other two are joined on), LinkedIn, then Crunchbase by domain. Report matched and mismatches_dropped. A mismatch is a namesake with a different website. It is expected for common names and is not an error.

Step 4. People and emails

Emails first, because the email step also returns named people for the ICP's departments:

find_emails(top=5, leads_per_domain=3, verify=true)
find_contacts(top=5)

find_contacts takes names from the enrichment data for free, then searches LinkedIn for the ICP's titles. Companies with no website are skipped by find_emails and listed in skipped_no_website.

Step 5. Show and export

list_leads(limit=10), then for the top 3 call get_lead and give, per account: score, the two strongest reasons, open roles, best contact, email and its verification status. Then export_leads() and give the file path and get_spend.

Judgment calls

• Score under 40 after enrichment: say it is a weak fit and why. Do not pad the list.
• No funding data does not mean no money. Bootstrapped and public companies score zero on funding. Say "no round on record".
• The user asks you to email the leads: refuse that part. This tool researches and sends nothing.