• Speed raises the bar on clarity and judgment simultaneously — you cannot go faster than both of those can keep up
• A framework for deciding what not to automate: Rohan's perception data stays human while the interpretation layer around it runs roughly 80 percent automated
• How to tell "it looks good" from "it is good" when AI produces polished output faster than you can evaluate it
• The instinct to automate everything destroys the one asset that does not commoditize — the judgment and proprietary data only your firm holds
• Every firm is feeling the speed; few founders have noticed it is quietly outrunning their clarity
• Founders who define what "good" looks like before the work speeds up catch problems while they are still cheap to fix
• When things speed up, two things have to rise at the same rate: an abundance of clarity and much better judgment

Jeff Klaumann: Hey everybody, welcome to the Pro Serv Podcast, brought to you by Collective 54. I’m Jeff Klaumann, I’m the president of Collective 54, and I’m your host. If you’re new to the show, here’s what we’re all about — helping you do three things: make more money, make scaling easier, and make an exit achievable. Everything we record here is built exclusively for boutique professional services firms, so if you’re in the expertise business, if you market, sell, and deliver expertise, this show is for you.

We have a great topic today, and it’s not the AI story we usually tell. We often assume AI made everybody faster, so the founder’s job got a little easier. That is not what happened. One of us gets faster, and that is exactly where the strain starts to show up. My guest today put a rule on it: when things speed up, two things have to rise at the same rate, or you cannot actually go faster — a lot of clarity, and very good judgment. He also says he was the biggest fan of Fuzzy, and at about 30 people, Fuzzy stopped working. That’s the show today.

To get into this, I have Rohan Paul, who is the president of Team Analytics. Team Analytics works with leadership teams at private equity firms and their portfolio companies. What they measure is perception data — what the people on the management team actually think of each other. Rohan has spent more than 25 years in leadership teams, and he’s a great Collective 54 member. Rohan, welcome to the show, great to have you here. I’ve given everybody the headlines, so in about a minute, take a moment to describe what you do and the problem you solve for your clients.

Rohan Paul: Yeah, thanks, Jeff, great to be here. What we really do is help leadership teams that are under pressure execute faster. If you think about a leadership team, you could have new ownership, new leaders on the team, or even a new CEO, a new strategy, a new direction, a massive transformation going on — and in private equity, usually it’s all of the above — with some very clear, time-bound milestones that are expected to be hit. Typically, PE firms will spend a lot of time and money hiring the right people, the best people they can, putting them all together, and then the team component is just sort of like, okay, work together and go make this happen. And they only get involved when the wheels start coming off.

I like to remind all of our PE clients that this environment is not normal. You don’t just put people together — some of whom know each other or whatever — in this much pressure, and expect them to just execute. So we are really that execution support system that comes around it.

Jeff Klaumann: Fantastic, can truly see the need. So, Rohan, you put it that you can build things over a weekend that used to take weeks of conversation. Take us to that moment. What could you suddenly do, and what started breaking right after that?

Rohan Paul: Well, I would say “build” is a generous term. In the old world, I would have thoughts and ideas that I really needed a whole bunch of people’s input and help to actually implement and execute. Whereas now, with Codex or Claude, just like everybody, I can vibe code something, I can build a working prototype, I can change a process — pretty much by myself.

I was sitting down with Derek, who I run the business with, and we were just acknowledging the fact that for the last 15 years, he was my primary thought partner. We would work on ideas and concepts together, bounce them off each other. And now, if I talk to him today and then talk to him again tomorrow, I’ve been bouncing a million ideas off Claude and Codex — and not just bouncing those ideas off, maybe even taking them pretty far down the road. So the dynamic of what we’re able to do and produce has completely changed, as I’m sure you have experienced, and so have many of our listeners.

I think I heard Balaji Srinivasan say it first on an a16z podcast — that as AI gets smarter and smarter, the need for verification rises just as quickly. The need to really look at things critically and go, okay, this looks great, it’s a pretty website, it’s well designed — but is it actually just AI slop with lipstick on a pig? Or is it actually something worth saying? I’ve started to see it not only in the things I can build, but I’m recognizing it in the emails I get, I’m seeing it on LinkedIn posts, everywhere you look. The more time I spend with Claude or Codex, the easier it is to start to see it everywhere. It’s definitely a problem I’m wrestling with, and something I see all around me.

Jeff Klaumann: Yeah, I couldn’t agree more. It’s almost like a fingerprint now — you can kind of see it. You can almost tell by the language that’s used, or the output, you can tell which one was providing it.

So, you have called yourself the biggest fan of Fuzzy, and then at around 30 people, that stopped working. Tell us what you mean by Fuzzy, and what gets written down or said now before the work starts that never used to.

