How to Teach AI Your Business Processes, Starting With the Hidden Ones
Your most valuable workflows run on instinct and live in no manual. Here's how to find them, teach AI your reasoning, and keep the judgment that clients pay for.
The short answer: The most valuable processes in a business are usually the ones the owner runs on instinct and has never written down. To get real value from AI, find those hidden workflows, explain your reasoning out loud until AI can follow it, and keep the final review for yourself.
In my business, the moment looks like this. A home inspection report comes back on a property my clients are buying.
Before I've finished the summary page, I've already started deciding. Which items are real and which are noise. How my buyers will hear the word defect. What the listing agent will accept, what the seller will resent, and which repairs need a bid before we ask for anything.
Ask an experienced agent what they do when a report lands and you'll get one sentence: I review it with my clients and we negotiate.
That sentence hides about 30 decisions. Those decisions are most of what the client is paying for.
Every business has its own version of that moment. A contractor gets a change request mid-project. An accountant opens a client's disorganized year-end file. A practice owner reads a frustrated customer's email. Something arrives, and an experienced person starts a chain of decisions they would struggle to write down.
That chain is where AI can help most, and it's usually the part AI never sees.
The Hidden Workflow Method
The Hidden Workflow Method: find a process you run on instinct but have never written down, teach AI the reasoning behind it out loud, and run it 10-80-10: you spend 10% of the effort setting the goal, AI does 80% of the work, and you spend the last 10% reviewing and adjusting before anything goes out.
Why your most valuable processes are the hardest to see
Experience compresses into reflex. The more often you've done something, the less of it you can see.
A new hire feels an inspection response as a long, nervous checklist. A seasoned agent feels a single motion: a trigger, then a chain of decisions that runs on its own.
So the workflows that produce your best outcomes are usually the least documented, because you don't experience them as workflows at all.
That has three costs. You can't delegate what you can't describe. You can't train a team member on it. And you can't show a client what they're paying for.
So when owners ask me how to use AI, I start with the decisions they make without noticing. Tools come later.
How to spot hidden workflows in any business
You won't find them by asking what are my processes? You'll list the obvious ones. These signals work better:
A document or message arrives and you start deciding. An inspection report, a contract redline, a complaint, a change order. The trigger is visible. The decisions after it usually aren't.
People ask you how did you know? If a colleague can't follow how you got to an answer, there are steps they can't see.
Work comes back from a team member and you redo part of it. The part you redo is the part you never explained.
It recurs and only you do it. Weekly, monthly, every time a certain kind of client shows up. Recurring work that only the owner can do is a ceiling on the business.
Not every hidden workflow is a negotiation. One of mine is a monthly job: I pull production reports for the top 90 or so agents in our office, out of about 320, and send each of them a personal message with their numbers. I taught AI the process. It now pulls the reports, keeps track of who has been sent theirs, and puts each message in my drafts for me to review and send. My part is now the review.
A wider net helps too. Take a normal month and write down everything you actually spend time on. Then ask one question about each item: am I investing this time or just spending it? The items you resent, and the ones only you can do, are where to look first.
How to teach AI your reasoning, not just your instructions
I have a reusable AI skill for inspection negotiations here in Utah. I didn't write it, and I didn't engineer a prompt.
I turned on my voice and talked through what actually happens in my head when a report comes back. I'm managing my buyers' emotions so they don't overreact. I'm negotiating with the listing agent and, through them, the seller. I need realistic repair costs so my clients understand their real exposure. Then I need a strategy for the ask.
AI turned that into a skill. Now I upload the report and it sorts the findings into what's worth negotiating (health, safety, and big-ticket items) and what not to ask for, because asking for everything weakens the requests that matter. It researches repair ranges, cites the page numbers, and drafts the emails: one to my clients and one to the listing agent.
On one report, it suggested asking for 7,500 to land at 5,000, so the sellers had room to come down and still feel they'd met us in the middle. That was its read on that deal, not a rule.
The part that impressed me most was the framing for the listing agent. It explained that my buyers started with a longer list and deliberately narrowed it to health and safety issues and material repair costs. That's very hard to push back on, and it's exactly how I'd want it positioned.
Saying your reasoning out loud is the real lesson. It turns instinct into steps, and steps can be checked, improved, taught to a new hire, and handed to AI.
One more thing: ask, don't just tell. Tell AI how to build something and it builds exactly that. Ask it what the workflow should include, or have it interview you, and it often surfaces steps you skipped.
What to capture before you hand a decision to AI
When you talk a workflow through, make sure these pieces come out. They apply to any business.
- The trigger. What arrives that starts the work.
- The goal. What a good outcome looks like for each person involved, not just the task.
