← Using AI at Work: AI in the Workplace & Generative AI for Business Leaders
Using AI at Work: AI in the Workplace & Generative AI for Business Leaders
62: Using AI for Custom Worklow Solutions with Andrew Amann
00:49:08
Loading summary
Transcript105 lines
- [00:00]
A
This podcast episode is sponsored by chiefaiofficer.com offering training and certification through the International association of Chief AI Officers. Interested in a new career or leveling up your value in the marketplace? Chiefaiofficer.com can help. Welcome to Using AI at Work. I'm your host, Chris Daigle. Each week we'll be learning how today's business owners, entrepreneurs and ambitious professionals are getting more done with smart use of tomorrow's tech. Let's get started. Welcome everybody to the latest episode of Using AI at Work. Andrew is our guest today and before we started this, we were just having a quick conversation about some of the cool things that he's doing. And I think that for those of you who are listening to these podcast episodes looking for, I don't know, direction, clarity, solution, anecdotal evidence on why you need to do this, with what Andrew and his team have been doing for companies from startup all the way up to enterprise, I think is going to address that for you. So, Andrew, before we get started, maybe just share a little bit about what you're up to, how you arrived on the podcast today and all that good stuff.
- [01:14]
B
Sure. So We've been building AI products since 2016. We founded the company in 2012. So I've been deep into AI for a long time now. I've also been building products. So since I was basically 2008, when I graduated school, I was a nuclear submarine engineer for the first six years of my life. And then I transitioned to the company that we ran, that I run now, which is a now considered a consultancy or an agency. But at the beginning it was a startup studio where we built 14 products from zero to one, sold a few of them, lost a bunch of them, but had a lot of fun along the way. And what I'm thinking about now is, you know, the same thing I was thinking about when I was in the nuclear submarine space. I was responsible for improving the supply chain. They had a system called Primavera, which is a giant manufacturing board of scheduling and when stuff needs to get done and at what time so that the submarine can be built in a five year period. And my job was to automate that and make sure that when something broke, how do you bring it back on schedule, but also how to systematize it, what, what stacks up against the thing that happened before and where are the bottlenecks, what is the throughput? And so learning all that, I became a supply chain expert, left that job in 2016, started this company and didn't know that supply chain was so important for AI automations and AI workflows.
- [02:30]
A
Yeah.
- [02:30]
B
And so having that brain and that knowledge around systems thinking has allowed me to go into a lot of these organizations, understand their main problem, not the problem they come to us with, but their main problem, and go in and fix that. And we're seeing a lot of six figure investments, sometimes even for eight figure returns.
- [02:45]
A
Yeah, not surprised. You know, that's one of the things that when I'm having conversations with executives, the people that you're dealing with, they just want, quote, unquote, AI. Yeah, they don't really understand. And the reality is it's not like an event that occurs, it is a process. And being a supply chain background, being a systems thinking individual certainly allows you to have better impact than somebody that didn't necessarily have that experience. So I'm glad to hear that. So I want to talk about the startup experience. This is my first startup that I've taken from zero to one and it sucks, man. It's a lot of work. I mean, we're getting over the hump pump now, but doing that 14 times seems like stupid. No. You know, certainly an opportunity to learn a lot of lessons because there's very little forgiveness in that environment for mistakes. So I would imagine. So for anybody who's listening to this that maybe wants to pursue an AI consultancy or something to do with AI, combine those two experiences that, that a rapid velocity of startup iteration combined with, I guess, you know, over a decade in AI, what advice would you give them?
- [04:00]
B
Yeah, I do get a lot of questions from mentors. When I mentor people around, should I be starting my startup? How should I be operating this startup? When I see a young person, you know, just graduating college, high school, and they're like, hey, I want to build my first startup, I always tell them, go join a company in that space because there is a probably a problem that you are willing to solve. But some of the times that you need, the most important time of your life is your age, 20 to 26, when you need to learn how business operate. And if you miss that period and you stay in startup mode, you're not going to understand how an enterprise company operates because you're never going to be inside of an enterprise company. One of the best examples I've given, I haven't given this in a while on a podcast, but if you came up with like the greatest dog toy that's ever been known to man or woman, and you're like, hey, I want to put this into the market, the ability for you to create that dog toy and sell it and market it and bring it from zero to one is going to be very challenging. It's the road that you're going down and that you just explained. It's like, it's a really hard process. But if you were to join a company and you say, I don't care if I own this, but I have an idea for you, and maybe you don't tell them in the first year, but you tell them in the second year, they probably have some sort of studio or warehouse where they have a bunch of dogs come and test that toy, and then they probably have a place where that prototype can be made in seven different ways or 20 different ways with machines and different types of squeakers and all of these different variations that you could do that you as a startup would never have access to. And so now all of a sudden, you're in a building, it's your idea. People are looking at you saying, like, what should we do Next? And you're 22, 23 years old, and you're like, hey, why don't we try this squeaker? Why don't we try this end? And golden retrievers love it, or huskies don't. But you're experimenting. But you're also, more importantly, learning the system. You're learning what it's like to actually do a product from 0 to 1 correctly. And big companies, as antiquated, as antiquated as they are and as slow as they are, they do still have a process to get products out the door. Yeah. And some of the greatest companies, like Apple, has done it for many, many years. And some of the worst companies, like Pets.com wasn't able to do that. Right. So you. There's a, there's a polarity, right? There's an end to end, like distance of what, what is a good company, what's a bad company. And just joining one puts you into that scale and you're going to learn and you're going to educate yourself. And maybe you don't own that dog toy, but you've learned the process. And now after your first dog toy went through that, now you can invent your second dog toy. Because I guarantee, if you have a dog toy idea, you're going to have two dog toy ideas.
