What Automating a Due Diligence Review Taught Me About Commercial Real Estate

A screener can only be as good as the rules it is given. The first job was writing the fund's criteria in plain words, with a version number, so that when the rules change you can tell which version judged which deal.

Every fund says it has criteria. Ask to see them written down and you usually get a mix of a pitch deck slide, a few emails and the managing partner's instinct. A person can work from that. A screening system cannot, and neither can a new analyst.

So the first real deliverable was a document: the fund's criteria in plain words, each version dated, so a score on any deal could be traced back to the exact rules that produced it. The criteria belong to the fund. How the screener is graded against them is the part I bring.

An accuracy claim means nothing until it meets a real decision. Start with a deal the team signed and ask whether the screening step would have flagged it. Then keep grading, one criteria version at a time.

Here is the moment that changed how I work. In July, the screening step had scored a deal at zero. It was the one deal the team went on to sign. The system had been running, producing scores, looking busy, and it would have buried the one deal the team went on to sign.

Nothing was broken in the usual sense. The criteria it was given simply did not describe what the team valued in that deal. The only way to catch that is to grade the screener against real decisions, starting with every signed agreement, and to keep grading each time the criteria change. A score you have never checked against an outcome is a guess with a number on it.

Documents scattered across inboxes are why the next deal starts from zero. Originals get a second home in storage the company owns, so the file outlives any one person's mailbox.

Due diligence produces paper: offering memoranda, rent rolls, operating statements, loan documents. In most small firms those files live in whoever's inbox they landed in. That works until someone leaves, or until the team needs to compare this deal with one from last spring.

The fix is boring and does more than any model: originals get a second home in storage the company owns, named consistently, linked to the deal they belong to. Everything else in the system reads from there.

Say it before the build, not after an incident: who can query the data, who holds the credentials, who keeps the backups and who fixes a change someone else made.

The part of the proposal I rewrote the most was the clause on data, access and responsibility. Anyone the company admits can query the database directly. Access, credentials, backups and changes made by other people are the company's to manage, and correcting someone else's change is separate work.

That sounds cold on paper. In practice it is what lets an owner trust the system, because nobody is guessing later about who should have kept the backup. The same thinking sits behind a line I now put in every agreement: the client owns its data, criteria and outcomes, and I keep my methods.

Dallas wanted every deliverable explained in detail before the next step, and he was right to. The first worry on the table was monthly cost, before accuracy even came up. So the work was split into fixed modules with the hours shown on each, a clear list of what done looks like, and acceptance a set number of business days after handover.

The lesson for any firm hiring help: if a consultant cannot tell you what each piece costs to run and how you will know it is finished, the scope is not ready.

The investment decision. A screener sorts and flags; a person moves a deal forward or passes. Nothing the system produces is advice.

The investment decision. A screener sorts incoming deals and flags the ones worth reading first; a person moves a deal into the pipeline or passes. The system also does not give investment advice, and the agreement says so.

If the partners do not agree on the criteria yet, automate nothing and write the criteria first. If the firm sees a handful of deals a month, a good analyst and a shared folder will beat any system. And if nobody will own the documents, the screener will only move the mess faster.

Where I would start next time is where the next audit, at Carter Funds itself, starts: recorded conversations with the people who do the work today, and one real example of each document they touch.