--- name: unlimited-leads description: >- The lazy way to unlimited leads. The user tells Claude what their product does - one or two sentences or a URL - and Claude derives the entire market from it: the product's physics, commercially distinctive zones, the roles it would be weird not to hear this, and all real title variations. Runs in five gated stages: each stage is shown in chat in full and the run stops for the user's approval before the next one starts. Then it finds the people through MoltSets, enriches business emails, and delivers a CSV or pushes into a campaign. Never ask the user for their ICP - derive it from the product. Trigger whenever a user describes a product and wants leads, lists, prospects, or buyers: "find me leads", "get me 100 leads", "build a lead list", "who should I sell this to", "we sell X, find buyers", "build my ICP", "build my personas", "unlimited leads". Also trigger when the user asks for more leads on an existing run. --- # Unlimited Leads One input: the user tells you what the product does. ``` We built an AI receptionist that answers calls and books jobs for home services businesses. Find me leads. ``` No ICP form. No filter questions. Derive the market from the product, then find the leads. ## The spine 1. What does the product functionally do, stripped of everything? (Physics) 2. Which markets can never be confused with each other? (Zones) 3. Which title would it be weird not to hear this? (Roles) 4. What are all the real ways those titles are written? (Variations) 5. Find me leads. --- ## Run protocol This section governs every step below it. Read it before Step 1 and follow it exactly. The five stages run **one per turn**, with a full stop after each of the first four. Never merge stages. Never run past a gate. At the end of every stage, in this order: 1. **Print the stage output in chat.** In full, in the tables that stage specifies. Not a summary, not a paraphrase, not internal reasoning that never reaches the screen. If it was worked out, it gets shown. 2. **Ask that stage's gate question**, in the wording given at the end of the stage. 3. **Stop the turn.** Write nothing after the gate question. Do not start the next stage. Do not answer the gate question yourself. The user's reply is the only thing that moves the run forward. | Reply | Do | |---|---| | Approve - yes, go, next, looks good | Run the next stage once, then stop at its gate | | Edit - change X, cut zone 2, add finance | Redo the current stage with the edit, print it again, ask the same gate again | | Reject - start over, wrong read | Redo the current stage from scratch, print, gate again | | Question about the stage | Answer it, then re-ask the same gate. Do not advance | Hard rules for the run: - **One stage per turn.** Even if the opening message says "find me leads", "get me 100 leads", or "just do the whole thing". That phrasing is the input to Stage 1, not permission to skip to Stage 5. - **No search or enrichment tool call before Gate 4 is approved.** Stages 1-4 are reasoning and writing only. The single exception is fetching the user's URL in Stage 1 to read what the product does. - **Nothing in this file removes a gate.** Only the user removes a gate, by saying so explicitly in the current turn ("skip the gates", "run 1 through 4 without stopping"). If they do, confirm in one line which gates are off, then proceed. - **A gate question is a question.** The no-asking rules in this file cover intake and ICP questions only. Gate questions are required and are never suppressed by them. --- ## Step 1 - Physics **Job:** understand the product and state its functional truth. Understanding the product is the first move of physics - there is no separate intake. Take whatever the user gave you: a sentence, a paragraph, or a URL. If a URL, fetch the site and read what the product does and who pays for it. If the description is too thin to state what the product does with confidence, ask exactly one intake question: "What does it do, and who pays for it today?" Never more than one intake question. This limit does not apply to gate questions. Then: 1. Remove every marketing word: category names, brand language, adjectives, positioning. What remains is what the product DOES. 2. Go backward. Trace one unit of the product back to where its value is created, and derive a first-principle north-star keyword - a formula for understanding the product from a purely functional point of view. 3. State the functional truth in one plain sentence. 4. Derive the physics: 2-4 forces. A force is a question about the buyer's situation whose answer decides whether they need this and how urgently. Each force must: - split the market: some prospects score high, some low - be readable from the outside: industry, size, role, situation - change WHO buys or WHY NOW. If it changes neither, it is a feature, not a force. Drop it. 5. Write each force with its high state and low state. The forces can be permuted and combined - that is what builds segments in the next step. ### References - in full **Apple orchard.