Ryan Conti (00:00)
Best and most powerful users of RAI chatbot are just like, go find me some money. This

Product's doing great, but are you aware that you're actually losing money on your whole foods contract? And maybe that's intentional, but if you think about repackaging it this way, then it can start to give you these recommendations and you can have this long running conversation. I mean, everyone I'm assuming has used Chat GPT or Claude in some capacity. Totally. I mean, all these models are goal-oriented, right? So at the end of the day, they try to figure out what the goal is, or you tell them explicitly what the goal is, and then they find any means to get there, right? And that's obviously like the AI doomerism is like, well, if you tell it you want to like never

have anyone be hungry and they're like, Well the easiest way is to eradicate humanity, then no one will ever be hungry. You're like, God, that could go really wrong.

Jordan Buckner (00:42)
Building a CPG brand comes with thousands of questions you never thought you'd have to answer. How do I find the right co-manufacturer? Which 3PL can actually handle my product? How should I manage lock codes? What happens if I want to change my recipe? And for years, founders have been asking these questions by messaging other founders and groups, posting in Slack, or asking around until someone kind of gives them the right direction. Recently, tools like ChatGPT can really help, but it's also, you know.

Confusing if you're actually getting the right answer or just what you want to hear. So today I've invited on Ryan, who's the co-founder of Peasy, talking through a slightly different approach of how he's been able to help train a chatbot kind of using the Peasy insights and understanding supply chain logistics, partners with CPG brands to be able to get you the best answer to your questions. Ryan, welcome to the show.

Ryan Conti (01:37)
Thank you, Jordan. Happy to be here.

Jordan Buckner (01:39)
So tell me a little bit first around the work that you've been doing with Peasy because I think it helps inform where you're coming from with how you're thinking about answering these questions. Yeah.

Ryan Conti (01:47)
Absolutely. Peasy was born out of this journey that my co-founder had building a mustard brand. He spent 10 years building this brand. He went looking for software to track all the physical inventory through its life cycle and just manage the day-to-day operations. He quickly realized that there wasn't a lot of great options out there. they were either too expensive or required dedicated resources to building it out. Maybe they were overbuilt or just didn't really speak to him.

And so we set out a few years ago to build the software that we believed always needed to exist, custom build for the CPG industry. and then through that journey, yeah, there's a couple of things that became very clear. One of which was that because there's it's super expensive, you need to find a different way to make money while still giving this product to people and getting it in their hands earlier in their journey. And so we gave the product away for free. We still give it away for free. We plan to always give it away for free.

we make our money on the payment transactions. And part of that is forced us to make difficult decisions in building the product. And one of those is how do you help people onboard to the product? How do you support them? And what we realized is that there's new tools out there like AI. And we layered on a chatbot. You know, at first it was basically just a glorified version of Claude. Now it's much more sophisticated. And in order to make it useful, we had to train it on.

Three different pieces of data. so we trained it, well, really, we trained it on two pieces. We trained it on how Peasy works, and then we had to go train it on how CPG works. Cause people are asking nuanced questions with these acronyms that are very specific. You know, you're like, what's a COA, what's a B O L, what's SQF? And it needed to understand all of that. And over the last several months, the thing that's been very, you know, frankly, a little bit surprising is just the sheer quantity of use.

Of this. And over time, we've seen brands once they use it to help onboard and get it get acclimated with the product, they just start asking it operational questions. They start asking it, you know, how do I find this command? How do I think about getting this certification? And we realize that there's a lot of value to this custom trained product that is Chat GPT or Claude, but it's routing to the right models with all the right context.

Tailored specifically for this industry. And so we've just continued to lean into adding AI to the product, as well as now we're we're launching a product that's standalone that's going to be a chat interface, chat GPT style, but again, trained specifically on this industry for these particular users.

Jordan Buckner (04:30)
Yeah, I think that's really powerful. And, I've always been of two minds to this, right? Like a lot of AI tools are fairly widely available now. Most people have a subscription to Chat GPT or Claude or something. And it's good at answering questions, a lot of general questions, but it gets better and better with context. And part of that is training it on like your individual company data and information that is not widely available on the internet.

The other part is maybe doing a little bit of filtering for the internet, right? There's so much out there that it can only synthesize like so much information at one time. And so like what's the specific type of information or answers that you're gonna be looking for? How is it relevant? What are the sources that it's using? Because if you ask around like, you know, who's the best co-manufacturer, there's thousands of co-manufacturers, there's different products, there's different ones, like how do you determine the best? Is it reviews from this site or that sort of aggregate, right?

And so by limiting down and filtering into a certain set, you can get more precise and better answers. I think that's really important. I'm curious how you're finding feeding in the right amount of context and information where it's able to provide like the best answer possible versus like using someone's generic chat TBT. Like, why should they use like this versus something else? Because like what what's special about the context you're adding in?