- The people and what each one needs. In my example, buyers need perspective, and the other side needs a request that's easy to accept.
- Where the information lives. The documents, systems, and history AI should look at. In my experience, AI can go much further once you've taught it your thinking and told it where to find things.
- Your sorting rules. What matters, what doesn't, and what you deliberately leave out.
- How it should sound. The framing you'd use yourself.
- What it must not touch. Decide what never goes in. I don't put things like bank account or Social Security numbers into AI.
- What you'll check before anything goes out. Decide in advance what you'll review.
If you can't say one of these out loud, you've found the step you'd struggle to delegate to anyone. That's worth knowing even if you never use AI.
The 10-80-10 rule, and where it breaks
I run these skills on a rule Dan Martell calls 10-80-10 in his book Buy Back Your Time. I spend 10% of the effort stating the goal clearly, let AI handle 80% of the execution, and spend the last 10% adjusting and auditing.
"Rarely do I just hit send." Editing beats starting from scratch, and the model moves my attention to the end, where judgment matters most.
But I want to be honest about its limits, because a tidy ratio can make it sound safer than it is.
The audit is only as good as the auditor. On repair costs, my experience is that the skill gets most of the way there. I can tell because we own rental properties, so I know roughly what it costs to replace an AC unit. Someone without that background will read the same numbers and nod. In that case, 10-80-10 doesn't remove risk. It produces confident drafts faster.
The first 10% is the hard part. Stating the goal sounds simple. It isn't, if you've never put your own reasoning into words. When the output disappoints, the cause is usually a vague destination, not a bad tool.
The ratio describes effort, not risk. On anything that touches a client's money or a negotiation, the audit should be the slowest part of the job, whatever the percentage says.
AI doesn't see everything you see. My skill tailors the client email because it knows how each of my clients prefers to communicate. It has no such read on the listing agent.
What your expertise is worth when customers use AI too
Customers in every industry will increasingly arrive with their AI research done, and some will wonder what they need you for. The honest answer starts with an admission I made to a room of agents: "if a client can do the same thing with ChatGPT, that part of our industry is no longer of value."
Swap in your own industry and the sentence still holds. In mine, finding homes is already there. And I think AI will be better than us at parts of negotiation within three years. I use it on almost every negotiation now.
So what's left? The hidden workflows. Knowing which repair to leave off so the rest of the request holds. Knowing that two streets over is a different market, for reasons AI tools don't pick up. Deciding what a nervous client needs to hear first. Standing behind the call.
Make those workflows visible and two things happen. AI can carry the execution, which buys back your time, and you can show customers where your judgment lives.
The owners who struggle in the next few years will be the ones who never figured out which part of their work was the valuable part.
Where to start this week
Pick one trigger, ideally where your business is most constrained right now. In my business that might be an inspection report, a price-reduction conversation, or a new listing appointment. In yours it's whatever arrives and sets off decisions only you know how to make.
Talk through everything you do when it happens, out loud, as if you were training your best new hire. Ask AI to interview you until the process is complete. Save it as a reusable skill. Then audit the first few outputs harder than feels necessary.
You'll end up with a faster process and a clearer picture of what you're actually good at.
FAQ: teaching AI your business processes
How do I teach AI my business processes? Talk through what you actually do when the work arrives, out loud, including how you decide, not just the steps. Have AI turn that into a reusable skill or set of instructions, ask it what's missing, and audit the first several outputs closely.
What is the 10-80-10 rule for AI? It's a rule from Dan Martell's book Buy Back Your Time, and it's how I split the effort: about 10% stating the goal clearly, about 80% AI doing the execution, and about 10% for me adjusting and auditing before anything goes out. It describes effort, not risk. On anything involving money or a negotiation, the final review deserves the most care.
What should I capture before handing a decision to AI? The trigger, the goal, the people involved and what each needs, where the information lives, your rules for what matters and what to leave out, how it should sound, what should never go in (things like bank account or Social Security numbers), and what you'll check before anything goes out.
Is my expertise still worth paying for if customers use AI? The parts a customer can do themselves with AI won't be worth much for long. What stays valuable is judgment: knowing what to leave out, reading the people involved, local knowledge AI tools don't pick up, and owning the decision. Making that judgment visible is part of the job now.
Get the next one. In the newsletter I break down workflows I've handed to AI, and where my judgment still sits in them. Join the newsletter
About me: I'm Shoney Ivens, CEO of a brokerage that's produced over $5.4B in sold real estate volume in the last 5 years.
Want help finding the workflows worth handing to AI? That's the work inside The Leverage Collective, where we install these systems in 90 days for established agents, team leaders and business owners. See if it's a fit