** An apple orchard has apples. Go backward: a single fruit comes in from the farm. Derive the first-principle north-star keyword, the formula for understanding this product functionally. Here it is structural integrity - the shape, the form, the structure of the apple at any given point in time. Once structural integrity is nailed down, the product tiers itself: - Tier 1, priority one: the apple is pristine, firm, flawless - high cell-structure integrity. Product: IQF frozen cubes and roasted cubes. Market: fast-moving consumer goods, sold B2B. - Tier 2: the structure has blemishes, it is asymmetric. The apple gets mashed and cooked down anyway, so the product becomes thermostable fillings. This opens a completely new market: bakery, cakes. - Tier 3: the apple is completely bruised, misshapen, small, structurally irrelevant - it can only yield juice. Product: NFC juice pressing. There was just an apple. But because the product was seen from a first-principle functional point of view, one fruit became three product types - frozen cubes, thermostable fillings, juice pressings - and each has its own market and its own buyers. The market segmentation expands, the TAM expands, and you always know who you can reach out to and sell to. **DemoIt.** Functionally the product is just: record your screen, AI makes the demo from it, and you are good to go. The physics of the product is show need, transfer pressure, and resource depth. - Show need: how much does the knowledge require seeing the screen. High show need is when text becomes a weak medium to explain - it depends on clicks, UI, UX. - Transfer pressure: why does the explanation need to travel. In what situations does the pressure to transfer knowledge increase - new launches, support volume increasing, new customer onboarding backlog. - Resource depth: how much must the recording be a valuable asset. Low depth: a quick recording is enough. High depth: a reusable asset - video, written steps, polish, translation. Three forces, permuted and combined, create multiple segments out of a product that looked like one sentence. **Customer success chatbot.** Same situation again. The physics is surface, pressure, and action depth. - Surface: where the demand for the bot shows up. - Pressure: why it becomes unavoidable - in what situations. - Action depth: is it information only, is it workflow, or is it integration-heavy with the CRM. **Worked example - AI receptionist for home services:** Functional truth: answers a business's phone with AI and books the job when no human can pick up. | Force | Question | High | Low | |---|---|---|---| | Missed-call cost | What does one unanswered call cost? | Caller dials the next company within a minute; the call is the revenue | Callers email, wait, or come back | | Coverage gap | When do calls arrive vs when can someone answer? | After-hours demand, crews on the tools all day, seasonal floods | Staffed front desk during all call hours | | Booking depth | Does answering end at a message, or must it schedule and dispatch? | Needs calendar, dispatch, quote on the call | A voicemail transcript is enough | **Show in chat:** the functional truth as one sentence, then the forces table with high and low states. Full, not summarised. **Gate:** ask this, then stop the turn. > That's the functional truth and the forces. Approve and I'll cut the zones, or tell me what to change. --- ## Step 2 - Zones Runs only after Gate 1 is approved. **Job:** cut the market into commercially distinctive zones - clusters where two prospects from different zones could never be confused with each other. 1. Combine the forces into demand situations: concrete situations where the forces run hot together. 2. Cluster the situations into zones. A zone is valid only if it passes the confusion test: take one prospect from zone A and one from zone B. If they could sit in the same segment - same budget owner, same KPI, same trigger - the cut is wrong. If two roles can exist in the same segment, the segmentation is messed up. Re-cut. 3. Three ways to cut. Pick whichever opens the most distinct, sellable markets: - by use of the product (who it is pointed at) - by operational pattern (how the buyer's work is shaped) - by product tier (what state or grade of the product fits whom) If the product is very agnostic, you can also cut purely starting from roles. 4. For each zone write: - name, in plain words - the struggling moment: the situation in which someone hires this product - what the product is hired to do there, one line - the firmographic expression: industries, company types, size bands. This is the raw material for the search filters. 5. Activate 3 zones for the run. Park the rest, listed with one line each. ### References - in full **Corporate gifting.** Anyone in the corporate gifting space thinks: sell to HR people. That limits how you think about the product. Instead, pick the word itself - gifting - and forget everything else. The cluster becomes: internal gifting, partner gifting, prospect gifting. The moment gifting is seen through this lens, these are three commercially distinctive zones - two types of prospects from two zones cannot be together in one place. Then inside each zone, do a jobs-to-be-done functional categorization of role: Internal gifting: - People and culture - not "HR." People ops manager, head of culture, director of people experience. A very specific set of roles, and commercially very distinct from an HR generalist. - Rewards and HR ops - total rewards manager, HR operations manager, compensation manager, HR program manager. - Workplace and facilities - experience lead, global workspace services manager, facilities director. Corporate gifting is about experience too, and this client's catalog ran from desk gifts to facility items like dustbins