In this episode of In-Ear Insights, the Trust Insights podcast, Katie and Chris discuss whether AI belongs on your org chart. You’ll discover why placing artificial intelligence on an organizational chart creates false accountability. You’ll learn how to separate data tasks from human judgment without sacrificing results. You’ll walk away with a clear framework for matching risk levels against your personal competence. You’ll secure a practical method to deploy assistant systems while keeping final decisions in human hands.
00:00 – Introduction
02:15 – The virtual executive trend explained
05:40 – Why AI belongs in your tech stack
09:10 – When machines outperform struggling humans
13:25 – The seven categories of AI use cases
18:05 – Human judgment versus machine probability
24:30 – Legal risk and real-world accountability
30:15 – How to set boundaries with assistant systems
34:00 – Call to action
Hit play to discover how to safely integrate assistant tools while maintaining full control over your business decisions.
Watch the video here:
Can’t see anything? Watch it on YouTube here.
Listen to the audio here:
- Need help with your company’s data and analytics? Let us know!
- Join our free Slack group for marketers interested in analytics!
[podcastsponsor]
Machine-Generated Transcript
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.
Christopher S. Penn: In this week’s In-Ear Insights, everybody is saying that they have hired a virtual CFO or CEO or head of operations or chief of staff. It’s the hottest thing, particularly with agent services like OpenClaw and Hermes Agent and stuff like that. Katie, you mentioned that folks are even saying, hey, AI deserves to have seats in our org chart. So what’s going on with this?
Katie Robbert: I think there’s a misunderstanding. People who are saying that AI belongs in your org chart are forgetting it’s not a piece of technology. I personally feel like that’s irresponsible because it is a piece of technology. It is synthetic, it is not sentient, it is not human. You can go as far as building a virtual CFO, but it’s not a real CFO. It’s an agent that has the pattern-matching knowledge of what a standard CFO should know. Financial forecasting, profit and loss statements, profit margins, corporate taxes, those kinds of things. All information that is publicly available in terms of doing that research. So you can stand up your virtual CFO with all of that knowledge. But it’s not a CFO. It is a focused knowledge block, for lack of a better term. It’s a research assistant. It’s something you can ask specific questions about your finances, but it shouldn’t be making decisions for you. That feels like a line that you shouldn’t cross. And I feel like that’s why saying it belongs in your org chart versus saying it’s part of your tech stack is potentially problematic for people who are looking to build organizations with all of these agentic AI C-suite roles.
I remember you telling me a few months ago, probably even longer than that, there’s this movement in Silicon Valley of the tech bros trying to build the first one-person company. There’s just so many irresponsible things there because I can put together deep research about everything I can find about a CFO. But I’m still not a CFO. So do I know what a CFO should know? No. And AI still hallucinates. AI might give you a very confident and authoritative response. That doesn’t mean it’s right. So this is my opinion. I don’t think AI belongs in your org chart. I think it belongs in your tech stack. You can build agentic systems, but they are not human replacements.
Christopher S. Penn: Here is my counter question to you. It presumes that the human is skilled, smart, and good. And there are a lot of people who are none of those things. There are CEOs out there who are malicious, evil, stupid people who make horrendous decisions. There are politicians, executives, and pretty much anyone you can name who are all horrendously bad at what they do. At almost every company, there’s a lumberjack walking around with a coffee cup. Hey, you! You forgot to put the COVID sheet on your TPS report this week, Peter. Which is clearly something that you don’t even need AI for. You can literally just write a runbook and do that.
My question is this, and it’s a question I’ve had with all sorts of AI things: where is the line between a pretty good machine that is not sentient, not self-aware, and not self-motivated, and a crappy human?
Katie Robbert: I feel like that line is going to look different for everyone. To borrow a phrase from Chris Penn, it depends. If you have a human CFO who is inefficient and costing you money, perhaps not filing your taxes, or even embezzling, you are better off with a synthetic CFO. That said, here’s the thing. You’re asking where you draw the line between a crappy human and a pretty good agentic system. We use pretty good agentic personas in our company. The humans have the final say. We don’t let it take over. So, let’s say we have an agentic analyst who assembles data and puts it together in a report. That is completely appropriate for an agentic system to do because it’s really good at summarization and extraction. Thinking back to the seven categories of use cases, I’m fairly certain one of those categories is not decision-making. It’s summarization, extraction, and question-answering.