- [06:22]
A
Yeah.
- [06:23]
B
And so that's, that's the advice that I give to young individuals, is learn the system, learn the process, get beat up a bunch, get a bad boss, know what it's like to be mistreated a little bit and then say, I'm not going to do that to my employees and then go start a company.
- [06:35]
A
I think that's good advice. So some of the listeners that we have here are individuals who are, have established careers. They are enthusiastic about AI and they're looking for perhaps an opportunity to take that experience they've got, which is exactly what you were talking about combined with that AI Interesting. And have a career change. Have a, I don't want to call it a side hustle, but an opportunity to, I don't know, take advantage of the time that we're in right now with high demand, low supply, inefficient market of like, how do I even find that talent? So for somebody listening, because it, for a lot of them that maybe they've never been an entrepreneur even, I mean an intrapreneur entrepreneur.
- [07:20]
B
Yeah.
- [07:21]
A
So there is a, a lot of mistakes that they will have to or things that they don't know. They just don't know what they don't know. So for anybody like that, I know you guys are doing a lot of consulting with your agency and that you've got, you've seen this process with others when they've joined your team. For those who are listening that that's their path or their intention. Maybe some advice there.
- [07:46]
B
Yeah. So if you're coming up with an idea inside of your company, there is a process that is associated to what you're doing at your job. And if you're trying to figure out like how do I extract what I know about this accounting firm or this legal firm or if you're in manufacturing, which is a great space to be, to intercept AI AI ideas, some of the best products we built. And I just got a proposal today. I can't tell you who it is, but I'll explain. The idea of this is in every big company there is a playbook that is 400 pages long. And every time a contract gets sent to this company, there is a person at that company that has to read this 400 page document, understand the rules of what that document needs to do and they need to write a human result like a memo and then that memo actually gets submitted to the government with your taxes. And so if the tax man ever came and said, hey, why did you do accounting this way? Well, you got an answer. You say this person did it a certain way. We did a prison project where every, every time somebody goes to jail, there's a thing called the PSR massive document. Takes 40 hours to do that. The company's called Prisonology and 40 hours of work For a consultant to read through all of these documents. And Then there's Rules 1 through 27 is a rating of how severe it can be for this person to go to jail. And that rating system is based on a set of rules. One is like a misdemeanor and 27 is murder. Well, that rule can be trained to the AI. And if you know that, and if you're doing that process day in and day out and there's 40 people next to you doing that process day in and day out, take that guidebook and automate it and you don't have to leave your company to do this, right? You could use lovable and some AI tools. But that, that is a public domain document. That is a guidebook for your industry. Manufacturing is like the best place. Why? When I'm doing tooling and I want to put a screw onto the edge of one of the widgets that I'm building out, there is a very specific machine book that tells me what screw, what size, how long, how length, what thread length. Like every single dimension of that screw is determined by the machiner's handbook. I don't need to go look in the handbook anymore. I need to ask in Natural language to ChatGPT what screw goes in this hole. But I need to train the natural language of what my widget does. And so if you take the knowledge of what a great machinist knows when they're looking at that widget and say, tell me all the things you know about that widget. Great, here's a screw that goes with it. Because I don't need to know anything besides what you know to figure out the screw in that machine who's handbook. Those are the places that are going to get the biggest returns. And that is where AI excels is summarizing and refining giant handbooks, giant processes, giant workflows, and for the prison one, there's 40 hours of work down to four hours. That's a 90% savings.
- [10:24]
A
Yeah, for sure. No, that's a great point. So I guess it doesn't even really matter what industry you're in. If, if you are familiar enough with your domain and you can see like everybody hates doing that, it's a dumb process that we have to follow. How can I introduce automation, AI, human augmentation, whatever, into making it better, that's a great place to start for sure. And obviously if you do it within a company where you're already at, even if you're not looking to go out and be an entrepreneur, career trajectory is guaranteed. If You're a bringer of those AI solutions for sure.
- [10:59]
B
Yeah. In the dog example that I gave, if you started that company instead of going to the big company with your dog toy, if you fail on the top of eating ramen every single night, you have nothing to fall back on. If you failed at the company, they're going to say, cool, why don't you go to this department instead and just go back to your day job. You don't lose a job, you don't lose an opportunity. You're still building your career. And Outside of a LinkedIn resume, no one's going to know that you actually even failed that unless you tell them. Right. You are protected by the ceiling of that company and it's a great place to be because people don't. People always think they want their own idea. And when we're young, we have that ego and we really want people to know that we have that idea. If you can learn to give that up at a young age, you learn so much more.
- [11:40]
A
Yeah. Had a mentor tell me, lose the ego and laugh all the way to the bank.
- [11:46]
B
Yes.