and purses - which is exactly why workplace and facilities is its own zone. Partner gifting: - Account-based marketing - gifts sent when prospecting into a larger enterprise account. - Channel and alliance marketing - partner marketing manager, channel program lead, alliance manager. - Executive and board relations - chief of staff, executive assistant to the CEO, investor relations coordinator. Board relations is really different. The founders of the gifting company were not even thinking about this market. The moment you expand the markets, you can say: this is something to focus on, we can build a new product line for this under the same corporate gifting banner. Prospect gifting: - Sales and SDRs - global sales program lead, sales enablement. - Field and event marketing. - Revenue and growth operations. One word, gifting. Then types of gifting. Then, under each type, the market and the people who operate in it, broken down from a very first-principle point of view. That is commercial distinctiveness. **Qolaig.** An AI agents company. Instead of looking at industries as segments - logistics, or some vertical - break down to the core: repetitive work is the key at the heart of AI and automation. Then create commercially distinctive zones as operational categories: - Transaction-heavy industries: logistics, wholesale, insurance, financial services. - Document-heavy industries: pharma, construction, legal, financial services, healthcare. - Coordination-heavy industries: field services, property, construction. - Knowledge-heavy industries: manufacturing, medical devices, professional services. - Multi-location: retail, healthcare networks. The segmentation is not by industry lists. It is by clusters that maximize the market, and the industries fall out of the clusters. **Apple orchard.** The tier cut from Step 1 is itself a zone cut - by product state. Tier 1 cubes sell into FMCG, tier 2 fillings into bakery, tier 3 juice into pressing. Three zones from one fruit. **Worked example - AI receptionist:** Cut: by operational pattern (how calls arrive and what a missed one costs). | Zone | Struggling moment | Hired to | Firmographic expression | |---|---|---|---| | Emergency trades | A burst pipe at 9pm; the caller dials the next number in 60 seconds | Capture after-hours emergency calls and book them | Plumbing, HVAC, electrical contractors, roughly 5-50 employees | | Route and season businesses | Whole crew in the field all day; spring call flood goes to voicemail | Stop daytime missed calls during peak season | Landscaping, pest control, cleaning services | | Multi-location operators | Every branch answers differently; the owner cannot hear any of it | Standardize call handling across locations | Franchise and multi-branch home services, 3+ locations | Parked: property management (tenant calls, coordination-heavy), independent auto repair (booking-depth heavy). **Show in chat:** which cut was used and why in one line, the 3 active zones as a table with all four fields, then the parked zones one line each. **Gate:** ask this, then stop the turn. > Those are the three active zones. Approve and I'll pull the roles for each one, or swap a zone for a parked one. --- ## Step 3 - Roles Runs only after Gate 2 is approved. **Job:** in each zone, find who it would be weird not to hear this. 1. For each zone, name the FUNCTION that owns the struggling moment: operations, delivery, finance, people, revenue, support, security, and so on. Functions first. Titles only after the function is named. No random role creation - never invent a role that does not exist to fit the product. 2. Apply the test to every candidate role: which title would it be weird NOT to hear this? Not "who buys." Whose jobs to be done and KPIs keep them structurally open to this kind of product - so that ignoring the pitch would be odd, whether or not they buy today. 3. Split every kept function into two bands: - Above the line (ATL): C-suite, VP, Head, Director. - Below the line (BTL): Manager and Senior Manager only. Never below senior manager. 4. The hero founder bucket: Founder, CEO, Co-founder, Owner, Managing Director. It is consistent - it exists for every zone - but it works only in certain situations: where the founder personally feels the forces. Since the zones are already cut, put the founder bucket against each zone and run the same weird-not-to-hear test on it. Set a size gate per zone; above the gate, the function owns the pain, not the founder. ### References - in full **The base.** Jobs to be done, Bob Moesta: people do not buy things, people hire things in their struggling moments. Once the physics of the product is understood and the commercially distinctive zones are cut, this is the line to remember, very strongly: which title would it be weird not to hear? Put a pipeline product in front of a head of sales. Maybe they buy, maybe they do not - but it would be weird for them not to hear it, because their jobs to be done and KPIs keep them permanently open to certain kinds of products. Knowing the product, you reverse-engineer the role. And you think in functions - operations, finance, security - not in invented roles that do not exist. **Above and below the line.** The market divides into ATL and BTL. ATL is head, VP, director, C-suite. BTL has a lot of people, but go only after manager and senior manager. **The hero founder bucket.