By Dan Donovan, @askdandonovan
Every business has a job it still runs on phone calls, a spreadsheet, or a whiteboard nobody looks at. These prompts find the most expensive one and turn it into a brief an AI builder can build from without guessing.
Four copy and paste prompts. The first two are the ones from the video: one finds the app-shaped problems hiding in your business, the other turns the best one into a build brief. The other two are the ones I wish someone had given me: a safety check before anyone relies on the thing, and an adoption plan so it doesn't end up like the whiteboard.
You don't need to code. You don't need to know what a database is. You need to be able to describe how your business actually runs, which is the one thing no AI can do for you.
Open Claude or ChatGPT, paste this, and answer the questions honestly. It interviews you about how your week actually runs, then scores every candidate it hears. Free plans handle this fine.
You're helping me find the highest value internal app I could build for my business. I'm not technical, so keep your language plain.
Interview me first. Ask these one at a time and wait for my answer before moving on:
1. What does my business do, how many people work here, and what are their roles?
2. Walk me through a normal week. What happens Monday morning? What happens at the end of the month?
3. Where does information actually live right now? Name every spreadsheet, whiteboard, group chat, paper form, shared inbox and notebook.
4. What do my people phone or message each other about most often?
5. What question gets asked over and over that nobody can answer without interrupting someone else?
6. What has someone already tried to fix here that didn't stick, and why didn't it stick?
If I give you a vague answer, push back and ask for a specific example from last week.
Then build a table of every candidate process you heard. Score each one 1 to 5 on:
- FREQUENCY: how many times a week does this happen?
- PEOPLE BLOCKED: how many people are stuck waiting while it gets resolved?
- ANSWER ALREADY EXISTS: is the information already known by someone, and the only problem is reaching them? (5 = yes, someone always knows)
- ONE SCREEN: could this be a single screen with a list and a button? (5 = yes)
- LOW RISK: what happens if the app is wrong for a day? (5 = nothing serious)
Total the scores. Show me the top 3, each with one line on what the app would actually do, and then tell me plainly which one you'd build first and why.
Rules:
- Ignore anything that touches money, payroll, or customer records on day one.
- Strongly prefer problems where the answer already exists in someone's head and the app just makes it visible to everyone.
- Do not suggest anything that requires my team to type paragraphs. Taps only.Answer the six questions properly and this takes about fifteen minutes. Rush them and you'll get generic ideas back, because the model can only work with what you told it.
Before you run anything, know what you're hunting. Automation assessment work consistently points at the same markers: high frequency, high manual effort, standard inputs, and two or more people repeating the same task. Here's the version that matters for a small business.
Open a brand new conversation. Not one you've been using for something else. The interview in Prompt 1 depends on the model asking you clean questions, and old context makes it assume things about your business that aren't true.
Sit somewhere you can think for fifteen minutes. This is the step people skip, and it's the one that decides whether the output is useful or generic.
Paste the Opportunity Finder from the top of this page. Answer each question as it comes. Use real names, real numbers, real examples from last week.
Two things that ruin the output: describing how the business is meant to work instead of how it does work, and leaving out the embarrassing bits. The messy workaround somebody invented three years ago is usually the best idea in the room.
You'll get three options back with scores. The temptation is to pick the ambitious one. Don't. Run the shortlist past these four questions and take the boring winner.
This is the one that saves you the most money. An AI builder cannot infer what you left out. If your brief doesn't say "no user accounts in version one", most tools will build user accounts anyway, because most apps have them. Every boundary has to be written down on purpose.
Turn this into a build brief that an AI app builder can build from without guessing anything.
THE PROBLEM WE'RE SOLVING:
[paste the winning option from the last prompt, or describe it in your own words]
Before you write a single word of the brief, ask me every question you need answered. Do not assume anything. Ask them one at a time.
When you have enough, output the brief in exactly these eight sections:
1. ONE SENTENCE. What the app does, written in the words my staff would actually use.
2. WHO USES IT. Each type of user, what device they're holding, where they are when they use it, and how they get in (login, or a personal link they don't have to remember).
3. SCREENS. Every screen listed out. For each one: what's on it, and the single action a person takes there. If a screen has two actions, question whether it should be two screens.
4. THE DATA. Everything the app stores, written as plain nouns with their fields. Example: Tool = name, category, who has it now, status, last updated.