- [11:46]
A
So, yeah, same situation there. So you guys are bringing solutions to clients that I'm very curious because we talked to a lot of clients as well. And here's what we're hearing now, which we didn't hear even four months ago. We talked to a client, a client says, oh yeah, we're using AI a lot right now. First, they're you, they're meaning they're using generative AI, not data science, machine learning, anything like that. And secondly, when I say that's fantastic, tell me more, they say, well, we use it to the two things we always hear. Right. Emails, summarize, documents.
- [12:23]
B
Yes.
- [12:23]
A
And as you know, and anybody who's listening knows that's serious about this, there's a lot more that ChatGPT can do than those two things. What are, what's your kind of temperature as you're talking to clients? Have you seen them self identify as maybe more sophisticated now than they were at the beginning of the year?
- [12:41]
B
Yes, but in the wrong direction. And it's been the hardest sales cycle of the last six months. So we've been selling AI products since 2016. These last six months have been the hardest because there is a $10,000 solution or a free solution to the problem that they're solving for on Twitter and it's some person that created one app or a flow in zapier type tools that they look at and they're like, oh, can you. Johnny in engineering can you go recreate this so that we never have to do invoicing again? And it's no different than duct taping and band aiding stuff together. When we had software tools and before APIs, we were like, oh, why don't we just talk to the sales department through this chat bot called Lotus Notes or Instant messenger, right? Like, we tried all of that and AI is not the deterministic results that everybody dreamed it was going to be. It's undeterministic. It guesses, it tries to solve, and it tries to figure it out. You can't look at AI. And when, when we go into these companies and they're like, oh, yeah, we're trying AI in the places that you said customer service, writing emails and all that.
- [13:42]
A
Yeah.
- [13:42]
B
What they forget about is the direction that AI is heading. And I always tell them, in one year, you will not be copy pasting out of ChatGPT. I can promise you that. That the tool you're using is going to have a large language model baked in. And for a product like Contently, we built Contently for Contently AI for this giant company called Contently. The people that are using that tool don't fully know they're using AI because it's baked into the process of the tool that's there. That is a great use case for AI because the humans are doing something that used to take them a long time and a very short time, and the AI is baked into the process. And that's going to continue to happen. Copilot and Gemini are going to come out in workspace and inside of Office365. And when that happens, you're no longer going to need to go to ChatGPT to copy paste into your workflow. Now OpenAI is going to have their own workflow. You might be an OpenAI company, right? And you use everything in OpenAI and all their connectors to all these MCPS and all the tools. And that's a great use case too. And you're not copying out of ChatGPT anymore. It is your home base. But I can promise you in one year that your phone. Well, I don't know about Apple, but if you're on Android, you're going to have a personal large language model that knows everything about your calendar, your Google mail, your notes, your wife's calendar, your wife's notes, your text messages. And when I ask it to pick my kids up at school or I tell my wife to, it's going to give all the context necessary for my calendar and the times and all the different things that I used to have to remember myself, use all the databases and send it to her. So that will happen next year. And now you get to think in my workplace as an, as an executive, where am I investing right now? Where I can silo every single person in my company down to this workflow where I can have user permissions, controls, access control, I can turn things on and off, I can, if somebody quits, I can see all their prompts. How do I stop people from copy pasting because I don't want my data on some random thing that I can't control. How do I keep it in house secure and how do I make sure they still get the same value from AI that they're experiencing right now?
- [15:44]
A
Is that what you just described? Is that the type of thing that you're starting to do with clients, that level of granularity on AI application?
- [15:53]
B
Yes. That has been the big unlock for us in 2025 is we're moving from large enterprise companies that could afford giant platforms like we built at the beginning to mid sized companies and you know, the smaller $5 million companies gaining access to build workflow automations that are saving them millions of dollars. And that only is true because they're bringing stuff to the workflow level, not copy pasting from some random tool.
- [16:17]
A
Interesting. So you mentioned a few tools. Are you guys doing this in N8N make Zapier?
- [16:24]
B
No. So we, we are interested in that for even smaller companies or even smaller departments. We have yet to figure out how the department works. I mean we have PhD level engineers and elite product managers that have gone to school. So those tools are great and they solve a specific problem, but it's a $5,000 problem. We're solving hundreds of thousands of dollars per problem. And when it takes that, it takes a lot of thinking of what is the human going to react to when they go through this process, what are the customers going to see and what are all the edge cases. And when you're working with AI, you need that product manager to see those edge cases. And that's why those tools, they break with any edge case.
- [17:03]
A
Yeah, people, an experience that we've had with clients is they were like, oh no, no, we got this off Twitter. We're good.
- [17:11]
B
Yeah.
- [17:12]
A
And they can get it about 90%, maybe somewhere around there, but it never ships, it never goes into production.
- [17:19]
B
Right.
- [17:20]
A
So for those who are listening to this right now and they're like, we can do this in the garage, we don't need any help with this Because I can, I can get that JSON code from a Twitter reply or something like that. What do you, are you a, are you seeing any success at all, even at like maybe some small levels with that? And at what point would you say that that would start to break and they would need to explore solutions similar to what you guys bring?
