Building the First Tool in an AI Portfolio Management System
A hands-on exercise in translating one step of quarterly investor reporting into a repeatable technical workflow.
1. Why I did it
I lead a company that combines institutional real estate and AI. I wanted to understand, firsthand, what actually happens between identifying a workflow and building a tool that holds up—not a demo, but something built from a real piece of institutional work translated into technical instructions.
This was my version of “Hello World,” pulled from an actual portfolio-management task rather than a tutorial.
2. Start with the function, not the technology
We’d already mapped the portfolio-management function—the recurring work, what it runs on, the judgment calls it involves, and what investment committees, senior management, and limited partners actually expect to see. From that map I picked one recurring workflow: the quarterly investor report.
I didn’t go looking for something to automate. I started from a process I already understood, and only brought in the technology once the function, the workflow, and the expected output were clear.
3. Break the workflow into steps
A quarterly report isn’t a single task—it’s a chain of tasks and subtasks, each with its own inputs, controls, and stakes:
- Gather the current-quarter materials
- Identify and categorize the files
- Extract the relevant information
- Validate it against other sources
- Investigate inconsistencies
- Determine what is material
- Populate the report
- Draft commentary
- Review and approve the output
I picked just the first step: open every file and figure out what it is. That narrowness was deliberate—before a system can extract, reconcile, or communicate anything, it has to know what evidence it’s actually looking at.
4. Build the first tool
The tool reviews the current-quarter folder and categorizes its contents. Those contents may include:
Even this simple-looking step involves real judgment. File names aren’t consistent, and formats vary across property managers, operating partners, assets, and reporting periods—a spreadsheet might have half a dozen tabs, a PDF might be scanned, an email screenshot might carry a real update with no useful metadata attached to it at all.
So the system can’t rely on filenames—it has to look inside the file and recognize what a document actually looks like. A portfolio manager can spot a rent roll almost on sight. Making that recognition repeatable means spelling out what it’s actually looking for: tenant or unit identifiers, lease dates, square footage, rental amounts, and how those fields relate to each other.
5. Explain where it fits
The file-classification tool is one small component of a larger system:
This first tool isn’t producing the quarterly report itself. It’s setting up an organized, traceable input layer that later tools can build on—the layer that ties every extracted fact, and eventually every conclusion, back to a specific source.
6. What I learned
The exercise made several principles of institutional AI implementation more concrete:
The hardest part is defining the work precisely. Coding doesn’t start until the objective, the evidence, the classifications, the edge cases, and what counts as an acceptable output are all spelled out.
A workflow that looks obvious is full of implicit decisions. Experienced portfolio managers resolve inconsistencies without even noticing; a system needs those judgment calls pulled out into the open.
File classification establishes provenance—it’s not paperwork, it’s the foundation. Every extraction, variance, and written conclusion downstream should trace back to an identifiable source.
Reliable automation depends on narrow, testable components. A bounded tool can be checked, corrected, and supervised in a way an open-ended request never can.
Domain experts and engineers need a shared way of describing the process. Workflow maps, classification rules, expected fields, and exception paths are what give them a common language.
Executive leaders need real technical fluency. Not enough to build every tool themselves, but enough to tell a demo apart from something that can actually run as an institutional operating system.
If there’s one lesson that keeps showing up, it’s that reliability comes less from the model and more from what’s built around it: narrow scope, explicit rules, traceable evidence, visible uncertainty, human approval.
7. What comes next
- The next tool extracts defined information from each file.
- Later components validate and reconcile the extracted information across conflicting sources.
- Approved data can eventually populate defined fields within the quarterly report.
- The portfolio manager remains responsible for interpretation, materiality, and final communication.
The system gets built from the function outward, one narrow and accountable piece at a time.