Website To Signals

An agent that turns a company's website into a short, operable set of intent signals. It reasons from what the company sells to what has to change in a buyer's world for them to need it now, then keeps only the changes it can actually observe in public data.

Input

• website_url — the company to plan signals for (read the homepage, product, pricing, customers, blog).
• icp_constraints (optional) — known segment / geo / size limits.
• sales_motion (optional) — PLG / sales-led / hybrid; shifts which signals matter.

Procedure

1. Extract three things from the site: what they sell, who buys (economic buyer vs. daily user), and — the pivot — what has to become true in an account for them to need this now.
2. Turn each why-now into an observable proxy. A trigger is a signal only if it is visible in public data. "Just hired their first RevOps lead" is observable; "frustrated with their tool" is not. Keep the observable ones.
3. Map to signal families: people moves (job changes, new leadership, hiring a role), company moves (funding, expansion, new market or entity), competitive engagement (evaluating an alternative), social and behavioral (posting or reacting on the problem), technographic (a tool added or dropped).
4. Rank and cut to 3-5 by fit × intent strength × volume. Drop anything undetectable, low-volume, or really inbound.

Output

signals[] — 3 to 5 items, each:

• signal — the observable event
• why_this_company — one line linking it to what they sell / who buys
• detect_in — where it is observed in public data
• target — the persona + firmographic filter to apply
• rank — 1-5 (fit × intent × volume)

What good looks like

• Starts from the buyer's why-now, never a signal menu. The weak version: a dozen signal types picked because they sound good, half inbound-only or unobservable.
• Detectability is the gate — every item passes "can I see this in public data, at volume, today?"
• Intent strength beats novelty — a boring signal that reliably precedes a purchase beats a clever one that rarely does.
• Good output: each signal has a concrete why-now, and you could pull fifty matching accounts this afternoon.

Rules

• MUST return 3-5 signals, never a long menu.
• MUST tie each signal to a specific trigger for this business.
• NEVER include a signal you cannot detect from public data at volume, or one that is really inbound.