- [17:48]
B
Yeah, I see agencies that are seemingly having a lot of success with the AI automation tools that you mentioned. Is it successful for the enterprise company? I don't talk to their clients. And is it successful for their clients? I'm not sure. I don't, I don't know their clients well enough to know that they're happy with the results. But I do know there's a lot of new jobs appearing on the, on the marketplace. AI automation is going to be a job inside of every zapier department. AI automation engineer. And so if you're going to hire for that specific job type, that person should be responsible for understanding the domain and applying those types of tools to simplify people's lives. And I think that's a very valid, you know, higher. I think you're gonna get a big return on investment on that. Where it breaks is the communication between departments. I like to give the example that like when companies started back in the 90s and 2000s, they had all these departments that were like, cool, we need a software tool to help our department. And so finance got QuickBooks and sales got HubSpot and engineering got all their IDEs and they all had these tools and then they're like, ah, how do we talk to each other? And so they, they slacked and Lotus Notes and Instant messenger and you know, all this stuff came along and then Slack came along. But they also realized they need the data to talk to each other. So you created all those APIs to communicate between all the different departments and it worked pretty well. Like you could submit an invoice and accounting and it would show up in the HubSpot framework. Right. Or Engineering could ship code and it would change things. So like that worked well and APIs like became this, this pipeline of data between all these departments. When AI comes along, we're trying to just use the same framework that like human in department A needs to work with human, department B just use AI to communicate. And you're like, well, I already do. I already like send my invoice to HubSpot. I don't need AI to do that. We need to rethink what AI is good at. It's a different form of computer language. It's a different form of ones and zeros. It's predicting the next word. It's good at summaries like you said. It's good at analyzing. It's good at taking a lot of data and simplifying it so that we can understand. It's good at taking PowerPoint presentations and reports and analyzing those reports and giving me the exact answer that I need to for my clients. If you think of it that way, it's a new department, it's a new place to send the data to, to do the work, to come back with the answer that I need. So I'm not sitting there writing a report all day long. I'm sending it to the AI department to come back with some sort of answer to me and say this is the result you were looking for. Oh, and I just saved you 40 hours of work. And when you think of it that way as a CEO, you're like, cool, go staff this department with all my knowledge, all my data, everything about my company, but don't train it on my competitors company, don't train it on my competitors data, don't train it on ChatGPT, train it on me. Because I built this company with all this data, all this engineering, all this ideas, how we do business, how we communicate. Train it on that so that when I get my response back, I can trust my employees are going to trust it to send it out. And that's the difference in thinking. I think that needs to happen.
- [20:38]
A
So is that a like a house LLM for your company?
- [20:42]
B
It's a knowledge base. It's a place to like, I call it vectorizing your data, but it's a place to communicate with the meta descriptions of where you're storing information. Because we take information, we vectorize and it's a giant summary of a document or an invoice or whatever it happens to be. But the engineering that needs to happen from an agency level still to this day is that when I recall something in my company from natural language and I say, hey, can you give me all the invoices from Chicago? Well, from sales department A or sales department B, right. There's too many nuances still. So what custom code needs to do is create the orchestration between the agents. And so instead of having a one shot prompt, hopefully find the right answer in the sea of data, you're creating a deterministic path for an agent to talk to an agent to talk to an agent to talk to an agent with deterministic results all the way back up to the front that says I got you your answer and I'm 95% sure this is what you're looking for.
- [21:36]
A
Is that, is that level of like this agent is very specific to this domain.
- [21:41]
B
Yes.
- [21:42]
A
This one's very specific to this small domain.
- [21:45]
B
Yes, that's exactly right.
- [21:47]
A
Interesting.
- [21:48]
B
When you break it down to small parts, you can actually do QA on it. Because if I was to read a document and tell you, give me the date back on this document.
- [21:55]
A
Yes.
- [21:56]
B
And I ask a large language model, hey, summarize this document and give me the date. Well, if there's five dates on it, it's going to be super competent and say your date is this. And then you're like, no, it's this one. It's like, oh, I'm so sorry. Yeah, like I gave you the wrong date.
- [22:06]
A
You're right.
- [22:07]
B
Yeah, you were right the whole time. But if you break it down to problems and you say the date of the document in the top left corner and it's read either, you know, in the European way or the American way, it looks in this format and you code that when it comes back and it says, I can't find a date in the top left corner, but I can find it in other places in the document. Do you want that date? Because it knows it failed. But if it didn't fail, we also can QA it and say, hey, you gave us the wrong date. Why'd you give us the wrong date? Oh, the document has it in the bottom right corner all the time. So let's reprogram this to look in the bottom right corner every time.
- [22:40]
A
Interesting. Okay, so that is a lot more sophisticated than most users of AI that say, oh yeah, we're, we're, we're power users, we're using it. Yeah, of course they're not doing that.
- [22:50]
B
No. And they're failing. Right. I'm sure a Twitter zapier connection is fun to play with and it looks really cool in the boardroom, but yeah, that's interesting.
- [23:01]
A
So for any of the. In those of you who are listening, who. And I've got a few of you in mind, I'm not going to name your names, but who think, hey, you know what, it's just cheaper and easier if we learn how to build this stuff ourselves. Unless that that process that you're trying to do in house involves a multi year education resulting in a computer science degree or a data science degree. You can only get so far. So please. Doesn't mean that you can't get benefits from copying and pasting the JSON from the Twitter thing, maybe you save yourself, you know, four hours a week on something, and that's fantastic. But if it's something that's mission critical to your business or it's something that is going to be a foundational piece of your business, the, the, the influencer on Tik Tok, sharing a architecture of a process is not going to cut it. So thank you for that clarity, Andrew.
- [23:53]
B
No problem.
