The AI Prompt That Finds What to Automate in Your Business

TL;DR. Paste the interview prompt into ChatGPT, Claude, or Gemini and answer its questions for about ten minutes. You get a ranked shortlist table: task, hours per week, who does it, feasibility, automate-it vs AI-feature, DIY vs needs a build. Run the 90-day abandonment test before spending any money. Most first automations should be DIY. When the top item needs a real build, our $500 AI Game Plan scopes it in a week, credited 100% against the build.

The audit prompt: paste this into ChatGPT, Claude, or Gemini

Copy the block below into a fresh chat. It works the same in ChatGPT, Claude, and Gemini. Answer one question at a time, honestly, and in about ten minutes you'll have a ranked shortlist of what's actually worth automating, built from your answers, not the model's guesses.

You are a senior automation consultant who audits small businesses for a living. Interview me about how my business actually runs, then produce a ranked automation shortlist I can act on.

Rules for the interview:
- Ask ONE question at a time and wait for my answer. No multi-part questions.
- Ask at most 8 main questions. A follow-up to pin down a vague answer doesn't count against the 8.
- Push back when I'm vague. "Admin stuff" is not a task. "A lot" is not a number of hours.
- Don't suggest a single automation until the interview is done.

Cover, in this order: my industry and what I sell; team size and roles; the top 10 recurring tasks we do every week, with rough hours per week for each; who does each task; the tools we already use (CRM, email, spreadsheets, WhatsApp, accounting); and what breaks down or gets dropped most often.

Then output ONE artifact, the automation shortlist table, with exactly these columns:
Task | Hours per week | Who does it | Feasibility (high/medium/low) | Tag (automate it / AI feature) | DIY tool or needs a build

Rules for the table:
- Max 10 rows, ranked by hours per week times feasibility.
- Feasibility is high when the task follows the same steps every time and the data lives in a tool with integrations. It's low when the task needs judgment or the data lives in someone's head.
- Tag "automate it" for internal workflows and "AI feature" for anything a customer would touch.
- In the last column, name a real tool (Zapier, Make, my CRM's built-in rules), write "needs a build", or write "not yet" plus a one-line reason. Never write "various tools".
- Under the table, add exactly 3 lines: the biggest time sink, the easiest win, and the one task I shouldn't bother automating. Nothing else.

Copy this prompt

Two things make the output better. Answer with real tasks, not categories: "chasing unpaid invoices every Friday" beats "admin". And don't flatter the hours column. Count the interruptions, the re-dos, the WhatsApp back-and-forth. Owners routinely underestimate their biggest time sink, and the ranking is only as honest as those numbers.

The artifact: a ranked automation shortlist

The prompt ends in one artifact you can act on. Here's what a filled-in shortlist looks like for a five-person trades business:

Each column earns its place. Hours per week is the size of the prize. Feasibility is whether the task follows the same steps every time and lives in a tool something can connect to. The tag tells you whether it's internal plumbing or something your customers would touch. And the last column is the honest one: most rows don't need a developer, they need an afternoon in Zapier or your CRM's built-in rules. If you're still wondering what's even on the menu, what AI can do for your business maps the terrain.

Why one interview beats ten template prompts

Most pages targeting this search hand you ten template prompts: one for email, one for scheduling, one for reporting. They read well and produce the same list for every business on earth, because the model doesn't know your team, your tools, or where your hours actually go. So it guesses, confidently. You get a document full of automations nobody needs.

The interview flips it. The model asks, you answer, and every row in the table traces back to something you said. The AI contributes structure and pushback, which it's genuinely good at. You contribute the facts, which only you have. If you searched "ChatGPT prompt to find what to automate in my business", this is that prompt. It just works by asking first.

The same audit finds your AI use cases

If you came here hunting "AI use cases for my business", you're the same buyer, and you don't need a second audit. The shortlist is your use-case list. Every opportunity gets one of two tags: automate it, an internal workflow like invoice chasing or data entry between tools, or AI feature, something customer-facing like a quote bot on your site or an assistant that answers before your competitors wake up.

