A commercial real estate team in Austin — 52 people, established firm, serious operation — told me last week that their quarterly investor reporting process takes hours every cycle.
Pull data from Yardi. Reformat it in Excel. Write the narrative sections in PowerPoint. Add property maps in Canva. Merge everything together. Review the whole thing. Send.
Six steps. Three tools. Same process every quarter.
One of their team members said something that crystallized the problem for me: “If someone could teach us the method to automate this, it would be transformational.”
He wasn't wrong. But the bigger insight isn't about that specific workflow. It's about the mental model most people are missing when they think about using AI.
The Wrong Way to Use AI
Most people use AI as a helper. They write a draft, then ask AI to improve it. They do research, then ask AI to summarize what they found. They build a report, then ask AI to check it.
This keeps you in the seat for 100% of the work. AI is an occasional assistant. You're still the one doing the job.
That's not the worst thing — AI as a helper is genuinely useful. But it's not the most valuable way to use the technology. You're working the same way you always did, just with a smarter spell-checker.
The 10/80/10 Rule
When I work with clients on building AI into their workflows, I use a framework I call the 10/80/10 rule.
The first 10% is yours. This is where you set the direction. You define what a good output looks like. You give the AI the template, the data sources, the narrative guidelines, the context about who the audience is and what they need to see. You decide what “done” means.
This part can't be automated. It requires judgment, domain expertise, and taste — things that are uniquely yours at this point in AI's development.
The middle 80% is AI's. This is the execution layer: the data pulling, the formatting, the assembling, the first drafting, the populating of templates, the checking of sources. It's repetitive. It's time-consuming. It doesn't require creative judgment — just careful, consistent execution of a defined process.
AI does this better than humans at scale. Faster, cheaper, and without the cognitive cost.
The last 10% is yours again. This is where you review the output, apply aesthetic judgment, catch anything that's off, adjust tone, and give it the final polish before it goes out the door.
This is also the part where AI currently falls short. The technical term I use with clients is taste. AI can build the skeleton of a PowerPoint. It can draft a narrative. But it can't reliably tell you whether something looks polished, whether the language feels right for a specific relationship, or whether the framing is exactly on. That's still a human call.
What This Looks Like in Practice
For the real estate team, the 10/80/10 breakdown for their investor reports would look something like this.
The first 10%: a team member defines the report structure, sets up the template, and tells the AI what data sources to pull from (Yardi, the Excel files, specific property records). They specify what kinds of language investors expect and what the tone should be for each property type.
The middle 80%: the AI pulls the financial data, populates the template sections, drafts the narrative for each property, adds relevant details from the source files, and assembles the full document.
The last 10%: the team member reviews the draft, adjusts any numbers that need double-checking, tightens the language in a few places, and handles the final layout before it goes to investors.
The total human time on a report that used to take several hours becomes maybe 30-40 minutes of actual judgment work, bookending an AI-executed middle.
I've seen a similar pattern in a medical office building brokerage I worked with. Before each sales call, their team used to spend time researching comparable sales data for the specific property. Pull the comps manually, format them, review them. Now the system pre-pulls and formats all of that before the call is even scheduled. The salesperson reviews it in two minutes. Then they go into the conversation already prepared.
Same output. A fraction of the human time.
The Part That Stays Yours
One thing I want to be clear about: the last 10% is not just a formality.
There's a version of this model where people assume AI will get to 100% eventually, and the review phase will disappear. Maybe — but not yet, and maybe not for the things that matter most.
The places where AI still struggles are judgment calls that require domain expertise and taste. Knowing that a particular investor prefers conservative language over optimistic projections. Knowing that a specific word choice will read as too formal for a longtime relationship. Sensing when a document's structure is technically correct but doesn't quite flow.
These are things you know because of who you are and what you've seen. They're not easily transferable to a model.
The good news: that last 10% is often the most valuable part of what you do. The AI handles the 80% of work that, honestly, you could train almost anyone to do if you had time to train them. What you keep is the piece that actually requires you.
The Setup Upfront
The honest part of this model: getting to 20% of the work requires real investment upfront.
You have to map the process clearly enough that AI can follow it. You have to build the templates. You have to connect the data sources. You have to write the guidelines that define what a good output looks like.
This is what I call the Automation Spectrum — the process of moving work from manual execution to AI execution requires documenting the repeatable backstage layers first. You can't automate what you haven't defined.
That documentation phase takes time. But most people already know their process well enough to describe it — they just haven't written it down.
The output of that work: a workflow where you make the call at the start and review the result at the end. AI handles everything in the middle.
Start With Your Bottleneck
The simplest way to begin: think about your most repetitive work output. The report you assemble every week. The briefing document that requires pulling from three different sources. The proposal template you customize for each client.
Pick one. Map the steps. Identify which of them require judgment and which are just assembly.
That middle layer — the assembly, the formatting, the first draft, the data consolidation — is your 80%.
Start there.
The 4-Day AI Sprint covers how to build the kind of AI workflows that handle the middle 80% — from mapping your process to connecting your tools to setting up the review layer that keeps you in control.