- [23:55]
A
So what with the stuff that you're doing with clients, like, how do they find you? Because this is my experience is that there's no question the media and results are causing FOMO in the marketplace. From the, the have nots. Right. The haves are the people with the knowledge, but not just AI knowledge, but also like a business acumen. You understand business.
- [24:23]
B
That's right.
- [24:25]
A
These people exist, but there's not that many of them that I would trust. And if I'm a business and I want to find those people, like, I can't go to ebay, I'm not going to find them on upwork, you know, if I have to, I guess I would. How are you, I don't know, benefiting or exploiting this gap with supply and demand and inefficiency in, you know, pairing the two.
- [24:52]
B
Yeah, we actually talked prior to this call about some of the books that are behind me, and one of them is Alan Weiss's book about Million Dollar Consulting.
- [24:59]
A
Yes.
- [24:59]
B
And if you focus on his teachings, we're a service agency that delivers a specific service that when it fits the client, they understand the exact purpose of it. And the best example I can give is there's two types of plumbers. I'm sure there's like a hundred types of plumbers, but in my example, there's two. There's somebody that I can call at midnight and say, hey, the water broke. Can you be here tomorrow morning? And they're like, sure, fifteen hundred dollars. I'll show up in the morning, I'll fix whatever you got. And so you pay that money and you get that person there. There's another type of plumber that I can call and say, hey, my, my water broke. And they'll say, hey, I'll be there in two weeks. It'll be $300. And depending on your scenario, you might not have that big of a leak. And you say, great, $300 sounds like a great scenario here. I'll put some duct tape around it or shut the pipe off and wait for you to show up. But if you have a lot of water on the floor, you need that plumber there, the next day, that service offering that he has, that he answered the phone at midnight. Because I bet you that other plumbing company does not answer the phone. They only work from eight to five. They schedule two weeks out. That's their service model. And they're cheaper and they get the job done. But that is the offering that the client needed at the time when something happened in their life. We offer the same thing for AI services. We are that AI like consultancy where you get top level product managers that are going to train the people that are using the product on how to use an AI product, or sometimes not even an AI product. An AI workflow automation that has nothing to do with AI but solves your digital transformation problem. So we have the product managers to do that, we have to hire them and train them. And they're elite. They go through a lot and lot of experience to get where they are. Then we have PhD level engineers that are phenomenal at communicating, at understanding, at seeing the problem and then saying, this is how we should solve it the easier way, not the way that like AI tells us to solve it for. But this is more deterministic, this gives better results. This is how we've solved it for 20 years. There's a giant GitHub repository of like, exactly how we can test this, right? So that knowledge of here's a problem, here's a solution. You come to us and in three months we can launch some sort of digital transformation that's going to make you your return on investment. If that's the service you want, that's us. Now, how do they find us? Hopefully there's a way like podcasts or LinkedIn or I'm better at shouting on the mountaintops. But there's got to be a better way to go through the weeds right now because I feel like it's the Wild West. There's agencies popping up every day, there's people going into AI automation every day. Upwork is flooded with engineers that say they can do exactly what we do. I don't know. I don't know how you go through the noise unless you have a trusted referral.
- [27:30]
A
If you're enjoying this episode and want to learn more about how to start using AI at work, we've made it easy for you. For just $1, you can have full access to the Chief AI Officer community, which will give you additional training, custom software, daily Training, calls on AI tools, using AI automations, getting more from your ChatGPT sessions and the business of being an AI consultant. Simply go to chiefaiofficer.com insiders to accelerate your AI journey. Now back to the episode. And how are you guys? Is it word of mouth? Is it people, previous clients? Is it. Where are you getting business from now?
- [28:09]
B
Yeah, we split everything. We try not to rely on one source. So we do Google Ads, SEO, organic, social. I write on LinkedIn every single day. I also have a great funnel on my website. So if you ever visit my website, there's a good chance that we can source those demographics as well. But we have all these shots on goal that we try to find our right client at the right time.
- [28:32]
A
And one of the things that I've noticed with it's still happening, but early on in this, somebody would hang their shingle and say, we're an AI consultant. And if they had distribution very quickly, their bandwidth was absorbed, just gone. Right. And they weren't necessarily hadn't had the foresight to say, okay, how do I scale an agency? Which isn't. It's not easy.
- [28:54]
B
No.
- [28:55]
A
So are you guys finding the same situation to where pretty much as soon as you've got bandwidth, it's absorbed and you're kind of at capacity, or how are you handling the scaling challenge?
- [29:09]
B
Yeah, I will say that there are agencies that do scale in this area and they do it their way, but they add people to it that aren't experienced. And it depends on the level of depth of what you're asking that agency to do. If it's a hard engineering problem, that takes practice and other clients, you know, trials or things we've solved in other clients, to solve the current problem, you're going to have to find an agency where that experience is stuck inside the agency. And so one of the things we're very proud of and very observant of is that our culture is something that people want to stay for and work for. And so we focus on keeping people as long as possible through all the different offerings that we have to be a great remote company and a European and American company. And so if we can keep the great talent and we can hire the great talent now, our job is to apply those individuals to keep their mind stimulated. Because most of the great talent wants to solve really hard problems.
- [30:01]
A
Yeah.
