Look at your calendar from the past week. Count the hours you spent on tasks that follow the same pattern every time. Writing follow-up emails. Preparing client reports. Drafting proposals. Reformatting content. Summarizing calls.
These tasks are eating your week. And you keep doing them manually because "it only takes 20 minutes." Multiply that by five days. That is almost two hours a week on a single task. Over a year, that is 100 hours of your life spent on something a well-built prompt could do in 30 seconds.
The fix is not to "use AI more." Most people paste something into ChatGPT, get a mediocre result, and decide AI is not ready for their work. The fix is to build dedicated tools: prompts that are specifically engineered for your exact business context, your exact workflows, and your exact quality standards.
This guide gives you the system to build those tools.
You will use a three-role process. A Consultant that diagnoses which tasks to automate. An Architect that designs the prompt logic. An Engineer that builds the actual production-ready prompt. Run all three in the same chat window. The output is a library of custom prompt tools you can reuse every day.
The Consultant
Analyzes your business to find the exact manual tasks that should be automated. Returns a prioritized list of automation opportunities.
The Architect
Takes your chosen task and designs the logic skeleton. Maps inputs, outputs, and determines if you need one prompt or a chain of prompts.
The Engineer
Builds the actual production-ready prompt for each step in your chain. Run this once per prompt in your inventory.
Identify your automation targets
Before you touch any AI, you need to know what to automate. Most people skip this step and pick the wrong tasks.
Look at your weekly work and sort your repetitive tasks into these categories:
Write down your top 5 most time-consuming repetitive tasks. For each one, estimate how many minutes it takes per occurrence and how many times per week you do it. The task with the highest total minutes per week is your first automation target.
Prepare your inputs
You need two things before running the first prompt. Both are about giving the AI enough context to understand your business.
Your offer description
What you sell, who you sell it to, and how you deliver it. This gives the Consultant context about your business model so it can identify the right automation opportunities.
"We run paid media campaigns for e-commerce brands doing $1-10M/year. We charge $3k/month retainer. Deliverables include weekly performance reports, ad creative briefs, and monthly strategy reviews."
Your lead magnet or entry point (optional but recommended)
How leads first encounter your business. If you have an [ICP_MEMO] from the Forensic AI Research guide or an [OFFER_MEMO] from the Create Your B2B Offer guide, paste those in. They significantly improve output quality.
Pick your model: This workflow requires a model that can hold context across a long conversation. You will run all three prompts in the same chat window.
Keep the same chat window
You must run all three prompts in a single conversation. Do not start a new chat between steps. The Architect needs the Consultant's output in memory, and the Engineer needs both. If you break the chain, the system loses context.
The Consultant
You should not guess what to automate. You are too close to your own operations. You accept slow, manual work because you have always done it that way.
The Consultant analyzes your business model and identifies the specific manual tasks where AI automation will save you the most time. It returns a prioritized list of "Automation Opportunities," each with a description of the current friction and a proposed AI solution.
Copy the prompt below. Paste your offer description (and ICP_MEMO/OFFER_MEMO if you have them) into the input sections. Run it in your chosen model.
The Consultant
Prompt 1Works with Gemini, Claude, ChatGPT
### ROLE DEFINITION You are the **AI Operations Architect & Conversion Systems Analyst**. Your expertise lies in deconstructing B2B sales funnels, identifying "operational drag" (repetitive, high-cognitive-load manual tasks), and designing generative AI workflows to eliminate that friction. You view business processes not as abstract concepts, but as sequences of logic that can be engineered into prompt chains. ### CONTEXT & OBJECTIVE The user will provide two foundational business assets: 1. **The Offer Memo:** Defines the product/service, the target pricing, the mechanism, and the ideal client. 2. **The Lead Magnet:** Defines the entry point, the free value provided, and the initial intent of the lead. Your mission is to analyze the "Conversion Gap" between these two assets. You must identify 5-10 specific, high-value opportunities where manual human effort is currently wasted and can be replaced or optimized by a **Generative AI Prompt Sequence**. ### CRITICAL DISTINCTION Focus on **"Prompt-able" Automation**, not just software automation. * Software Automation: Moving data from A to B (e.g., Zapier). * Prompt-able Automation: Tasks requiring reasoning, creativity, synthesis, or analysis (e.g., "Read this lead's LinkedIn profile and write a personalized connection request based on my Offer"). Focus solely on this category. ### STEP-BY-STEP ANALYSIS LOGIC 1. **Asset Decoding:** Analyze the `{{ Offer_Memo }}` to understand the complexity and cost of the sale. Analyze the `{{ Lead_Magnet }}` to understand the sophistication level of the incoming lead. 2. **Gap Mapping:** Visualize the necessary steps to move a user from consuming the Lead Magnet to purchasing the Offer. (e.g., Nurturing, Qualification, Objection Handling, Meeting Prep, Proposal Writing). 3. **Friction Identification:** Pinpoint steps that currently require: * Subjective decision-making. * Custom content creation. * Data synthesis/research. 4. **Solution Engineering:** Formulate how a specific AI prompt chain could execute this task instantly. ### INPUT DATA **[OFFER MEMO]** "" **[LEAD MAGNET CONTEXT]** "" ### OUTPUT FORMAT REQUIREMENTS Present your findings as a structured list of 5-10 "Automation Opportunities." You must strictly adhere to the following layout for each item: ## Opportunity [X] **1. Name of Task:** [Concise, Action-Oriented Title] **2. The Current Friction:** [Explain specifically why this task is a bottleneck. Is it slow? Does it require too much research? Is it mentally draining?] **3. The AI Solution:** [Describe the specific Prompt Logic. E.g., "A prompt that takes [Input A] and generates [Output B] using [Framework C]."] ### CONSTRAINT CHECKLIST * Do not suggest general business advice; stick to **workflow automation**. * Ensure every suggestion can actually be executed by an LLM (Large Language Model). * Prioritize tasks that directly impact revenue or save significant founder time. * Keep the tone professional, analytical, and strategic.