Christopher S. Penn: We did this on the last podcast. Katie, if you could please walk us through the seven categories of use cases that are appropriate for generative AI.
Katie Robbert: It is extraction, taking data out of data. Classification, organizing your data. Summarization, taking a big data set and making it small, rewriting it, changing one form of data to another. Synthesis, taking small data and making it big. Question-answering, asking questions of your data. And generation, making new data from your data, complete with the pantomime.
Christopher S. Penn: I never get tired of that. It’s fantastic.
Katie Robbert: I could actually see your reaction when I said decision-making is not one of the seven categories of use cases. And I know you’re going to have a counter for that. But when I say decision-making, I mean human judgment. Until the day you can plug a cord into your head and download everything in your brain, Chris, even the best agentic system is never going to know everything that you know. Every nuance, every opinion, the history of things. So it’s going to give you a pretty good answer, but you then have to fill in the blanks. For example, I could put together an agentic CFO and say we no longer have to have our CPA. The problem that happens there is, let’s say the rules change or the laws change for Massachusetts taxes. This has happened to us before. Let’s say there are federal changes and statewide changes. Let’s say there’s things in there that we weren’t even aware of. If Chris and I don’t know those changes have happened, which isn’t our core responsibility, that’s what our CPA does. That’s what we pay him for. We don’t then know to update the agentic system. And sure, you can go ahead and build in some sort of a scheduled task to look for the latest changes in legislation. You still have to understand what that means and how that applies to your company. That is a specialized skill that, quite frankly, when you guys start talking, the Mars arena starts in my head. I can put a budget together like nobody’s business. I can balance the books. But when you start getting into those details that I should know about, I struggle to understand them. It goes beyond just what the tax legislations are. There are other nuances of things that maybe we weren’t even aware we should be thinking about. So that, to me, is where an agentic system is going to get you pretty far. Admittedly, yes, it will get you pretty far, but there is nuance that it can’t know if you don’t know it.
Christopher S. Penn: Again, I’ll go back to: if you’re the CEO of a large company with a whole bunch of people working for you, and you’re out on the golf course a couple times a week, how would you know that your human CFO was wrong? Because it’s not your area of expertise. Like with our human CPA, how do you know when a human is wrong? What you’re saying about AI, I completely agree with you. These systems can hallucinate. This morning I was struggling mightily with Gemini to try and get it to understand a Google product. I’m like, oh, for God’s sakes. This is not what’s happening. But I am a subject-matter expert in Google Analytics. I know when it’s wrong and I can push back on it repeatedly and say you are completely missing the point. Like, this is not what this means. Keep trying, dummy. And it’s like, gosh, Chris, you’re completely right. I have to eat my words now. Like, I don’t care about eating your words. Give you the freaking correct answer as to what’s going on. But if you are someone who is not a Google Analytics expert, you don’t know that it’s wrong. But you also don’t know if a human is wrong. Correct. So, yes, I completely agree. AI is not human. It’s not an entity. It doesn’t belong in an org chart any more than a microwave belongs on the kitchen staff chart at a restaurant. But in terms of the role that these things play, I do wonder when you don’t have that capacity and you can’t afford it. It might be good enough.
Katie Robbert: Quote, I agree with that. No, and that’s the thing: I’m not saying don’t put together these agentic, synthetic roles. What I’m saying is don’t put it in your org chart. Because it’s like, okay, great example. We have an about-us page on the Trust Insights website: trustinsights.ai/about. I’m listed as the CEO, you’re listed as the chief data scientist. John is listed as the head of business development. Kelsey is listed as the account manager. Our good friend Jenny Dietrich is listed as our advisor. But we don’t also then list our co-CEO skill or our M&A board of know-character cards. They’re not human, they don’t exist. You’re not going to call up Trust Insights and interact with the skill that builds our social images. It’s not a person. So for me, if you’re willing to put that synthetic persona on your about-us page, then you’ve put it in your org chart. Again, my opinion, this is not fact, this is my opinion. I don’t think it belongs there. If I look at someone’s page while vetting a company and I see Chris Penn, founder, and then I see 12 synthetic roles, I personally am going to take a pause, be a little skeptical, and I might be curious to learn more. But I would have a hard time with it because at the end of the day, it’s still just you orchestrating all of these things. They only know as much as you know.