- [30:02]
B
They also want to be challenged, and they also want to be on the cutting edge of technology. So my job is to bring them those clients that are thinking two years ahead. And to find those clients, I have to match them with these great engineers that are great communicators and saying, I'm up for the challenge. I will try this. And so we have to price and find the people. And it's all a math problem. And I think the angle that I try to do to scale the agency is we're not venture backed, we're not PE backed. We can grow at our own pace and we grow at 20 to 30% a year. And most importantly, we've been on the Inc 5000 for four years actually. This will be our fifth year in a row.
- [30:39]
A
Congratulations.
- [30:40]
B
Thank you. I appreciate that. And we do that at our pace. And so if something was to dip, we have a month or two months or three months to recover. And what do we do? We have an action team. And that action team goes out and solves some of the coolest AI problems that are in the space right now. And they love it. They get to experiment, they get to like play with the toys that they weren't experimenting with while they're with customers. Now who does that benefit? It benefits the next customer. Yeah. And so we do that across the board where you have great people that really love what they do. And if you can give them those types of resources and the ability to work remote and the ability to be challenged, we hopefully those dips last shorter and shorter every single time it happens.
- [31:21]
A
So your agency is going to be a lot different than some of the other agencies that are listening here that don't have, you know, advanced degrees in data science supporting their solutions. They're using off the shelf generative AI stuff. Right. How much of what your team is doing internally are you leveraging, leveraging generative AI and then for the clients, are you leveraging generative AI?
- [31:47]
B
I think the easy answer is speed. I think the way that our team has learned it has allowed them to take an eight month project and turn it into a four month. And then even a four month, now I'm seeing turn into a three. So I think their ability to adapt to the new IDEs and working with cursor and some of those other tools at still coding. But coding with speed has allowed us to get products out quicker. Yeah. And it benefits the client, it benefits the price, and it benefits us because we get to run through more clients every year and we love getting new clients. So it all is like a self fulfilling prophecy. Right. Because as AI gets better and as we deploy code better, we get to see the next round earlier. We're not building a software, waiting a year being like, oh yeah, six months into that AI release this cool thing that you could have just used from the beginning. We're actually like Releasing it in three months and then saying, hey, this next thing we can build faster for you.
- [32:40]
A
So one of the things that has been coming up is I've got some relationships with teams very similar to what you described. They've got the strong technical background, but they're leveraging the accelerant in some portion of generative AI to get them there. Pricing their, their results is challenging for them. And then to reference Alan Weiss with that results based, you know, like compensation or value based compensation I guess is how he calls it. How much of your pricing model includes that as compared to simply an arbitrage of I pay my team X and I charge X plus?
- [33:20]
B
Yeah. So you're saying like how do we price projects based on the known, known when we enter the project or based on what we knew from the clients?
- [33:28]
A
Well, one, one of the biggest challenges because we, we, you know, we have a big community of individuals who are out there making a living as an AI professional. Right?
- [33:36]
B
Yes.
- [33:38]
A
And pricing is, there's, there's no like standard thing. I know how much this widget costs. This widget + five, four years of college applied in. Your business is 65 an hour, whatever, right?
- [33:50]
B
Yes.
- [33:51]
A
With AI, because, and I'll give you a perfect example, we had a client that wanted some help in building a tool that was going to compress the result cycle for them to do due diligence on activities they were performing. I don't want to give away too much. That solution was going to be worth if applied, if it, if AI delivered on what we suspected it would, that solution would mean that they were spending about 90% referencing the same number that you talked about earlier, 90% less resources required to get that same result and get it faster.
- [34:30]
B
Yes.
- [34:30]
A
A price was presented to them and, but their challenge was you're going to charge me this much, but I can, you can build it in, you know, a week or two weeks or whatever. Because this wasn't some sophisticated situation that needed like, like precise tolerance on screw selection or something. Right. This was, this was knowledge work.
- [34:53]
B
Yes.
- [34:54]
A
And even though it would save them that they, they had a hard time connecting result with price and time.
- [35:05]
B
Right.
- [35:05]
A
You know this maxim, you're from engineering. Good, fast, cheap, pick two. Right. That paradigm doesn't exist anymore. I can still get good quality very quickly and it doesn't cost me as much. So I, I guess in a way I'm asking how are you encountering those types of like people going to. Doesn't compute. I don't understand how you can charge this much but it only takes this, you know, two weeks or whatever to build. And if so, what is the, what's your perspective on that? Or perhaps suggestions on being on the other side of the table and giving a response to that when a client says that?