Ryan Conti (05:50)
Yeah, I think that there's there's two things that I would say is if you're using Peasy to manage your operations, then your agent has access to your data. We don't train anything on that data, obviously. It's incredibly secure. But when you start asking questions, it has that context about your business. And I think that that's one of the magical things that AI has is its ability to process lots of contexts and provide very tailored answers. And then the other thing that we've noticed is that.

the the the depth of question is actually quite great. And so when I look at, you know, the questions that are being asked in some of these larger communities or on Reddit, they're so multi-layered. It's not like, hey, I need a coman for my product. It's like I need a co-man that does frozen food that specializes in these locations, this packaging type, and has like, you know, the ability to meet this small production run.

And has these lead times. And it's like to be able to process all of that information, you actually need to have not just the right contacts, but you need to train the model and in how to think about breaking down that question, answering all the pieces, and then evaluating that answer to make sure it's giving you the right response. and that's what we used to do with Google, right? Like you'd be all right, this is what I need, and then you'd pull up 20 tabs and you'd go through each one and you'd X out of them and you'd keep the ones open. That can all be streamlined.

And it's not just about the power of the model, but it's like what are the parameters and the structure and then the evaluation layer on the back end to validate that it actually gave you the right response.

Jordan Buckner (07:27)
Yeah, I think that's super helpful. And then, I'm also curious, right, there's the general industry questions and then there's specific questions that every founder operator has around their business and how it's performing. So how are you incorporating the AI into PZ's operations, right? To give specific and detailed answers about your brand. The thing

Ryan Conti (07:47)
That's

very clear about AI is that it's it's going to be pervasive if it isn't already, but people are really sensitive with their data. And if you're not on these enterprise contracts with some of these large providers, you're giving them your data and they can do what they want with it. And that's really uncomfortable for a lot of people for good reason. And so with Peasy, what we're able to do is bring the model in-house, reference their existing data.

And then answer those complex questions. And so once you're using Peasy, we have an AI chatbot built into it. function the best chief of staff or COO you've ever had. And the questions you can ask can be much more open-ended than you used to be able to ask when you're like pulling reports. So in the past, you'd be like, I just want to see the price of this thing over time. And then I have to qualitatively figure it out. And now the best and most powerful users of our AI chatbot are

Just go find me some money. Right. I'm I feel like it's like, tell me if I'm pricing my items correctly. And it'll go through and it can solve that question for you in such a high level way that you actually get better answers when you don't constrain the model and you let it use its own like computational power. And it'll go in and look and it'll say, Hey, you have these items and they might look like they're priced correctly, but the inputs to them.

Have these wildly varying commodity prices. Or your prices make a lot of sense based on the input cogs. But when you start factoring in your transportation costs and your labor costs, you're actually a little bit underwater here. Or you've adjusted your pricing down for because you can have custom price lists in Peasy. And it might tell you like, look, this product's doing great, but are you aware that you're actually losing money on your whole foods contract? And maybe that's intentional.

But if you think about repackaging it, this way, then it can start to give you these recommendations and you can have this long-running conversation. everyone I'm assuming has used ChatGPT or Claude in some capacity. But when you have your own data in there and it has all of the transactional history and all of the recipes and the bill of materials and the bills and the invoices, it can tell you the like truth based on the actual transactions rather than

when you're giving it like, Well, here's my recipe, tell me if this is good and it doesn't know all the other context around it.

Jordan Buckner (10:10)
Yeah, I think that's super helpful. One thing I'm trying to find more experience with and curious about your interaction with it is a lot of the AI tools right now are built to be responsive, right? You ask a question, it'll go out, get an answer, you can kind of clarify. I'm curious if you're seeing ways that the AI and agents are becoming more proactive, where instead of asking it like, hey, how can I improve my pricing?

is actively coming to you and saying, hey, notice this month that our transportation costs were high. I looked up this other these other providers and we should go with these options moving forward and giving you the information as the CEO or operator versus you always asking and knowing the right questions to ask.

Ryan Conti (10:54)
Yeah, it's it's such a hard question to solve on a product level because I've, you know, we've experimented with tons of AI products and some of them are a little like overzealous. And you're like, dude, leave me alone. Like I don't want you constantly telling me that you can do this thing that I don't want you to do and and it becomes noisy. And so we've tried to take this approach of you still have to initiate the effort, but how do you then set these things on autopilot? And

We see, we see people will go and be like, Hey, how can you help me? What are some of the things you can do? And then that is like the seed that spawns some really incredible idea. And the the best users are they're identifying an opportunity, specifically a recurring opportunity that generally like they just can't do because it's too, you know, cost prohibitive or time prohibitive for these busy founders. And then they're setting it on autopilot. So it'll be like, Hey, every week review all of my receiving costs.