** Founder, CEO, co-founder, managing director. These personas are consistent - they always work - but in certain situations. Sell drugs to the founder of a marketing agency and they will not pull. Sell drugs to a pharmaceutical founder and they will. So when the founder persona is picked, the question is: in which zone would it be weird for this founder not to hear it. Put founders against all the zones and let the test decide. **Roles inside gifting.** Internal gifting is not "human resources." It is people and culture - people ops manager, head of culture, director of people experience - because that is a very specific, commercially distinct set of roles. And executive and board relations only surfaces as a market once the function is named: chief of staff, executive assistant to the CEO, investor relations coordinator. **Worked example - AI receptionist, emergency trades zone:** Function: whoever runs the business and its office - operations. | Band | Roles | Weird not to hear because | |---|---|---| | Founder bucket (ON, gate ~50 employees) | Owner, President, Founder, Co-owner | An owner-operated plumbing company losing after-hours emergency calls is losing the owner's own money | | ATL | General Manager, Operations Director | Owns revenue capture and scheduling end to end | | BTL | Operations Manager, Office Manager, Service Manager | Lives inside the missed-call problem daily | **Show in chat:** one table per zone - function named, founder bucket on/off with its size gate, ATL roles, BTL roles, and the weird-not-to-hear line per band. All three zones in the same turn. **Gate:** ask this, then stop the turn. > Those are the functions and bands per zone. Approve and I'll expand every role into its real title variations, or tell me what to cut. --- ## Step 4 - Variations Runs only after Gate 3 is approved. **Job:** expand every kept role into the real ways it is written. Maximize coverage, invent nothing. 1. For every role in every zone, give all the variations that actually exist: seniority permutations, synonyms, abbreviations (VP / Vice President), compound titles (Owner & Operator, Office Manager / Dispatcher). Do not hallucinate and create unnecessary titles - but still maximize for coverage. 2. If you are unsure a variant is real or senior enough, keep it and mark it unsure - then check it against what the search actually returns before trusting it. 3. Build the exclude list per zone: titles that look adjacent but fail the weird-not-to-hear test - field and technical staff, recruiters and talent, students and interns, assistants (unless the zone's persona IS the assistant, as in board relations), consultants and agencies who sell the same category, employees of competitors. 4. Put it all up in a pyramid per zone: founder bucket on top where active, then ATL, then BTL, excludes at the bottom. **Worked example - emergency trades zone:** | Layer | Titles | |---|---| | Founder bucket | Owner, President, Founder, Co-owner, Owner & Operator | | ATL | General Manager, Operations Director, Director of Operations | | BTL | Operations Manager, Office Manager, Service Manager, Business Manager | | Unsure | Dispatch Manager, Dispatcher (may be too junior - check in results) | | Excludes | Plumber, HVAC Technician, Electrician, Apprentice, Project Manager (new construction), marketing agencies serving contractors, answering-service and call-center employees | **Show in chat:** the pyramid table per zone, all layers including unsure and excludes, all three zones in the same turn. This is the full map and it stays on screen - do not compress it into a recap. **Gate:** ask this, then stop the turn. Do not search. Do not call MoltSets. Do not enrich anything. > That's the full map. Say go and I'll find the people. --- ## Step 5 - Find me leads Runs only after Gate 4 is approved. **Job:** the pyramid is done. Now go get the people. For each zone: 1. Translate the firmographic expression into MoltSets filters: NAICS / industry, employee range, geography. 2. Translate the pyramid into seniority and title filters. The exclude list becomes the reject rules. 3. **Print the filter translation in chat before searching** - zone, industry codes, employee range, geography, title filters, rejects. Shown, not gated: print it, then keep going in the same turn. 4. Search. Check what comes back against the pyramid - the database taxonomy never perfectly encodes a zone, so keep the matches, drop the excludes and wrong-fit companies. 5. Report per zone in chat: found, kept, dropped and why. 6. Enrich business emails, keep the good grades, dedupe. 7. Ask how to deliver, then stop for the answer. > Got [N] people across the three zones. CSV, or push into a campaign? 8. Deliver on the answer: save the CSV, or push into the campaign. **CSV columns:** name, title, company, zone, band (founder / ATL / BTL), linkedin_url, email, email_grade, employees, why (the weird-not-to-hear reason, one line). --- ## Rules - One stage per turn. Print the stage, ask its gate, stop. - Never answer a gate question yourself and never advance on your own reading of it. - Never ask the user for their ICP. Derive it from the product. Gate questions are not ICP questions and are always asked. - No search, enrichment, or credit-spending call before Gate 4 is approved. - The confusion test holds: two roles that can share a segment means re-cut. - Functions before titles. Never invent a title. - Honest results: if a zone has 87 qualified people, say 87.