The distinction matters because the economics differ. Workflow automations save hours and usually start DIY. AI features make money and usually need a build: real engineering, error handling, AI agent development cost territory. We've built the second kind. Henny Bot, a trading bot that lives inside a Telegram chat, took on-chain volume up 1,200%, carried about $300,000 of trading a day at its peak, and saved five minutes on every transaction. That started life as one row on a list like yours.

When an AI-feature row ranks high, don't spec it in your head. Run the app requirements prompt to turn it into a one-page sheet a developer can price, and read how to integrate AI into your business before you commit to anything.

The reality check: the 90-day abandonment test

Most small business automations don't fail on technology. They get abandoned. Someone renames a spreadsheet column, the Zap breaks silently, and six weeks later everyone's back to doing it by hand and nobody remembers why the tool subscription still bills. So before you build anything, make your AI try to kill the list.

Here is my automation shortlist: [paste your table]

Run the 90-day abandonment test on it. Assume every automation on this list fails or gets abandoned within 90 days until my answers prove otherwise, and say KEEP only on specific evidence, never on my enthusiasm.

Go row by row. For each one, ask me, one at a time:
1. What exact event triggers it? ("a new row lands in the form responses sheet", not "when we get a lead")
2. Which specific tool runs it, and does that tool actually connect to where the data lives today?
3. Who fixes it when it silently breaks in week 6, and do they know they own it?

If I can't name the trigger, the tool, and the maintainer, mark the row FANTASY and give the reason in one line. Watch for the classic tells: the task needs judgment nobody has written down, the data lives in screenshots or someone's memory, or the "automation" is secretly a new process nobody has agreed to follow.

Output my table again with a final Verdict column: KEEP or FANTASY, each with its one-line reason. End with the count: X of Y survive.

Copy this prompt

A FANTASY verdict isn't an insult. It's a saved subscription and a saved weekend. The rows that survive have a named trigger, a named tool, and a named human. Those are the only ones worth money.

The ROI ranking: what to do first

Survivors get ranked by one blunt equation: hours saved, times what those hours cost you, minus what the automation costs to build and keep alive. Maintenance is the part every listicle skips, and it's the part that quietly kills the ROI on marginal automations.

Here are the KEEP rows from my automation shortlist: [paste them]

Rank them by rough monthly ROI. For each row, calculate: hours saved per month, times the hourly value of the person doing the task (what I'd pay to buy back an hour of their time; ask me if you don't know), minus a rough monthly cost to build and maintain it. A DIY tool counts its subscription plus an hour of upkeep. A custom build costs real money up front, so amortize it over 12 months.

Output a ranked table with these columns:
Task | Hours saved per month | Rough monthly value | Rough monthly cost | Net | Do it how

In "Do it how", make one honest call per row: "a Zapier afternoon handles this" when an off-the-shelf tool covers the whole flow, or "this needs a real build" when it doesn't (multiple systems, judgment steps, anything customer-facing).

Then name the top 3 to do first, in order, with one line each on why. If only one row clears the bar, say so. Don't pad the list to 3, and if nothing is worth doing yet, tell me what to fix in the business first.

Copy this prompt

The "do it how" column is the decision that matters. A Zapier afternoon is something you do this week, and you can do it without hiring anyone. A real build is a purchase, and a purchase deserves a real scope.

DIY your first automation. Hire for the build.

Here's the honest routing, and it starts with: don't hire us yet. Most small businesses should DIY their first automation. Pick the top "Zapier afternoon" row, ship it, and watch whether it survives 90 days. You'll learn more about your own operations from one live automation than from any audit, ours included.

Hire when the top item is a real build: multiple systems, judgment steps written into logic, or a customer-facing AI feature. That's what our business automation service exists for. The AI Game Plan is $500 flat for one week: we take your shortlist, map it against your actual systems, and hand back a fixed-price plan for the build that's worth doing, with the $500 credited 100% against any build. Installs start at $2,000. Fixed price agreed before we start, senior engineers only, a working demo every week before each payment.

And if you'd rather shop around, do. Best AI automation agencies for small business and best business automation agencies rank the field, us included. A recent lead found us by asking Gemini for help with their problem, which tells you where buying starts now. It starts where you already are: in a chat window, with a good prompt.