- [35:41]
B
Sure. I look at it a lot this year because I think there was challenges in the transformation of how people thought about Software. I think SaaS is going to have a tough time going forward. And the model of SaaS is starting to be questioned. And so I think people were stuck in this mindset of like, cost per seat. And they knew that the software cost millions if they wanted it for themselves. But they're like, if I just pay cost per seat, that makes sense because the salesperson now has four pieces of software, add them up. I can do the math very easily to make it valuable for that salesperson. When you look at AI and you look at it as a tool and you think about it from a cost per seat perspective, you start to get those objections that you just suggested. People look at it as, I'm not going to see the returns because this person or this tool that I'm actually implementing it for is fragile based around what they're going to tell the AI to do. It's a tool for them, it's not a tool for the business. Most Twitter tools that you see in AI are productivity hacks. Yeah. And what you forget about is that employees are hired to do the work because they like doing the work. There are parts of everyone today that they enjoy. And if you come in and say this tool is going to take away that, they're not going to let you take that part of their day. What you have to find as a good consultant is the part of the day they don't want to do, the part that they leave till Fridays because they didn't want to do it the rest of the week. And it's something that they wish could just be done automatically. And that might not be a high value return. Which means that these smaller AI tools don't actually have a place to be because that person should just do the work on Fridays. It's not that big of a return on investment. And so that concept now brings us up to the enterprise level. Where is AI live in the enterprise level? And when you start looking at 40 or 50 people having that Friday task that they hate now, the tool becomes valuable, but you have to orchestrate it between 40 or 50 people with different opinions, with different ways of working to do that job. Yep. And so the way I look at it, and the way it's very easily seen to me when I'm on both sides of the table is that when you're hiring humans at a consultancy to do the work with other humans at a department, then you're going to solve a problem for everybody. If you hire a robot to do the solution you're hiring, you're, you're solving for nobody. And the humans will hate it. They will revolt. They want to go back to doing the jobs that they actually like doing anyways. And the AI will be left on the corners collecting dust. The what we sell are great product managers and great engineers and they're communicative. They have been in the other side of the shoes that the, the other person's shoes and they've realized the problems that they had and they're an enhancement to that department. They are getting along together, they are talking every single week, they're solving problems together. It is hiring an extension of that department so that we can build the rocket ship, put the rocket fuel in it, teach them how to fly it, and then say this is how you fly to the moon. And when you do that, it takes humans, not robots, to do that consulting.
- [38:36]
A
That's an interesting perspective. You know, I, I think I track very similarly with you on making sure that AI is handling the crap work. Yes, right. Because my experience is the same thing. If they don't like it and they're not good at it, following from traction, maybe the delegate and elevate exercise.
- [38:53]
B
Right, that's exactly right.
- [38:54]
A
If, if they don't like it and they're not good at it, it's going to be Friday afternoon. It's going to be the last thing they do. They're going to put the least amount of time into it, the least amount of quality, and that's risky. And we, we kind of look at that by starting there as almost like a change management tool. If I can get the crap work off of your plate really quickly and easily, you start looking around going, hey guys, this AI stuff is pretty good. Let's do more.
- [39:22]
B
That's right, right.
- [39:23]
A
So very interesting. We kind of discovered that on our own. But it's good to see that others are having that, that discovery in the wild as well.
- [39:32]
B
The most amazing thing is when you work closely with other humans and you become part of their team just for even a three month period, you get to witness their energy levels. And when you start, it's curiosity and kind of fear, you can always tell they're like, what's going to happen with this Agency coming in. I can tell you, by the middle, it starts getting exciting. And by the end of it, there are so many ideas of what they can do with the product. It re energizes that department, and they're like, now we can do this. Now we can sell it here. Now we can do this. I mean, we did Charles Barkley, Chuck GBT for FanDuel. That was a product where we didn't even know if it was possible to control a large language model, to never say anything negative. And we couldn't talk about specific gambling. We couldn't talk about DraftKings. Right. We had to specifically talk about FanDuel, but we also had to talk about, like, how to make a peanut butter and jelly sandwich. How is it fun? How can you, like, go to a taco stand and order churros?
- [40:25]
A
Right.
- [40:25]
B
We wanted. We wanted it to be an entertainment piece, not a restrictive thing that just talks about sports. And at the beginning, we're all worried about this, but right towards the end, we're like, oh, we can put Chuck here. We can put him in a commercial. We can add this type of language to it. We can, like, make it fun. And those ideas. And everybody came together as an entire group. And when we got on Reddit, we had 3 million people that were talking and chatting with Chuck, all in positive connotations. And that's when you're like, this is the right place for AI. And I know it's a fun entertainment piece, but it enabled people to have human connections with this really fun person that was based off of Charles Barkley, who's a phenomenal person. And you're, like, just experiencing the human connection between a robot, a human to a robot, to a human. And it's just fun. And it's. It's. It makes life good. Right. So that's why I appreciate it when teams come together like that.
- [41:15]
A
Yeah, that's a great point. I can certainly concur that once they kind of get past the skepticism and they're like, wait a minute, it just did what exactly? Show me that again. Yeah, they get real. They start using it at home.
- [41:29]
B
Yes. Right.
- [41:30]
A
And that's when you've got a winner. So one of the things that comes up a lot with companies is, well, we don't hear it as much, but we still hear it. We don't allow our employees to use AI.
- [41:41]
B
Mm.
- [41:42]
A
It's a risk.
- [41:43]
B
Right? They don't know they already are.
- [41:45]
A
Exactly.
- [41:46]
B
Not even generative AI. They don't know that they're already are using AI. Inside of the software tools they're already using.
- [41:51]
A
Yeah. So how do you, if you have a client or a prospect that has some resistance to the governance side, the security side, what's kind of your go to narrative around that?
