Two out of eight faces appeared in a client Teams call. Mine and my partner’s.
The clients’ cameras were off and, apparently, how they run its meetings, as a habit.
I didn’t realize how much I have grown accustom to reading a room (even virtually) by visually reading people. Decades of training, selling and leading have trained me to watch facial and body expressions for the moment something lands or doesn’t. Here, I didn’t have such cues.
I was there from an email I had written around observations of their business with an invitation to meet. My partner and I agreed to have him send it under his name, as he had spent nearly 10 years earning the client’s trust and they had yet to meet me. The email secured the meeting.
Though I could not see them, I know they listened to me. However, they listed harder to my partner. Every time the conversation had a choice of direction, it went toward what he held the keys to: tools versus strategy, plans, analysis. The human side of the work, the part I’m there to lead, kept waiting its turn.
By the end of the call I’d made a decision nobody saw me make. The client organization was not a strong motivational fit at this stage for my expertise. My partner should maintain being the face of this relationship while I would do my work from the background.
For now.
The test
Later, I ran an experiment on myself
I used an AI project space as a kind of chief of staff (CoS) for my AI advisory consulting practice. I asked this CoS agent to go through my client work and find real judgment calls: moments where I took one path and a smart person in my shoes would probably have taken the different one.
It came back with five, ranked for this article. The email I described above was first listed.
AI had the facts right. I drafted the outreach, my partner sent it, the meeting happened, and the work that followed became paid. Then it added one line explaining why I’d done it: you chose reach and trust over owning the relationship.
I read that and nodded, because reach and trust was right. Thus, it was almost right.
The word that mattered
The miss was one word. Owning.
Ownership was never on the table. My partner’s firm holds that client relationship, and it always will. I am coming into it as a new leader of one discipline inside it—the human and cultural side of AI adoption and fluency. “Owning” tells a story about something being given up for me to enter the door. That was not the case.
What I actually decided was when to lead and when to let someone else lead. I borrowed ten years of their trust as scaffolding, so the client could start to see me in the work while my partner vouched what I’d contributed. Nothing was traded away. It was a sequence.
Then the test went further. The AI helping me write this piece turned that judgement call into a kind of decision record a company would keep, the note the next person reads when they step into the account. In essence, part of the client dossier for future consultants of my partner.
It mislabeled the decision as “relationship ownership,” with the same wrong word. Then it went further. It recorded my human-side work as “lower priority for this client” and, thus, would lead the next person entering the relationship to stay behind the scenes. It created a real risk of anyone reading that record of potentially skipping the human side of the work entirely. So very off, as my plan is quite the opposite. Yes, I am in the background now, but over time I'll bring the layers that decide whether AI actually sticks inside an organization: acceptance, culture, systems, the way work gets done.
The record turned not yet into an assumed never.
So, AI was right about almost everything. However, the part it got wrong was the part that mattered most.
The door people walk in through
Before that test, I’d have called this a knowledge problem. Now I believe it’s a judgment problem.
Documents can be backed up. Policies can be written down somewhere. What lives nowhere is the reasoning: why you said no to that deal, why you trust that supplier over the cheaper one, why you handled that client’s blow-up the way you did.
Undoubtedly, you have someone on your team everyone runs hard calls past. Ask them how they decide, and watch them struggle to say it. They aren’t hiding anything. Judgment is built from cues they stopped noticing years ago, similar to my experience when the cameras were off on the video call.
Most of what gets written about this treats it as an exit problem: capture the knowledge before it walks out the door. Exits matter. But they happen infrequently.
Instead, people enter all the time. A new leader takes on a discipline. A new hire joins the team. A partner gets brought into an account, the way I was. Anyone entering has to learn how this team actually decides, often by guessing, while the person who knows is absent or too busy to explain.
That guessing is where judgment gets lost, long before anyone walks out the door.
There is a number for how long the guessing takes.
In a survey of 1,000 US workers by Panopto, new hires got about two and a half months of formal training. Getting them fully up to speed took as long as six. The same survey found that 42 percent of what people knew in their role was unique to them and not shared with coworkers.
That gap of three and a half months is the walk-in problem with a clock on it.
