A useful agency workflow makes the next person’s job clear. Here we build a small lead-to-booking process and a connected campaign pack, showing the evidence, drafts, review, and delivery at each handoff. Follow along · One small project Finish with: a five-account research sample, three reviewed outreach drafts, and a clear response-and-handoff plan. Bring: your approved offer, buyer criteria, permitted research sources, exclusion list, and the person responsible for replies. First practice limit: one 90-minute preparation session, five accounts, and zero automatic sends. These are worked teaching examples, not customer results. The time boxes are limits for a first attempt, not promises of completion. Read left to right: the first specialist’s accepted output becomes the next specialist’s input. The dashed path sends incomplete work back for revision. Set up Cameron and the handoff rules Select and copy the prompt. Replace the bracketed fields with your own information. You are Cameron, the coordinator for this one workflow. Use only the context I supply or sources you can actually open. Tell me which tools and delegation features are available before starting. If you cannot create subagents, complete the roles sequentially and label each handoff. Do not imply a role prompt installed a persistent bot. Return: task ID, owner, input version, output, source references, unknowns, review status, and next action. Stop when essential information is missing. Keep client data limited to what this task needs. Do not send, publish, spend, create accounts, or change live systems without approval of the exact action. Use the run's stated time and cost limits. No invented sources, quotes, contacts, or results. Example 1: build a lead-to-booking workflow Our fictional agency offers a booking-flow review to local home-service businesses. It is looking for observable fit, not guessing which owners have money or manufacturing “urgent” problems. Lead research and message preparation are separate stages. 1. Define a qualified lead before collecting names You → Cameron Write inclusion and exclusion rules. A list is useful only if a reviewer can explain why each business belongs on it. Illustrative output What a useful result looks like Include: serves the chosen area; sells the relevant service; has a public request form. Research question: is the next step clear on that form? Exclude: existing clients, active opportunities, suppression-list entries, and businesses outside scope. Ready to continue when: two people using the rules would make similar inclusion decisions. Copy the step 1 prompt Select and copy the prompt. Replace the bracketed fields with your own information. Cameron: convert this approved offer and target [paste] into a fit checklist. Include positive evidence, exclusions, fields to collect, and what must remain unknown. Limit the first batch to five businesses. Do not purchase data or contact anyone. 2. Build a small evidence-backed shortlist Cameron → Scout Ask Scout for public business information only, with a source and checked date for each substantive claim. Leave unverified contact details blank. Illustrative output What a useful result looks like Practice account Observation Decision Example Plumbing A (fictional) Request form exists; next-step text is missing Review for fit; response process unknown Example HVAC B (fictional) Service area excludes our target city Exclude Example Repairs C (fictional) Already appears on our suppression list Exclude; no outreach Ready to continue when: the list is deduplicated, exclusions are honored, and no personal email address has been guessed. Copy the step 2 prompt Select and copy the prompt. Replace the bracketed fields with your own information. Scout: research up to five accounts using [permitted sources] and [fit rules]. Return business name, source URL, checked date, exact observation, fit rationale, known business contact channel, exclusions checked, and unknowns. Mark unverifiable details as unknown. Research permission is not permission to send. 3. Draft a message from a real observation Scout → Relay Give Relay only the accepted accounts and approved offer facts. Review whether the proposed help follows from what you actually observed. Illustrative output What a useful result looks like Illustrative draft: “Hi [name], I noticed the quote form on [page] does not explain what happens after someone submits it. I prepared a short example of how that next-step text could read. Would it be useful if I shared it?” Removed claim: “You are losing dozens of leads every week”—there is no evidence for that statement. Ready to continue when: every personalized statement can be traced to a checked source and the offer is accurate. Copy the step 3 prompt Select and copy the prompt. Replace the bracketed fields with your own information. Relay: draft three individual messages from these accepted account records [paste]. Use one concrete observation and one relevant question. Do not invent familiarity, urgency, customer results, or revenue loss. Return the draft beside the evidence it uses. Keep all messages unsent. 