Rohan Paul: Yeah, it’s not even that it breaks at 30 people or whatever. Maybe here’s how I’d put it. In our business, we work with highly capable people, so a lot of my teammates are very experienced, they exercise a lot of judgment, and they’ve got a lot of freedom to essentially use their own discernment in how they work with clients. And I’m great with that — in fact, I sort of like not being put in a box. I like having highly empowered people with a lot of room and latitude to operate.

What I started noticing, as I began to work with AI and to really see AI becoming an asset, was that — because it makes everything so much faster — it can go off course really quickly. The need to create the right guardrails, framework, and structure so that the AI could operate inside of that and not just go off the rails became really important. And so it sort of forced me to really go down and challenge a lot of the assumptions that were either fuzzy or unclear — even fundamentally around who we are, what we do, and why we do it. The method, the left and right pillars — not so much because the team needed that clarity, but because if we didn’t have that in place and we just sort of let these agents run, they will run very fast and go down all sorts of directions. As you framed it: the faster you want to go, the more clarity you have to have. The more you need to make sure that you’re compounding people’s judgment and not replacing it.

Those two concepts became very clear to me and took me on this very introspective journey. What do I believe? And if I rewind it, it really started over the Christmas holidays of ’25 — going, oh my goodness, these things are so smart and can do so much of what I can do, so much better than me, that I need to start really retooling my own identity. What am I actually good at? What value am I actually adding here? And those two parallel tracks landed on this idea of judgment, of intent, of vision, and clarity.

That’s where I started seeing the need for clarity, whereas in the past I would resist it and say, let’s just trust everybody’s judgment. And it’s not that I don’t trust everyone’s judgment — it’s more that I don’t trust AI to have good judgment.

Jeff Klaumann: Well, I couldn’t agree more, because AI is the greatest producer of words per minute of all time. If you’re fuzzy on your request and you keep going down with whatever agent you’re using, before you know it you can’t pull it back out if you don’t have those guardrails — the left and right pillars. Great guidance.

Your firm’s asset is perception data — what people on the management team really think of each other — and you’ve drawn a hard line around that. Where is that line, and how did you decide what goes there and when something doesn’t?

Rohan Paul: Great question. One of the things we started to do in December of last year, and especially through January, was really look at how we can protect what’s truly valuable and not get commoditized by AI like everybody else. We have a huge amount of proprietary data based on the perception data that our advisors use with clients. Because our services are at the higher end of the consulting range and we work with very large companies and large private equity firms, the premium is really around the judgment of our advisors.

We want to be really careful that we’re never replacing the judgment of those advisors — we’re augmenting, we’re compounding, we’re focusing and channeling that judgment. So when we’re looking at and analyzing the data, AI can really accelerate our ability to focus on the right things, but the final judgment — our goal is almost to get 80% of the work done with AI, and then force that 20% to be human, because that’s where all the value is. If one of our advisors could spend a month doing something, AI can do a lot of that in minutes. But that final piece that really requires human judgment has to be human.

That’s the way we’re thinking about it — not just the perception data, but really anything. Emails that go out, notes that get captured, anything we’re doing that touches a client. If it’s completely AI, we just tell them, hey, this is an AI summary or an AI-generated piece of text. But if it’s our work, I don’t want it to be 100% AI, because I think the value really lies in that space where there isn’t a right or wrong answer — there is a level of discernment and judgment, and a level of humanity that, especially for what we do, is super important.

Jeff Klaumann: I couldn’t agree more. The real superpower of the human is the judgment side of things. The real superpower of AI is the analysis. If you can put as much of the analysis on that 80% and then allow the human to do the judgment, you’re really putting both in their superpower.

Rohan Paul: Yeah, and the AI can push your judgment as well. It can surface things you may miss, or that you just don’t have enough time to go uncover every little area. That’s been another interesting thing we’ve been pushing into — how do I push your judgment as an advisor even further? Push back a little bit, get you to stop and think: hey, wait a second, you said this, but what about this? What about this?

The other thing I think about is that in our world especially, a lot of it is influence. We are trying to influence people to make a change. I like to use the analogy that even if you handed me the most customized diet matched to my DNA, perfectly aligned to my goals — if you give me that piece of paper with exactly what I need to eat and exactly the exercise I need to do — I could still drive down the road and eat some fried chicken for dinner and skip my workout tomorrow. In the world we’re in, it’s really about building trust and credibility and having the influence to help people make the right decision, sometimes the difficult decision, and sometimes change patterns and do things that are uncomfortable. That’s where we want to make sure we don’t assume that a better-generated report or website is going to take the place of that human-to-human interaction and influence.

Jeff Klaumann: Exactly. One of the things that you’ve identified, that I think we all have, is that AI hands you work that looks finished — it’s clean, beautifully formatted, incredibly confident in its work. How do you tell whether it looks good or it actually is good?

