What Is an Expert Judgment Platform?
Every company has a person like this. The machinist everyone asks. The consultant whose clients call at night. The attorney who knows which questions to ask before anyone drafts a word. They do not just know facts. They know what to do when the standard answer does not fit.
That is expert judgment, and it is the thing most software never touches. An expert judgment platform is software that captures how one specific expert thinks, then lets other people ask that expert's judgment by voice, whenever they need it, without interrupting the expert. Apex Replicant is the expert judgment platform we built. This post explains what that means, piece by piece.
What expert judgment is, and why it is not information
Information is what you can look up. Judgment is what an expert brings when the lookup does not fit: the question they always ask first, the pattern they recognize from years of watching what works, the moment they say "I need more context" instead of guessing.
A book holds information. So does a course, a wiki, or a shared drive. None of them can listen to a unique situation and ask the right clarifying question. In my own coaching work I have seen the best advisors refuse to answer a founder's question until they understood it, and that refusal was the most valuable thing in the room.
General-purpose AI is built to produce a plausible answer every time. An expert is built to know when they do not have one. A platform for judgment has to copy the second habit, not the first.
Why judgment walks out the door
When they leave or retire, the answers walk out the door.
This is not a rare event. Deloitte's 2025 research, which we cite on our continuity page, found that 83% of organizations call the loss of a departing expert's knowledge serious, while only about 7% consistently capture it. On the manufacturing side, The Manufacturing Institute reports that 1 in 4 U.S. manufacturing workers are 55 or older, and Peak 65 counts more than 11,000 Americans turning 65 every day.
Documentation does not close that gap. The standard operating procedure records the steps. It does not record why the experienced person skips one of them on a cold morning. That reasoning lives in a person, and until now the only way to reach it was to interrupt that person.
How capture works
Capture starts with one hour. We run a guided voice interview with your expert. Nothing to write, nothing to fill in. The expert talks, and the interview is designed to bring out the part of their practice that was never written down: their frameworks, their methods, the way they weigh a decision.
After the interview, we ingest the documents and recorded video you already have. For a plant that means SOPs, prints and handbooks. For a consultant it means their own material. The interview seeds the judgment and the voice. The documents add depth. Our knowledge-capture method is patent pending.
What comes out is a Digital Protégé: a voice-trained AI version of your expert that works only from what that expert approved. It is not a chatbot trained on the open internet. It answers by voice, and it only knows what it has been taught.
How a voice Digital Protégé answers
People talk to a Digital Protégé by voice, in a browser or on a phone. A session runs 15 to 60 minutes and has three parts: a light prep, the call, and a recap, which we cover below. You can see the full flow on our sessions page.
Each answer follows a small set of rules. It first works to understand the question, because context comes before any answer. Then it reaches for the expert's own validated knowledge first, in the expert's framing. And when it reaches the edge of what it knows, it hands off instead of guessing.
Escalation: what happens when it does not know
Escalation is the Digital Protégé handing a question to the human expert when the answer is not in what it was taught. This is the feature that makes the rest trustworthy.
When a Protégé escalates, the expert receives the question with the full session transcript. Over time those hand-offs become a map. They tell you precisely which knowledge exists in exactly one head, and where your documentation is thin. Then the expert answers once, and from that point on the Protégé knows. Every real conversation teaches it more.
I would rather a tool say "I don't have that, I've notified the expert" than give a confident wrong answer. In a machine shop, a confident wrong answer is scrap. In an advisory practice, it is a damaged relationship.
Where the post-session summary email fits
The judgment does not stop when the call ends. After every session, two people get something useful.
The person who asked gets a plain-English session summary, the insights and action items, and the recording with a searchable transcript, all in their client portal.
The expert gets a scored summary email: a session rating out of 10, a follow-up flag, an urgency level and a lead score, followed by sections on client satisfaction, how the Protégé performed, key topics, action items and recommendations. You read the summary, not the transcript, and a new lead arrives already scored. A weekly digest arrives on Mondays. You can see a sample on the for experts page.
Who it is for
There are two ways to use an expert judgment platform, and they share one engine.
For companies, the question is key-person risk: what breaks the week your best person is out. We offer the 30-Day Proof, a fixed-price pilot: one expert, one hour of their time, thirty days to find out. The Proof is $1,247 once, with no contract and no auto-renewal.
For experts who sell their time, the question is the ceiling on their calendar. You build a Digital Protégé, set your rate, and keep 70% of what clients pay, with a minimum rate of $60 an hour. Your client's first session is free. The details are on the for experts page, and you can browse available experts in the Expert Marketplace.
What to look for in an expert judgment platform
If you are comparing options, ask four questions.
- Does it capture how one person thinks, or does it only index their documents?
- Does it say when it does not know, and hand the question to a human?
- Does it only know what you taught it, and is your knowledge kept isolated and out of any public model?
- Does it show you what happened after each session, so the learning compounds?
If the answer to any of these is no, you are buying search, not judgment.
Start with one person
The way to understand this is to talk to one. Ask a Digital Protégé something only its expert would know, or book fifteen minutes with a founder from the Proof page. Start with the one person you cannot afford to lose.
Ask it something only the expert would know.
The fastest way to understand Apex Replicant is to have a conversation with one. It answers only from what its expert taught it. When it doesn’t know, it says so.
CEO and co-founder of Expert Scale Incorporated. Writes on how Apex Replicant, the expert judgment platform, captures what an expert knows and hands off what it does not.
More from Drew Harris→