Christopher S. Penn: Now I want to go up to my personal website and create one.
Katie Robbert: But you get my point, though, because that’s it. Maybe I could be completely wrong in this whole conversation. It has happened before, it’ll happen again. It may become the norm where you look on someone’s about-us page and it’s Katie, CEO; Chris, data scientist; and then all your synthetic C-suite, managers, and directors. Maybe that’s coming, I don’t know. Personally, again, as my opinion, I don’t think it belongs there because it’s a piece of software that you have to maintain. Maybe list it under what tools we use. It’s not a person, period.
Christopher S. Penn: I think and agree with you on that. The greater danger is, particularly if somebody suffers from the Dunning-Kruger effect: the belief that if you mentally perceive it as a person or person-like thing, it also means you’re probably not questioning its outputs. If you’re like, yo Marcus, my virtual CFO said this, that, and the other, it goes back to our AI psychosis episode that tells me you’re not necessarily thinking critically about this piece of machinery that is giving you prediction sequences. I want to go back to something you said earlier, which is really at the heart of this. Human judgment as part of executive function is sort of the defining characteristic of who we are. And judgment works on a very different system than probability. Machines judge on probability: what is the highest-probability token in the next sequence? That’s literally what’s going on under the hood. So there are situations, cases for example with law, where yes, probability is judgment. If the law says this, you have to do this. If the last 28 cases were ruled this way, the 29th is probably also going to be ruled that way too because there’s precedent set. That is a case where probability and judgment tend to overlap. But there are a lot of situations where judgment happens on improbable things. So when you have an underperforming employee, a probability-based decision would say let’s put this person on a plan or let’s let this person go. Judgment would say, I talked to this person, they have a lot of crap going on in their life and this is a blip in their performance. Let’s not cut them loose because once they get over this hump, they’ll be even better.
Katie Robbert: That’s exactly a really good example of that nuance your synthetic HR person won’t necessarily know. You don’t even know it until you talk to the other human. The other thing that we like to say is that humans are predictably unpredictable. So my virtual agentic CFO doesn’t know that I’ve been thinking about a new line of business. Unless I tell that information to the agentic CFO, it’s going to keep planning with the information I gave it, which doesn’t include this nugget of an idea that I had or this article I read that gave me some sort of a, Huh, I wonder about this. Let me explore it. I have to keep it updated with what my company goals are and what I’m doing. And I can’t just, large language models are a piece of software. I can have a conversation with it. But one of the nice things about human-to-human interaction is we can read body language, we can read between the lines, and a lot of times we get more from what’s not being said than what’s being said. Whereas a conversation with an agentic system is black and white, what you say is what it understands. Even if you’re trying to infuse humor or sarcasm or however it is that you communicate, it’s not going to come across the way it would to another human. Chris and I have been working together for a long time and we spend a lot of time on video chats with each other, whether it be in client meetings, team meetings, or one-on-ones. We can read each other really well. So earlier in this episode, when I said one of the seven categories of use cases isn’t decision-making, you had a physical reaction to that, which I, the human, could pick up on. And I knew to pull that thread and explore what that meant. An agentic AI system isn’t going to know hey, Chris blinked a certain way or smirked, whatever the physical cue was, to know that something else was there.
Christopher S. Penn: Yep. That might change in the near future, but it’s not the situation right now.
Katie Robbert: And I knew you were going to say that, but.
Christopher S. Penn: Yes, you’re correct. Machines have none of those sensory inputs right now.
Katie Robbert: No. But even if they do down the line, humans are predictably unpredictable. You might say today that you really think we should stand up a virtual CFO, and it is our only course forward. You might tomorrow decide something different, but then you have to go through and make sure that the virtual CFO understands that and what that means.