4. Approve the exact contact action You → Cameron Before any send, review recipients, source records, final copy, sender identity, channel requirements, exclusions, and follow-up limits. A separate sending tool or provider configuration may be required. Illustrative output What a useful result looks like Review record: batch B01; three named recipients; final copy version 2; one named sender; one initial message; no automatic follow-up. In this guide: the batch remains a practice draft until you replace the fictional records and explicitly authorize a real action. Ready to continue when: the review record identifies exactly what is allowed and who handles replies. Copy the step 4 prompt Select and copy the prompt. Replace the bracketed fields with your own information. Cameron: present the recipient list, final message per recipient, sender, channel, exclusion checks, follow-up limit, and reply owner. Highlight missing sending prerequisites. Stop for my decision. Do not treat approval of a draft as approval of unspecified future messages. 5. Route replies without overpromising Relay → You → delivery owner Prepare response categories before a campaign. Positive interest, questions, not-now responses, and opt-outs need different next actions. A human confirms scope, delivery dates, and any commitment. Illustrative output What a useful result looks like Reply Prepared next step Owner “Can you show me?” Review and share the agreed sample You “What does this cost?” Use the approved scope and current price You “Not now” Record the response; follow only the agreed policy Cameron “Remove me” Stop outreach and update suppression records Named campaign owner Ready to continue when: a reply cannot trigger an unapproved promise or an endless follow-up loop. Copy the step 5 prompt Select and copy the prompt. Replace the bracketed fields with your own information. Relay: classify these replies [paste approved data] and draft the next response. Quote the request that supports your classification. Route scope, price exceptions, complaints, and uncertain intent to me. Never promise a booking or delivery date that has not been confirmed. 6. Measure the handoffs, then improve one Gauge → Cameron Keep activity, opportunities, and money in different fields. Use the first batch to learn where the workflow needs correction. Illustrative output What a useful result looks like Fictional teaching batch: 5 accounts researched → 3 accepted → 3 drafts reviewed → 0 sent. Result: a prepared batch, not three leads contacted and not sales. Next improvement: tighten the fit rules if too many accounts fail review. Ready to continue when: your report can distinguish prepared drafts, sent messages, replies, booked meetings, accepted proposals, and payments. Copy the step 6 prompt Select and copy the prompt. Replace the bracketed fields with your own information. Gauge: reconcile this run [records]. Count researched, accepted, drafted, approved, sent, replied, booked, proposed, and paid as separate states. Include human review minutes, tool costs, and missing evidence. Recommend one correction for the next batch. Example 2: build a campaign pack from one approved brief Cameron · Brief v1 Lock the business facts Audience, offer, current price, inclusions, exclusions, permitted proof, destination, and owner. Resolve missing facts before drafting. Spark · Drafts Connect the assets Create a landing-page outline, two ad concepts, and three email drafts. Each points to the same action and uses the same approved terms. Reviewer · Checks Find contradictions Compare headline claims, price, CTA, links, form destination, mobile layout, and what the recipient receives next. You · Release decision Approve a defined batch Sign off on final assets, audience, channels, schedule, spend limit, and owner. Gauge compares the actual outcomes with the original objective. Copy the campaign-pack prompt Select and copy the prompt. Replace the bracketed fields with your own information. Spark: create a campaign preparation pack from this approved brief [paste]. Deliver one landing-page outline, two ad concepts, and three email drafts. Attach a claim-to-source checklist and list every unresolved business fact. Use [approved CTA/destination]. Do not launch ads, send emails, or publish. Cameron: return only the decisions I need to make. Add specialists only where they reduce rework Forge can propose a fact-checker and a destination checker for the campaign. They can work independently on the same final version, then return findings to Cameron. Cap the review at two specialists and one revision round. A failed destination check blocks release; adding more writers does not fix it. Copy the specialist request Select and copy the prompt. Replace the bracketed fields with your own information. Forge: if delegation is available, configure two read-only reviewers for campaign version [ID]: claim/source checker and link/form-destination checker. Each returns pass, fail, or blocked with evidence. No child agents or external writes. Time box: 10 minutes each. If unavailable, give Cameron the same checklists to run sequentially. Watch the handoff, then try it The shared 24-second AI Work Crew overview. Open the interactive animation for lead generation, marketing, resale, and video examples. Read the animation instead You set the goal; Cameron assigns bounded work; specialists return evidence and drafts; Cameron packages a decision; you approve the defined action; Gauge records what happened. Take the workflow with you Download this example and its prompts (TXT) · Get all ten role templates (ZIP) For a guided working session, explore Agency training . Bring one workflow and the inputs you are allowed to use. Try another worked example Shape an offer with Waymaker Run the Wealth Experiment cost-and-time review Deliver a first AI Side Hustle project By Ashley Kays Rabbitt. Updated November 18, 2026. The named specialist roles are workflow templates; available tools and automatic delegation depend on your chosen platform. Live guide: https://agency.waymaker.cx/blog/ai-workflow-walkthrough-leads-to-bookings