5. THE RULES. What happens on each action, in plain English. Include what the app must refuse to do.
6. NOT IN VERSION ONE. An explicit list of everything we are deliberately leaving out. Be aggressive. Assume that anything not on the "build" list will get built by accident unless you name it here.
7. DONE MEANS. One test that proves it works, written as a real scenario I can walk through on my phone in under a minute.
8. THE FALLBACK. What people do if the app is down or someone's phone is flat.
Constraints:
- Version one must be buildable in a day. If a feature can't be, move it to section 6.
- Every screen has to save a staff member time. Cut anything that only benefits me as the owner.
- Assume nobody gets trained. If a screen would need explaining, simplify it until it doesn't.
- No paragraphs of typing. Taps, searches and single fields only.
Finish by listing the three assumptions you made that are most likely to be wrong.Section 6 is the section that pays for itself. It's the difference between a builder producing what you asked for and a builder producing what it thinks apps normally look like.
You now have something worth handing over. Paste the whole brief in and add the line at the bottom of this step. Three honest options, cheapest first:
Add this line under your brief, whichever you use:
Build exactly what's in the brief above and nothing else. Section 6 is a hard boundary, not a suggestion.
Before you start, ask me any questions where the brief is ambiguous. Then tell me in plain English what you're about to build and wait for me to say go.
Build it mobile first. Everyone using this is on a phone, outdoors, on a bad connection, with one hand free.Two prompts, both run on the finished thing. Skipping these is how a handy internal tool turns into a problem you didn't know you had.
Review this app for the ways it could go wrong before my team relies on it. I'm not technical. Explain each risk in plain English, and for each one give me the exact sentence to send my builder to fix it.
THE APP:
[paste your brief, and describe what got built]
Go through these five:
1. ACCESS. Who can open this link? What can a stranger who guesses or is forwarded the URL see, change, or delete? If there's no login at all, say so bluntly and tell me whether that's acceptable for what this holds.
2. DATA. What personal or commercial information is in here? Staff names, customer details, addresses, pricing? Is there anything I'd be embarrassed to see leaked or that I'm legally obliged to protect?
3. THE DELETE PROBLEM. Can one person destroy real data with one tap? Is there any way to undo it? Is anything backed up, and if so, has anyone ever tested restoring it?
4. THE SILENT FAILURE. What could break without anyone noticing? What would tell me it had?
5. THE ONE PERSON PROBLEM. If the person who built this disappeared tomorrow, what would I not be able to change, and what should I write down now?
For each: the risk in one sentence, how serious it genuinely is for a business my size (don't inflate it), and the fix worded as a message I can paste straight to my builder.The app is built. Nobody has used it yet. The last attempt at solving this problem was a system everyone ignored, and I'm not repeating that.
THE APP:
[paste your brief]
THE THING THAT FAILED LAST TIME:
[what it was, and your honest guess at why nobody used it]
Give me:
1. THE FRICTION AUDIT. Count the taps from "person realises they need to do this" to "it's done". Include unlocking the phone and finding the link. If it's more than three taps, tell me exactly what to cut.
2. THE FIRST 60 SECONDS. What a worker sees the very first time they open the link, with no training, no app install and no account. Write the actual words that should be on that screen.
3. THE ONE RULE. The single behaviour change I need from the team, in one sentence I could say out loud on a job site without sounding like a manager.
4. WEEK ONE. A day by day plan. Who I show it to first and why that person. What I do on day three when someone hasn't used it once. What I stop doing myself so people can't route around it.
5. THE KILL TEST. One measurable thing that tells me in two weeks whether this worked. Include the number that means "shut it off", not just the number that means success.
Be blunt. If any part of this app is going to get ignored, tell me now rather than after I've rolled it out.Point 4 is the sharp one. If you keep answering the phone with the answer, nobody will ever open the app. The old way has to actually stop.
Same business, same day, two ideas. One of these is buildable this week and one is a project.
"An AI assistant that manages our whole job scheduling, talks to customers, quotes work, and syncs to the accounting system." Touches money and customers, needs everyone to change how they work, and there's no version of it that's finished in a day.
"A page that tells you who's got the core drill." One screen, one tap, zero training, nothing breaks if it's wrong for an afternoon. Saves a real person standing on a real site doing nothing.
If Prompt 1 comes back thin, it's usually because you described the business rather than the week. These are the places the answer is normally sitting.