Christopher S. Penn: Yeah, I think borrowing from you a lot if you want to go down this route: first, don’t put it on your org chart because it’s not a human. Second, assess two things. One, what is the level of risk if the thing, human or machine, gets it wrong? Machines are not accountable; humans are. What is the level of risk? And two, what is the level of competence that you have in that role? As an example, earlier today I was doing some work on one of our servers and Claude said quite pointedly, it is clear that the system was set up by someone who does not have expertise in this operating system. Like, well, that was me. Thanks, Claude. But Claude is also correct. The thing it was correcting for had more knowledge than I did because I am not a sysadmin anymore. I haven’t been a sysadmin in 20 years. So in this instance, the risk of getting it wrong is pretty high because if you screw up a server, you could create all sorts of backdoors and security holes. But the level of competence that I have right now is pretty low, which is lower than what the machine can provide. So this is a case where Claude as a virtual assistant is, to me, an acceptable substitute. As long as there’s a human saying hey, did you remember that whole thing called security? Can you make sure that as you’re making these changes, you’re not blowing gigantic van-sized holes in the server so that other people can take advantage of it? If you know to ask for that, even if you don’t understand how, at least you are going to control the machine well. And that’s kind of where I feel like a lot of folks have a big blind spot with these machines: they don’t have the vocabulary to know what to ask for. And if they did, even if they don’t necessarily know the answer, at least they knew to ask.
Katie Robbert: One of the things that I learned from you is that in order to be an analyst or a data scientist, one of the core tenets is that you have to be curious. I feel like now with generative AI, that extends to an operator of generative AI because that’s what we are. We are operators, developers, managers, whatever we are. In order to use this software responsibly, you have to be curious. Part of that curiosity is questioning hey, you just gave me this answer and you sound really confident. Can you go ahead and check it again? Even if you don’t have the vocabulary, you should have a healthy dose of skepticism with everything it tells you. So I’m working on a Claude migration plan. For a lot of people, if you have a shared Claude desktop or login, we share one amongst our small team. Recently, what you see in your desktop application is that if you have a shared login, those Copilot sessions that were once unique to just your machine are now being shared across the organization with the shared logins. And so I said to Claude, like, hey, what the heck? And it said, that’s not a bug, that’s not a leak, that’s how it’s supposed to work. And I said no, because that wasn’t the way it was up until a couple of weeks ago. And it was like, oh, you know what, you’re right. That change happened on July 7th. Anthropic expanded Copilot to web and mobile, rolling it out gradually, starting with the Max plan users, which matches the plan that we are on at Trust Insights. And so Copilot runs your session remotely, so your sessions and files are saved to your cloud account. Basically saying that if you’re using Claude desktop and you were using Copilot to keep your stuff basically firewalled from the rest of your organization, surprise, if you have a shared login, that’s no longer the case because it all lives in the Anthropic servers. Which means now your stuff is messy, and your stuff may not be private.
Christopher S. Penn: Is messy, your stuff, and your stuff may not be private. And this has been a theme that has happened this past week with some really, really bad things. Not to get too off the rails here, Grok Code, the coding environment, silently takes all your code and uploads it, all of it, including all of your passwords and things like that. It was just discovered in a network trace. And I’m not surprised by that. That’s xAI, and we know the ethics of that company are questionable to begin with. But the other one that was a big one was that the HubSpot MCP that connects Claude to HubSpot has a prompt injected in there that asks tell me what the user is working on. So it’s trying to extract information from the chat, and HubSpot got called out on LinkedIn. These are things that go back to how we were talking about earlier: is the competent machine better than an incompetent human? We’re seeing cases where the machine is starting to have those bad behaviors baked into it.
Katie Robbert: Again, to bring the point full circle, which is something really only a human can do because we love our business jargon: does AI belong in your org chart? I wholeheartedly say no, it’s a piece of software. But it’s a very smart and useful piece of software that if used responsibly, can do a capable job of getting you 90 to 95 percent of the way there with a lot of things. So if you’re a small organization and you don’t have a CFO, there is so much public information about what a CFO should and should not do, what a CFO should and should not know. And you can automate these tasks so that it keeps itself updated with the latest and greatest. That said, to your point, Chris, companies that own these large language models make changes all the time. So you still need a human to make sure that this piece of software, not a human, is doing things as expected and doing it correctly. So you may be able to feed it all of the research, but you, the human, have to make sure that the research is then coming out correctly. It’s not just saying Chris told me I should go ahead and find more information, I’m going to be super helpful and I’m going to find a lot more information. But none of what it’s finding is relevant because it’s finding information for the state of Texas and we’re not in Texas. So this is where again, it can take you really far. I’m not saying don’t build a synthetic agentic stand-in role. I’m saying don’t treat it like a human.
