Recipe library / review-only automation

Simple AI automation recipes that keep a human in control.

Use these when you know the workflow and want the smallest safe build. AI drafts, organizes, summarizes, scores, routes, or reminds. A person approves before anything risky happens.

Default operating rule

Draft first. Approve second. Update last.

Most small businesses should not start with fully autonomous agents. Start with a review-only recipe that makes a useful work product.

Allowed early

  • Draft a reply
  • Summarize a call
  • Score urgency
  • Create a checklist
  • Route a task

Approval required

  • Send a message
  • Change CRM data
  • Publish a post
  • Spend money
  • Make a promise
Risk: medium / Best first build

New lead comes in → AI drafts reply → human approves → CRM updates

Tools needed: form or inbox, CRM or spreadsheet, email draft, Zapier or Make after the manual version works.

  1. Send new lead data into a review folder or CRM view.
  2. Ask AI to summarize need, urgency, missing info, and next reply.
  3. Owner reviews and edits the reply.
  4. Only after approval, send the reply and update lead status.

Human approval point: before any customer-facing message and before any CRM status change.

Copy/paste prompt: You are my lead review assistant. Read this lead. Summarize what they need, what is missing, the safest next step, and draft a short reply for human approval. Do not say it was sent. Do not invent facts.

Success looks like: every new lead has a clear next step within one business day.

Risk: medium / Local service

Missed call → AI drafts follow-up text → owner approves

Tools needed: phone system or call log, note field, SMS app with draft mode, owner approval.

  1. Collect missed call time, caller number, voicemail summary if available.
  2. AI drafts a short callback text and call task.
  3. Owner reviews for tone and accuracy.
  4. Owner sends or calls back manually.

Human approval point: before any SMS or call task leaves the business.

Copy/paste prompt: Draft a friendly missed-call follow-up. Keep it short. Mention only the known facts below. Ask one simple question. Leave placeholders for any unknowns.

Success looks like: fewer missed calls disappear with no response.

Risk: low-medium / Delivery

New customer → onboarding email and checklist

Tools needed: checkout or CRM status, docs, email draft, delivery checklist.

  1. Trigger when a customer is marked sold or paid.
  2. AI prepares welcome email, input checklist, and first milestone.
  3. Human confirms promise, timeline, and customer details.
  4. Send and save the checklist in the delivery folder.

Human approval point: before timeline, price, or deliverable promises are sent.

Copy/paste prompt: Build a customer onboarding checklist from this offer and customer note. Include what we need from the customer, what we will do first, and what the customer should expect next. Do not add promises not listed.

Success looks like: every new customer knows the next step without a custom scramble.

Risk: medium-high / Realtor-safe draft only

Realtor lead → buyer calculator + follow-up sequence

Tools needed: buyer info, calculator worksheet, CRM note, email draft. Keep compliance-sensitive facts human-verified.

  1. Collect approved buyer inputs.
  2. AI prepares a buyer question list and follow-up draft.
  3. Human verifies numbers, financing language, community facts, and timing.
  4. Human sends and logs next step.

Human approval point: before any buyer-facing message, payment estimate, financing claim, or CRM update.

Copy/paste prompt: Draft a realtor follow-up note from these approved buyer details. Keep it helpful and cautious. Do not give loan advice, legal advice, or unverified community facts. Include a human review checklist.

Success looks like: buyer follow-up is faster, clearer, and still human-owned.

Risk: low-medium / Creator

Content idea → post draft → approval → scheduling

Tools needed: source note, content doc, scheduler, approval checklist.

  1. Start from a reviewed source, customer question, or work product.
  2. AI drafts post, email, and short checklist.
  3. Human verifies claims, privacy, examples, and CTA.
  4. Schedule only after approval.

Human approval point: before publishing or scheduling.

Copy/paste prompt: Turn this source note into one useful post, one email idea, and one checklist. Use plain words. Do not invent results or testimonials. Mark any claim that needs verification.

Success looks like: content comes from real proof, not generic AI filler.

Risk: medium / Support

Customer question → answer draft → approval

Tools needed: inbox or form, knowledge base, draft reply, owner review.

  1. Capture the question and any order or account context.
  2. AI searches approved knowledge and drafts an answer.
  3. Human checks accuracy and tone.
  4. Human sends, then updates the knowledge base if needed.

Human approval point: before any answer goes to a customer.

Copy/paste prompt: Draft an answer using only the approved notes below. If the answer is not in the notes, say what needs human review. Keep it short and helpful.

Success looks like: common questions get answered faster without making things up.

Risk: low / Owner cockpit

Weekly business review → AI summarizes numbers

Tools needed: spreadsheet exports, CRM report, bank or sales summary if manually provided, weekly review doc.

  1. Export or paste approved weekly numbers.
  2. AI summarizes wins, leaks, stuck leads, next tasks, and questions.
  3. Owner checks the summary against source numbers.
  4. Owner chooses three actions for the next week.

Human approval point: before decisions, spending, outreach, or strategy changes.

Copy/paste prompt: Summarize this weekly business data. Show what changed, what needs attention, what might be wrong in the data, and the three safest next actions. Do not make final decisions.

Success looks like: the owner sees the business clearly in 15 minutes.