You Are Not Bad at AI, AI Is Bad at Remembering
It was about a year ago. I had blocked off the morning. No client calls. No interruptions. Just a clean stretch of time to sit down with ChatGPT and finally build our client onboarding process from scratch.
Thirty minutes in, I was fishing.
Good answers were wrapped around silly mistakes. Strange details were landing in the document from nowhere, almost like the AI had decided to fill in the gaps with whatever felt right.
I closed the laptop and walked away.
Maybe you have done the same thing.
Three exchanges in, things are moving, you are getting somewhere, then something stops making sense.
Before long, you are reading the output and wondering why more and more random ideas are getting mixed in with your actual directions.
They sound confident, they look polished, but they do not make sense.
I have been there and I finally solved this puzzle for good.
This week, I’ll share how I used AI to rebuild our client onboarding process on the second attempt and got it right.
If you run an interior design studio, an architecture firm, or any other business where complexity never sleeps, what I am about to share may permanently change how you use AI.
Before You Proceed
Take a deep breath and ask yourself:
“Did I come here to solve a problem that is not actually mine to solve?”
You came here with something in mind.
A process to build. A system to fix. A procedure that has been sitting unfinished for longer than you want to admit.
And somewhere in the back of your mind, a small voice might be telling you this is not actually yours to solve.
If that voice is there, listen to it.
Send this article to the person it belongs to. Ask them what problem brought you here in the first place. Then go do the thing only you can do.
I delegate aggressively. I have not filled out vendor paperwork, filed an online application, or sent an invoice in a very long time.
But there are things in my business I will not hand off.
Client onboarding is one of them.
Not because I do not trust my team.
Because when a client says yes, they are not just writing a check. They are investing time, energy, and trust in us.
What our onboarding process says to them in those first moments reflects my values and my sense of what their commitment deserves.
I cannot delegate gratitude.
I cannot outsource the thing that communicates what I actually believe about the people who hire us.
So I built it myself.
If this is yours to build, keep reading.
What Nobody Told You About AI
Here is the thing that changes everything.
It is simple, and almost nobody says it directly.
AI does not have great memory.
More specifically, AI is not good at holding the full picture when a conversation drags on. The longer the exchange, the more context starts slipping through the cracks.
And when context disappears, something fills the gap.
The AI's imagination.
💡 It produces work that looks complete at first glance, but when you actually read it, it is not complete, accurate, or helpful.
When context disappears, something fills the gap. And that something is the AI's imagination.
That is not a flaw you caused.
That is the architecture.
The frustration most people feel is not because AI is bad at its job. It is because nobody told them it has a memory problem they can design around.
Once you understand that, the entire approach changes.
You stop fighting the tool and start building a smarter conversation.
The Framework That Fixed It
What I am about to describe is built around three ideas.
They are not complicated.
But together, they are the reason I went from walking away frustrated to finishing a complete, significantly improved client onboarding standard in a single afternoon.
1. Make your opening ask as complete as possible
The goal is to keep the overall conversation short.
A shorter conversation means less context is at risk of slipping away or being replaced by the AI's imagination.
Pack everything you know into the first message. Every relevant detail. Every constraint. Every thing you are not sure about. Every preference, concern, and messy reality the AI needs to understand.
The more you give upfront, the less the AI has to invent.
2. Let AI interview you
This is the shift most people never make.
Most people approach AI like a search engine.
Ask a question. Get an answer. Repeat until the results stop making sense.
What actually works is almost the opposite relationship.
Give AI your thinking. Then ask it to find the gaps, address the conflicts, and resolve anything that needs clarification.
One question at a time.
Not a list. Not five follow-ups. One question.
✅ Answer it. Walk away if you need to. Come back when you have seven minutes between calls. The conversation holds. The progress does not stall.
One question means you can answer it, close the laptop, and come back when you are ready. Progress does not have to wait for a clear calendar.
The questions also get sharper the further you go.
Each answer feeds the next one. By the end, the AI is not guessing about your business. It has been paying attention, building a picture from everything you have told it, and asking what is still missing.
3. Use the right model for the right job
If you are working with Claude, Anthropic's AI platform at claude.ai, there are two models worth knowing: Sonnet and Opus.
Think of Sonnet as a sharp senior engineer.
Structured. Logical. Built for execution.
Think of Opus as a brilliant scholar.
Deep. Nuanced. Built for ambiguity and analysis.
Neither is smarter than the other. They simply have different specialties.
Sonnet builds. Opus analyzes.
Use them that way, and the outcomes can genuinely surprise you.
If you are working with ChatGPT or Gemini, the framework still works. You will just run the whole process with one model from start to finish.
The principles do not change.
One Afternoon. Here Is What It Looked Like.
This framework works with ChatGPT, Gemini, or Claude.
For this example, I am going to use Claude because that is what I used. Specifically, I used Sonnet for building and Opus for deeper analysis.
You can still follow the same process with another AI tool. The names may change, but the principles stay the same.
Step 1: Start with a brain dump
Before opening any AI platform, open a blank document.
Notepad. Google Docs. Anything with an empty page and no distractions.
Write everything down about your client onboarding process.
What happens after a client says yes.
What the first email should say.
What information you need to collect.
What your team needs to do internally.
What clients tend to ask.
What feels clunky.
What keeps falling through the cracks.
What you have been meaning to fix for years.
Write until you feel lighter, like you finally let it all out onto the page instead of carrying it around in your head.
That document is your brain dump.