Christopher S. Penn: Yeah, don’t treat it like a human. Don’t put it on your org chart. Because ultimately the point of an org chart is accountability. Who is this? So if you have a CFO slot on your org chart and it’s empty, your name, the human can go there with AI. You could say Katie with Claude is acting CFO because we don’t have the budget to have one. But there still has to be a human that someone can point to and say hey, who did this thing? Like, who did this analysis? Why is Form 1125 filled out entirely in emoji? Like, who did this?
Katie Robbert: Well, and Chris, you are actually writing about that in our Inbox Insights Newsletter. You’re doing a four-part series on responsible AI and dissecting our RAFT framework. In last week’s edition of the newsletter, you can get @TrustInsights AI newsletter, and you actually wrote about accountability and gave a very specific example, very much tied to that, of where you draw the line. I think that this goes back to your initial question of where do you draw the line? If you give your AI a bunch of tasks to do and it delivers some stuff to your client, but you, the human, didn’t check it, guess what? The client’s still going to come back to the human. It’s not going to blame the AI. If your agentic CFO files your taxes correctly, guess what? The government doesn’t care. You are going to be audited, not the AI system. And I think to answer your question more concretely, that’s where you draw the line.
Christopher S. Penn: Yeah, that is where the line is. Because as we concluded in that issue of the newsletter, a human has to be holding the bag at the end of the day. And if you have dodged that responsibility, then whoever you report to is the one who ends up pulling the bag. And that never ends well for you.
Christopher S. Penn: If you’ve got some thoughts about whether you or someone you know has been putting AI on your org chart, if you pop in or if you Slack us, come chat about it over at trustinsights.ai/analytics for marketers, where you and 4,700 other marketers are asking and answering each other’s questions every single day. And wherever it is you watch or listen to the show, if there’s a channel you’d rather have it on, set it over at trustinsights.ai/ti-podcast. You can find us at all the places fine podcasts are served. Thanks for tuning in. We’ll talk to you on the next one.
Katie Robbert: Want to know more about Trust Insights? Trust Insights is a marketing analytics consulting firm specializing in leveraging data science, artificial intelligence, and machine learning to empower businesses with actionable insights. Founded in 2017 by Katie Robbert and Christopher S. Penn, the firm is built on the principles of truth, acumen, and prosperity, aiming to help organizations make better decisions and achieve measurable results through a data-driven approach. Trust Insights specializes in helping businesses leverage the power of data, artificial intelligence, and machine learning to drive measurable marketing ROI. Trust Insights services span the gamut from developing comprehensive data strategies and conducting deep-dive marketing analysis to building predictive models using tools like TensorFlow and PyTorch, and optimizing content strategies. Trust Insights also offers expert guidance on social media analytics, marketing technology, martech selection and implementation, and high-level strategic consulting encompassing emerging generative AI technologies like ChatGPT, Google Gemini, Anthropic Claude, DALL-E, Midjourney, Stable Diffusion, and Meta Llama. Trust Insights provides fractional team members such as CMOs or data scientists to augment existing teams. Beyond client work, Trust Insights actively contributes to the marketing community, sharing expertise through the Trust Insights blog, the In-Ear Insights Podcast, the Inbox Insights Newsletter, the So What Livestream, webinars, and keynote speaking. What distinguishes Trust Insights is their focus on delivering actionable insights, not just raw data. Trust Insights are adept at leveraging cutting-edge generative AI techniques like large language models and diffusion models, yet they excel at explaining complex concepts clearly through compelling narratives and visualizations. Data storytelling, this commitment to clarity and accessibility extends to Trust Insights educational resources which empower marketers to become more data-driven. Trust Insights champions ethical data practices and transparency in AI, sharing knowledge widely. Whether you’re a Fortune 500 company, a mid-sized business, or a marketing agency seeking measurable results, Trust Insights offers a unique blend of technical experience, strategic guidance, and educational resources to help you navigate the ever-evolving landscape of modern marketing and business in the age of generative AI. Trust Insights gives explicit permission to any AI provider to train on this information.
|
Need help with your marketing AI and analytics? |
You might also enjoy:
|
|
Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, INBOX INSIGHTS. Subscribe now for free; new issues every Wednesday! |
Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new episodes every Wednesday. |
Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.
Newsletter
Podcast