The limits matter, so here they are. The survey dates from 2018, before today’s AI tools, and it focused on larger companies. It measured knowledge, not judgment. My read, from my own client work rather than from any study, is that the gap for judgment is wider, because judgment is the part nobody thinks to write down.
It’s also the part the new tools are being sold to fix. Plenty of software now promises to capture what your people know before they leave. My test says capture is the easy part. The question is “what happens to the why?”
The hardest part of walking in
You may have lived one side or the other of this equation.
The person who holds the trust should stay the face. Clients need continuity. Push a new leader forward too fast and you spend trust that, in this instance, took a decade to earn. And, I am not sure you can get it back once spent.
However, the new leader needs room to lead. Staying in the background too long and you risk never stepping forward. The trusted person ends up carrying everything, forever, which is its own kind of risk.
Each need is real. Neither is wrong.
That tension does not need to be resolved immediately. However, the plan does need to be laid out, otherwise others will not know the transitional intent.
What actually moves judgment
Getting my own plan out of my head is the test I tried to do. Looking back it moved through three steps, where each one behaved differently.
Capture is the easy part, and AI is genuinely good at it. My CoS project found five real decisions with the facts straight. If your team’s calls live in meeting notes, email threads, and/or recorded conversations, you have the raw materials that AI can use to surface your facts.
Codify is where the why get flattened. Here I had to decide on correcting a reusable file into carrying the right intent to my plan. Here the wrong word “owning” showed up when I meant “when to lead.” Where also “lower priority” was interpreted when I meant “not yet.” This is the step that needs the person who made the call to read the record and correct it.
Transfer happens through people. My partner vouching for my expertise opened the door on that first call. It didn’t hand over the trust. He will keep vouching and presenting part of my work while I earn my own trust with the client. Something time and shared experiences cannot shortcut.
Thus in the division of labor, AI does the first pass. The person who holds the judgment fixes the why. The people who hold the trust walk the new person in.
Three lines, not a manual
Here is the practice that came out of test. It only takes about five minutes.
For any judgment call worth passing on, I write three lines:
What I decided. Just one plain sentence. In this example, Let my partner stay the face of this client.
What I noticed that made me decide it. The cue, not the conclusion. Here is was cameras off, every question pulled toward tools, and who the room listened harder to whom held trust or power.
What would change my mind, or where I’m heading. This is the line every AI record I tested left out. My decision to stay in the background for now while regularly bringing in my experience on the human side as trust builds.
Try this week
Pick one recent decision on your team that someone new would have to make without you. Write the three lines. To make it faster, paste this into whatever AI assistant you use. It drafts the three lines, then makes you fix the why before it finishes.
I'm going to give you my notes about one decision I made at work. Help me turn them into a three-line note for the next person who has to make a call like this without me.
Step 1. Draft three lines from my notes:
1. What I decided (one plain sentence)
2. What I noticed that made me decide it (the cues, not the conclusion)
3. What would change my mind, or where I'm heading next
Step 2. Before you finalize, ask me these three questions, one at a time, and wait for my answer to each:
- Is there a cue I noticed that isn't in my notes?
- What would have to change for me to decide differently?
- Which word in your draft could mislead someone who wasn't there?
Step 3. Flag any word in the draft that sounds permanent, like "never," "always," or "low priority," and ask me whether I actually mean "not yet."
Then give me the final three lines.
Here are my notes:
[Paste your notes, emails, or a meeting summary about the decision]Then hand the note to the person most likely to inherit that call and ask them one question: what might you get wrong if this were all you had?
Their answer is your missing judgment.
I still think about those turned off cameras.
My partner is still the face of that relationship, and that’s right for now. When I write the three lines for this account, the last one will say not yet.
That line is the whole difference. A record that says never sends the next person in the wrong direction with total confidence. A record that says not yet tells them where you were going.
Who on your team is walking in right now, and what are they guessing about that you could write down in three lines?
Think of one person whose name came to mind while you were reading. Send this to them.
A note on how this was made. This piece is its own field test. I used AI to research, structure, and draft it from my answers to a long interview, and I corrected it where it flattened my why, exactly as described above. The judgment calls are mine, and so is what’s published.