Rohan Paul: Yeah, that is so funny — and it’s a conversation Derek and I were having recently. I was saying to him, hey, I want you to read that thing I sent you, and he goes, yeah, it looked great. I go, I wanted you to actually read through it, and if you start digging into it — and he goes, oh, I didn’t realize you wanted me to actually dive in and take a look. It looked great on the surface and it looked finished. He and I will often also joke about how, especially when working with AI, you can produce something and read it and go, wow, this is great. And then you come back the next day and read it again and you’re like, what the heck is this?

I find myself having that experience more and more — if I just give it a break and come back and read something, I go, wait a second, that sounds like AI slop. But it looked so good in the moment.

One of the things that has been so cathartic for me with AI is that it has helped me articulate what I’m trying to say so well. When you’re trying to say something and then it appears captured in text for you, and you read it and go, yes, that’s what I was trying to say — that feels so good. But then you don’t realize: wait a second, if I actually came back and read it, even though it really resonated with me, that’s not how I talk. That’s not a phrase any human being would ever use. So I think we’ve got to separate those two things — okay, it captured the direction, the substance, the essence of what I was trying to say. But now, how would I actually put this in my own words?

And I think the other element is — just because you make it look pretty and formatted doesn’t mean it’s worth reading, and just because you can say it in 10 pages doesn’t mean it’s not a paragraph. I’ve also noticed that AI will default to these wordy structures: the eyebrow heading with the heading, then a subheading, then an italics intro, then the text, then the bottom paragraph. It’s like, oh my goodness, we’re trying to say one word. That’s the other thing I’ve been really pushing myself and the team on — let’s not default to these wordy, beautiful prose, and let’s just get really clear and concise, and distill down to the essence of what we’re trying to say.

Jeff Klaumann: 100%. I can’t remember if it was Mark Twain who said, if I had more time I would have written you a shorter letter.

Rohan Paul: Yeah, exactly.

Jeff Klaumann: Essentially, I think we need to subscribe AI to that. So, one of the things you’ve done is build an agent that interprets your perception data with business context, and you’ve really designed it around that 80-20 rule we talked about earlier. Tell us what happens when the human overrides it.

Rohan Paul: Yeah. We see every opportunity — when the agent does what we call a calibration of the data, the human reviews that and makes some changes, and essentially explains why they made those changes. Then we’re able to do a training run and debrief that. Usually after 10 or 15 of those, we debrief it with one of our chief learning officers who can essentially train the agent to understand where its judgment could get a little better. And we’re sort of at the point now where they’re honestly judgment calls — it’s not that it made a mistake, there was no missing piece of information, it was literally a judgment call. So we continue pushing and training it, both on the calibration side and on the report writing side — when we generate a recap email after a debrief call, for example.

We’re capturing the human-edited output, and then an agent runs the diff between what it developed and the actual output, and tries to find patterns and ways it can get better. So it’s essentially providing feedback. And again, I’m nervous of that. I’m nervous of getting to the point where the agent actually writes the email to the point that we think it looks great, because I just think we’re not trying hard enough. I want us to have that human touch at the end that’s going to push it a little bit further. Maybe that thinking will get outdated with GPT-10 or whatever — I don’t know — but we’ll see.

Jeff Klaumann: Yeah, exactly. So, one more and we’ll wrap up. If a founder did one thing this week to keep their own speed from breaking things in the firm, what should it be?

Rohan Paul: For me, it really comes back to that idea of clarity and alignment with the rest of the team. I read on a blog that when electricity was first invented, everyone thought it was going to make factory output go through the roof right away. And apparently, it took 30 years for factory output to actually go through the roof, because you needed power transmission, you needed to redo the supply chains, you needed to reorder the way work was done inside these organizations. And so for me, it was realizing that I need to have the clarity, I need the rest of the team to speak into what I’m not seeing, and we collectively need to be able to get on the same page faster. Then the systems and the structure and all of the complex stuff that needs to happen around it can happen. But as speed rises — come back to where we started — clarity and judgment have got to be matching, or maybe even going a little bit faster. And that’s what helps keep it together.

Jeff Klaumann: Fantastic. What I’m taking from our conversation is there is a natural tension as you navigate becoming an AI-native firm, and I’m sure our listeners can relate to it. Rohan, thanks for being here today. I am glad to have you on the show.

Rohan Paul: Thanks, Jeff, absolute pleasure.

Jeff Klaumann: Members, Rohan and Derek will be joining us for our private member Q&A session, where you can ask them your questions directly. And two calls to action for you before we wrap. First, if you’re not a member and you want to keep the conversation going, head over to Collective54.com and start a conversation with our AI agent. Second, our founder Greg Alexander has a new book coming soon — it’s called The AI Native Boutique Firm: How Founders Build More Valuable Firms When Services Become Software. It’s now available for pre-order on Amazon, so go check it out. Thanks for listening. Until next time, I wish you the best of luck as you grow, scale, and someday exit your firm.

‍