It is the raw material for everything that follows.
Step 2: Close the gaps with Sonnet
Open a fresh chat with Sonnet and give it this prompt:
Prompt to use:
I want to transform our client onboarding experience to remove friction for new clients, reduce effort for our team, and eliminate anything falling through the cracks. You are an expert at analyzing systems for small businesses in [your industry] with [your team size] employees. We are a small team serving [target market]. I am going to give you a brain dump I put together. Review everything in it and start asking me questions, one question at a time, to address any conflicts, gaps, and areas that need clarification while aligning every step to industry best practices.
Then paste your brain dump.
Let the interview begin.
Answer each question. Walk away when life interrupts. Come back when you can.
By the time the interview is complete, you will have a comprehensive, well-organized picture of what your onboarding process needs to look like, built from your thinking instead of the AI’s imagination.
Step 3: Get your complete summary
When the interview wraps up, the AI may ask if it should build the onboarding system now.
Say no.
This part matters.
Ask for a complete, detailed summary instead.
Every change.
Every addition.
Every decision.
Every open question.
Every point of clarification.
Captured in one document with nothing important left out.
Prompt to use:
Before building anything, create a complete and detailed summary of this conversation. Include every important decision, change, addition, open question, clarification, and recommendation related to our client onboarding process. I want this captured in one organized document with nothing important left out, so I can use it as the source material for the next step.
Copy that summary.
Save it.
Then open a brand new chat.
Step 4: Ask Sonnet to build the first version
Give Sonnet the summary from the previous chat and ask it to produce four things:
Complete client onboarding documentation organized in the correct hierarchy
A checklist your internal team can actually follow
Client communication email templates for every onboarding touchpoint
AI prompts for the steps between, such as call transcription analysis, client data organization, onboarding summaries, and handoff notes
Prompt to use:
Using the summary below, build the first complete version of our client onboarding system. Please produce four things: complete onboarding documentation organized in the correct hierarchy, a checklist our internal team can actually follow, client communication email templates for every major onboarding touchpoint, and AI prompts for the steps between, such as call transcription analysis, client data organization, onboarding summaries, and handoff notes. Keep the output practical, clear, and usable for [your industry] serving [your target market].
This is the moment that feels like a jackpot.
And you are only halfway there.
Step 5: Bring in Opus for deeper analysis
Take those four documents and open a new chat, this time with Opus.
Use the same opening prompt you used with Sonnet.
The only change is this:
Instead of saying, “I am going to give you a brain dump,” say, “I am going to give you the onboarding documents we created.”
Prompt to use:
I want to transform our client onboarding experience to remove friction for new clients, reduce effort for our team, and eliminate anything falling through the cracks. You are an expert at analyzing systems for small businesses in [your industry] with [your team size] employees. We are a small team serving [target market]. I am going to give you the onboarding documents we created. Review everything and start asking me questions, one question at a time, to address any conflicts, gaps, weak spots, missing details, and areas that need clarification while aligning every step to industry best practices.
Opus will ask questions.
The kind that surface things Sonnet missed because Sonnet was building, not analyzing.
Answer them one at a time.
Take your time.
This conversation is worth slowing down for.
Step 6: Ask Opus for the decisions document
When this interview ends, say no again.
Do not ask Opus to rebuild the whole onboarding system.
Ask it for one document: every decision made in this conversation, with the reason behind each one clearly stated.
Prompt to use:
Before building anything, create one organized decisions document from this conversation. Include every decision made, every improvement recommended, every issue resolved, and the reason behind each one. I want this to be clear enough that another AI model can use it to update the original onboarding system without losing any context.
Save that document.
You will use it in the final step.
Step 7: Open one final chat with Sonnet
Now take the original four onboarding documents from Sonnet and the decisions document from Opus.
Open one final chat with Sonnet.
Tell it you built the first version of the onboarding system, but a deeper review surfaced issues that needed resolving.
Then provide the original onboarding documents and the decisions document.
Ask Sonnet to produce a comprehensive, updated version.
Prompt to use:
I built the first version of our client onboarding system with the help of an expert AI conversation. After reviewing it more carefully, I used a second expert conversation to identify gaps, resolve issues, and make better decisions. I am giving you the original onboarding documents and the decisions document from that second conversation. Please take everything into account and produce a comprehensive, updated version of the client onboarding system. Keep it practical, clear, and aligned for [your vertical] serving [target market].
Then let it work.
What You Walk Away With
🎯 Set aside a couple of hours to review the documents it produces.
You will find a thing or two to adjust. And I promise, that is about all it will be.
What you will have at the end is a complete, significantly improved version of something that may have been sitting on the back burner for years.
Not a rough draft.
Not a framework that still needs to be populated.
A system. Done.
A process I had been meaning to build for years. Finished in one afternoon. And I was proud of every page of it.
For us, it was client onboarding.
The thing that reflects everything I believe about what a client's investment deserves. The thing I could not hand off because it needed to carry my values, not just my logo.
But this framework is not really about onboarding.
It is about anything in your business that lives in someone's head, or in a folder nobody has touched in two years, or in the gap between how things are supposed to work and how they actually work.
Any system that needs building, start here.
Any procedure that exists as a collection of tribal knowledge and good intentions, this is how you give it a structure worth using.
The tool was always capable.
You just needed to know how to talk to it.
And now you do.
Running an interior design studio, an architecture firm, or another project based business and ready to put this to work? We help teams build the systems their businesses actually run on. Let's talk.