After running the Consultant
You now have a prioritized list of automation opportunities. Read through the list. Find the one that takes up the most mental energy. The one you procrastinate on. That is your first target.
Pick one opportunity. Note the number. You will reference it in the next prompt. Do not try to automate all of them at once. Build one tool, test it, then come back and build the next.
The Architect
You have a list of targets. Pick one. Now we need to design the logic before writing any prompt code.
Most people skip straight to writing the prompt. This is why their results are mediocre. Complex tasks require structure. You cannot build a house without a blueprint.
The Architect looks at your chosen task and breaks it down into steps. It maps the inputs and outputs. It determines if you need one prompt or a sequence of prompts chained together. The output is a "Prompt Logic Skeleton" that the Engineer will use to build the actual tool.
The Architect
Prompt 2Run in the same chat window as Prompt 1
### ROLE DEFINITION You are the **AI Systems Architect & Logic Engineer**. Your current phase in the development lifecycle is **Technical Specification**. You do not build products yet; you design the architectural blueprints that ensure the final product is robust, logical, and error-free. Your output must be a structural document, not an executable script. ### CONTEXT & GOAL The user has identified a business process to automate ("The Opportunity") and is now requesting a technical breakdown of *how* that automation will function. **Your Objective:** Create a detailed **"Prompt Logic Skeleton"** (a technical spec sheet) for the user's selected option. This serves as a validation layer to ensure the workflow is sound before we write the code. ### INPUT DATA **User Selection:** [WRITE THE OPTION NUMBER YOU CHOSE FROM THE CONSULTANT'S LIST] ### OPERATIONAL CONSTRAINTS 1. **CHAIN VS. SINGLE:** Determine if this task is simple enough for one prompt, or if it requires a "Prompt Chain" (breaking the task into smaller steps for higher quality). If it is complex, break it down. 2. **NO GENERATION:** Do not write the final prompts yet. Write *about* them. 3. **REALISM:** Only request variables (inputs) that are reasonable for a human to have on hand. ### OUTPUT ARCHITECTURE Produce a "Prompt Specification Sheet" using the following Markdown structure exactly: ## 1. The Operational Directive *A single, precise sentence defining the "Definition of Done" for this workflow.* ## 2. The Prompt Chain Inventory (Crucial) *Define exactly how many separate prompts we need to build to execute this workflow.* * **Prompt #1 [Name]:** (e.g., "The Analyzer" - Extracts data from input) * **Prompt #2 [Name]:** (e.g., "The Drafter" - Uses data from Prompt 1 to write content) * **Prompt #3 [Name]:** (e.g., "The Formatter" - Turns content into JSON) *(If only one prompt is needed, just list Prompt #1).* ## 3. Dynamic Variable Schema (The Inputs) *List the distinct data points the user must provide to start the chain.* ## 4. Algorithmic Logic Flow (The "Brain") *Map the step-by-step cognitive process across the prompts.* ## 5. Output Interface Design *Describe the physical look of the final result.* * **Format:** [Table / Email / JSON / Report] --- ### VALIDATION PROTOCOL End your response with this exact question: *"Does this inventory accurately map the workflow? If yes, command me to 'Build Prompt #1'. If no, tell me what to adjust."*
After running the Architect
You now have a "Prompt Chain Inventory." This is your checklist. It tells you exactly how many prompts you need to build. Review the logic. Make sure it matches how you actually do the work. If it misses a step, tell the AI to fix it before moving on.
Once the spec looks right, reply with "Build Prompt #1" to trigger the Engineer.
The Engineer
You have the blueprints. Now we build the actual tool. Run this prompt once for every item in your Prompt Chain Inventory.
The Engineer takes the Architect's spec and produces a production-ready prompt. If the Architect said you need three prompts in your chain, you will run this three times.
Each prompt it builds will include your specific business context hard-coded in, so you never have to re-explain your offer or audience when using the tool.
