Score Incoming Leads So Follow-Up Time Goes to the Hottest First

You are a sales-operations adviser for an Australian small business drowning in mixed-quality enquiries. Your job: a lead-scoring model simple enough to apply in 30 seconds, honest enough to reflect how this business actually wins work — built from ITS patterns, not a SaaS template.

<context>
- Business and offer: [BUSINESS — e.g. 'Brightline Electrical, Cairns — resi jobs $200 to full rewires $15k']
- How leads arrive and what's captured at intake: [INTAKE — e.g. 'web form (suburb, job type, message), phone (whatever gets scribbled), Facebook DMs']
- What recent WON jobs had in common: [WON PATTERNS — e.g. 'named a specific job, local suburb, mentioned timeframe, referred or found us on Google']
- What LOST or bad leads had in common: [LOST PATTERNS — e.g. 'price-only first line, outside service area, vague "just getting quotes", wanted it done yesterday for nothing']
- Capacity reality: [CAPACITY — e.g. 'can properly chase ~10 leads/week; get ~25']
- Who applies the score: [SCORER — e.g. 'owner evenings, apprentice sometimes']
</context>

<task>
Before building, extract the signal set: from [WON PATTERNS] and [LOST PATTERNS], list the 6-8 observable-at-intake signals that actually separated winners from losers — each must be knowable from [INTAKE] data at the moment of scoring (a signal you can't see yet is a wish, not a criterion). Flag any pattern in the inputs that is NOT usable: anything scoring people by protected attributes (age, race, gender, etc.) is off the table — score the job and the fit, never the person's demographics.

Then:
1. Build the scorecard: each signal with its points (positive and negative — [LOST PATTERNS] earn deductions), weighted by how strongly the patterns support it (state the reasoning in one line each). Keep total arithmetic mental-maths friendly: a scorer in [SCORER]'s position must total it in their head.
2. Set the bands from [CAPACITY]: Hot / Warm / Cool thresholds chosen so the expected number of Hot+Warm leads per week roughly matches the 10 the business can chase — show the logic using the volumes given. Each band gets its follow-up rule: Hot = contact within [X hours] with a call; Warm = same-day message, call if reply; Cool = polite templated reply (draft the 3-sentence version) and a fortnightly batch check.
3. Stress-test the model on paper: score 3 hypothetical-but-realistic enquiries constructed ONLY from the patterns given (label them as constructed examples) — one obvious Hot, one borderline, one Cool — showing the arithmetic, so [SCORER] sees it applied.
4. Fix the intake to feed the model: the 2-3 questions to add to [INTAKE] channels that capture missing signals (worded for the web form and as natural phone questions), and the one-line scoring log (where the score gets written so it's not vibes-by-memory).
5. Set the review loop: monthly 20 minutes — compare scores against outcomes (did Hots convert? did any Cool turn into gold?), adjust one weight at a time, and the standing rule: the model routes attention, it doesn't refuse customers — a Cool lead still gets its reply.
</task>

<output_format>
'Usable signals' (with the exclusions noted) → 'Scorecard' (table: signal / points / why) → 'Bands and follow-up rules' (with the drafted Cool reply) → 'Three worked examples' → 'Intake fixes' → 'Review loop'. Under 1,100 words. Australian spelling, tradie-friendly plainness.
</output_format>

Rules: every signal and weight must trace to [WON PATTERNS]/[LOST PATTERNS] — no imported B2B-SaaS criteria (company size, job title) unless the inputs support them. Constructed examples must be clearly labelled and use no real names. If [WON PATTERNS]/[LOST PATTERNS] are empty, stop and set the 2-week homework instead: log every lead's details and outcome, then return — a model built on guesses just automates the guessing.

Copy the block above straight into Claude — anything in [BRACKETS] is yours to fill in.