- [42:06]
B
Most questions come of where is my data going to live? And we have to educate them that using generative AI, especially in the enterprise version of OpenAI or Microsoft or Google, your data is going to be destroyed the second it's prompted and that data is not going to be used to train the model. Now, is that true? I hope so. I hope these models and these frontier models, they abide to that rule that they put in place. And I hope that that's actually happening and we have run tests and seen that it has been destroyed. But as a company, you really should start thinking about, if you're an Office365 person. I love this scenario, right? Like, I was talking to a company and we were talking about the security aspect of their mail, and they wanted, they used Google as their workspace. And they're like, we have all of our employees right now copy and pasting from ChatGPT to help them write their emails, and they want to stop that. And so they said, we connected ChatGPT to Google using an MCP, which is a way to communicate between ChatGPT. And now we have employees asking a bunch of questions, but we're really worried about ChatGPT kind of crawling our emails and knowing things that we don't want them to know. And we tell them it's protected. And OpenAI does a good job storing all that and protecting it and making sure that it is security grade. And then I suggested, like, why don't you just use Gemini? And it's built into workspace, it's built into Google. And they're like, well, we're going to have the same problems. We're not going to trust Gemini to secure and monitor our data. And I said, but it's Google, right? Gmail is already known by Google. Every email you send, everything you send inside of that workspace is known by Google and Gemini is Google. So you're staying inside of the Google ecosystem. And if one thing was to happen to one company, whether it be Google or OpenAI, you're probably protected. If OpenAI all of a sudden leaks everything, because you're in Google now, if you're in two places, you have two things to worry about. And that's what we have to start thinking about, is what are the tools in our ecosystems that we can latch onto where the Data is already known and we can use the same tool. It might not be 80% good, it might be like 90% as good as OpenAI. But I don't have to worry about, like logging into two places or securing two places.
- [44:21]
A
It's a good answer. So when you say it, it's destroyed, essentially. You guys have been able to, because you're approaching it from more technical side, you're able to kind of like witness that this is not being stored on any type of server or anything like that.
- [44:36]
B
Yeah, that's correct. And we've read the documentation and the terms and conditions and stuff, and especially when we get into bigger companies that are requiring this. Yeah, the prompt, I mean, you remember the whole purpose. It's predicting the next word. Yeah. We don't care if that next word is a float number that's 95 digits long and we have to convert it or if it's literally the next word. But the data that is fed to get that next word is what's important. Right. And that's what the company's trying to protect. So there are even ways to obfuscate that so that we can change the concept of what is being sent out. Extract the pii, extract the information and send the essence. Yeah. To the prompting to say this is the type of response we're looking for you to generate, and then gathering that information back and destroying everything along the way. So there's, there's a bunch of methods that we've attempted throughout the past to make sure things can stay in the walled garden.
- [45:25]
A
Well, Andrew, this has been, you know, a different talk about the agency side of things, because most of the stuff that we do is we're not providing those solutions, at least in our business. Right. So it's been interesting to hear the differences, but also the similarities as well. And some of your experiences are very much in line with what we're seeing out there. So for anybody listening, I would say, look, imagine Andrew and I both have lots of conversations with experts in the, in the space and, you know, the intel that I'm getting from the field is that what you're hearing here today is pretty spot on with kind of the paradigm of what's happening. So it would have been even more interesting if we had completely different views on things. But it is good because it means that, like, we're, we're heading in the right direction. Andrew, who, who is, who is an ideal person to reach out to you guys about a solution?
- [46:19]
B
I'm the only salesperson in a company still. We've built 150 products over the last nine years now and we have a great product team, great engineering team and you can find me on LinkedIn. I post on it every day. But if you have a specific question or you want to see if it's a fit, you can go to our website, check out our portfolio. We have a lot of case studies. Our engineers write a lot of really great papers about what they're building as well. So you can check those out and see if there's a match of what you're thinking of. And if there is, we'd love to build it.
- [46:45]
A
Nice. Very cool. And all the, the links to Andrew's LinkedIn profile, the website, all that kind of stuff are going to be in the show notes here, so should be easy to find. Andrew, any, I don't know, closing advice or remarks for the non technical business professional who's listening to this?
- [47:02]
B
I'll say that there's a lot of people you can follow in social media. Stop following the frontier model leaders, the Sam Altman. It's cool to learn from them, but they have their own goals and their own profit sites of what they're trying to achieve. There is a lot of people in the workspace automation place and the AI place that have been here for 10, 15, 20 years. I mean some of these individuals have been the birth of AI or the grandfather of AI and I don't remember all their names but they exist and they're talking about real solutions for real businesses, not AGI. They're not talking about removing developers, they're not talking about the end of the world. They're talking about how businesses can make a bunch of money using agents. And that's where you should be getting your knowledge from is those types of individuals that have seen it before, have done it and are experiencing that this is the world we're starting to enter into, not AGI, because AGI is not something that's going to help your business anytime soon, maybe not ever help it. Yeah, yeah.
- [48:01]
A
So great stuff. All right, awesome. Andrew, thank you so much for being a guest on the show. And as I mentioned, we'll be getting all this information out to everybody who's listening to this, the links, the anything that was referenced in here so that you can kind of dig a little bit deeper on what we've talked about today. So thanks everybody. We'll see you on the next episode. Thanks for tuning in to using AI at work. Don't forget to subscribe for more conversations about how to use AI at work and a special thank you to our sponsor, Chief AI Officer for empowering businesses with AI education and training. Visit their website for a free AI readiness assessment and AI strategy guide to help you get started using AI at work. That's www.chief aiofficer.com. so thanks to our producer Evan Desaunier for making this episode possible. Follow us on Twitter at the handle Using AI at Work and visit www.usingaiatwork.com for free resources to help you harness AI in your role.