The Engineer
Prompt 3Run in the same chat window as Prompts 1 and 2
### ROLE DEFINITION You are the **Lead Prompt Engineer & Automation Pipeline Architect**. Your goal is to systematically build the tools required to execute the "Prompt Chain Inventory" we agreed upon in the previous step. You view prompts as modular code blocks that must function independently but integrate perfectly as a sequence. ### INPUT DATA (USER SELECTION) **TARGET_TASK:** [INSERT HERE WHICH PROMPT FROM THE INVENTORY TO BUILD NOW] ### CONTEXTUAL MEMORY & DATABANK 1. **Business DNA:** Recall the User's Offer Memo and Lead Magnet context provided in the Consultant step. 2. **The Inventory:** Recall the "Prompt Chain Inventory" (the list of prompts) defined in the Architect step. ### COMPILATION PROTOCOL (Internal Execution Logic) Based on the TARGET_TASK above, generate the **Production-Grade Prompt** for that specific step. Follow this engineering checklist: ## 1. Inventory Validation * Confirm which step in the chain this is (e.g., Step 1 of 3, or Step 2 of 3). ## 2. Input/Output Handshaking (Crucial) * **If this is Prompt #1:** Ensure the input variables are raw data the user possesses (e.g., {Raw_Transcript}, {Website_URL}). * **If this is Prompt #2 or later:** You must assume the input comes from the *output* of the previous prompt. Design the input variable to accept that specific format. * *Example:* If Prompt #1 output a JSON list, Prompt #2's input section must say: [PASTE_JSON_OUTPUT_FROM_STEP_1]. ## 3. Business Context Hard-Coding * Do not make the prompt generic. Implicitly "hard-code" the user's business constraints (Offer/Lead Magnet details) into the prompt instructions so they do not have to type them every time. ## 4. Chain of Thought Transcription * Translate the logic from the Blueprint into imperative, step-by-step instructions for the AI model that will run this prompt. ### RESPONSE FORMAT Your output must look like this: **[Prompt for TARGET_TASK]** ```markdown # ROLE [Specific Role] # CONTEXT [Business Context hard-coded here] # INPUT DATA [Variable_Name]: [INSERT DATA HERE] # INSTRUCTIONS 1. ... 2. ... # OUTPUT FORMAT [Specific format] ```
After running the Engineer
Testing your prompts
Open a new chat window. Do not test your prompts inside the Prompt Factory conversation. The AI holds too much context there. You want a clean environment that simulates what happens when you run this tool next week.
Copy the code block the Engineer gave you. Paste it into the fresh window. Replace the input brackets with real data from your business. Run it and check the output.
If your inventory has multiple prompts, take the result from Prompt #1, copy it, then paste the code for Prompt #2 and feed it the previous output. Continue until the chain is complete.
Building your library
Once a prompt produces output you are happy with, save it somewhere permanent. A Google Doc, a Notion page, a folder on your desktop. Label it clearly: what it does, what inputs it needs, and which model works best.
Then come back to this guide, pick the next task from the Consultant's list, and repeat the Architect + Engineer process for that task. Each cycle adds another tool to your library.
Organizing your tools
Group your prompts by function: email tools, content tools, client delivery tools, research tools. When a team member needs to do one of these tasks, hand them the prompt instead of explaining the process. The prompt IS the process.
30-minute walkthrough · Google Meet · Free
Red flags (bad output)
Your prompt tools are weak if any of these are true:
The Consultant returns generic advice like "use AI for email marketing" instead of specific tasks tied to your business model. If the opportunities are not specific to your offer, your input was too vague. Add more detail about your actual daily workflow.
The Architect's logic skeleton does not match your real process. If it describes steps you never do, or misses steps you always do, correct it before moving to the Engineer. Tell the AI what is wrong and ask it to revise.
The Engineer's prompt is too generic. It should contain your specific business context hard-coded into the instructions. If you see placeholder language like "insert your industry here," the Engineer failed to use the context from earlier in the conversation.
The output of your prompt chain sounds robotic or uses generic AI language. Go back to the Prompt Factory chat and tell the AI to rewrite the prompt with tighter tone constraints. Add a "Forbidden Words" list or a writing style section to the prompt.
Fix: debug inside the factory conversation
If your output has problems, do not try to fix the prompt code manually. Go back to the Prompt Factory chat window and tell the AI what went wrong. Say: "The output from Prompt #1 sounds too generic. Rewrite it with tighter tone constraints and add a forbidden vocabulary list." Let the AI repair its own work.
How to adapt this
If you want to automate multiple tasks
Run the Consultant once. It gives you a list. Then run the Architect + Engineer cycle separately for each task you want to automate. Each cycle adds one new tool to your library. Do not try to build all tools in a single conversation.
If you work with a team
Build the prompts yourself first. Test them with real data. Once a prompt consistently produces good output, save it as a shared document. Write a one-paragraph instruction above the prompt explaining what it does, what inputs to use, and which model to run it on. Your team members do not need to understand prompt engineering. They just need to paste and fill in the blanks.
If the output quality is inconsistent
Add more constraints to the prompt. The more specific the instructions, the more consistent the output. Add examples of good output. Add a "Do not do this" section. The Engineer prompt already does this, but you can ask it to tighten the constraints further after your first test run.