And tell me what's increased in price and where you would recommend or pick two things that you recommend I go try to find another supplier for. And then you can even tell the agent to go do that. And we have like a whole RFQ process where it can, you know, pre-populate everything. And we're big believers of human in the loop. Like we don't want to start spraying emails on your behalf. And, you know, I just think your business is too sensitive for that.

But it can prepopulate everything, be like, okay, cool. Here's the suppliers we recommend you reach out to, send them this email. It's gonna capture the the price today. And then we can identify if there's a new supplier worth onboarding. and the the other thing that I've found with agents in even my own use cases is that agents really want to be helpful, almost to a fault. And so you'll ask it a question and the

The success of your answer and the quality of your answer is so dependent on how good you are at at explaining and providing the context for that question. And the thing that's always frustrated me is I'm like, if I were to come to Jordan and say, Hey, Jordan, I need some advice and I were to ask you a question, there's no shot you're giving me an answer without asking me two or three follow-up questions to get to the heart of what I'm getting at and clarify. And so we've really tried to train our agent on, be be a little bit patient, right? don't just

Fill in the gaps to the best of your ability and provide the answer you think because it's how we so often get these unsatisfying answers. And, you know, ask, be comfortable asking one or two follow-ups. It's no different than like coaching a young employee and you're like, get get to the heart of the issue because generally what you're, you know, what they're really trying to solve for is like, how do I improve the margins of my business? How do I improve the operations? It's not just as simple as like, how do I get a cheaper price? And

Then you can start to understand okay, well, here are some trade-off decisions. Would you ever consider it not being organic? Would you ever consider longer lead times? You know, would you consider buying in bulk? And maybe there's opportunities with your current supply chain to improve your pricing that isn't just like go find 10 other vendors and start to query.

Jordan Buckner (13:59)
Ryan, that actually made me think around this tension that's happening with AI around what AI and agents can do without you versus bringing it down to like the speed of human level, right? Because like theoretically, since if you're in complete all these computations and faster than you can, like it could send hundreds of emails out, it can make all these responses, it can do all this sourcing.

But a lot of your times you're like, wait, wait, slow down. Right. let me check. Let's make sure we're doing it right. Let's make sure there's all these other context and nuance pieces. And there's this weird bounds of let's slow things down to human level, which I think is also important with business right now, to say, all right, how can we have more control and more say and make sure it's done right and not just done fast? And it's interesting to see how that's starting to develop. And I like your

analogy of let's have it ask questions back to make sure it's fully understanding the context versus assuming. That's another area of AI which has gotten a little better. I think Claude even used to do a better job in Chat GPT where it defaulted to asking more questions versus just giving you an answer even if it was like an incorrect r answer just in order to get something to you. And I think that's something that is a tension that's starting to happen more and more is like how do I make sure I give

the know the context fully to give you the right answer. And that's very important with business as well, because that also develops trust, right? when you're talking to somebody, as you mentioned with the example, having a dialogue, having questions back and forth before giving a final answer is really important to actually building that trust and rapport that it in the AI like knows what it's talking about, just like you would a a human member on your team.

Ryan Conti (15:39)
Totally. all these models they're they're goal oriented, right? So at the end of the day, they they try to figure out what the goal is, or you tell explicitly what the goal is, and then they find any means to get there, right? And that's obviously like the AI doomerism is like, well, if you tell it, you know, you want to like never have anyone be hungry and they're like, well, the easiest way is to eradicate humanity, then like no one will ever be hungry. You're like, God, that could go really wrong. But I think, you know, the thing that we've even learned just building the product is like

If you're using AI to do anything, spend three to four acts as much time planning. And then you don't have to get involved as much and iterate as much and you get the right answer the first time. And it's no different than working with a consultant. Like if you ask if a if a consultant you ask a consultant one question, they give you the answer, it's like they're probably a bad consultant. Any good consultant's gonna make sure they really understand the goals and then and then enter the work mode. And the the more time you spend on the planning mode.

The better your result is. And frankly, the faster you're gonna get the information you need and be able to move on. And so we really tried to tune our process to to mirror that so that you don't have to be an expert in prompting to get great value.

Jordan Buckner (16:54)
Do you ever see Peasy becoming not just the software center of your operations, but the actual center of operations where you can give all your operations to Peasy if maybe your code manufacturer and actually producing the stuff yourself and it manages all those relationships? Or do you think that for a long time now there's still gonna be like a person that needs to go in and interact and and affect all the data versus

It just being the centralized database that tells you things.

Ryan Conti (17:25)
my my general philosophy is if you if everyone's running on AI, then like nothing ends up being differentiated whatsoever. And there's still just such a human component to all of this. And there's so many variables to running these businesses. And you can't rely on AI to do all of it. Like there's just times where you have to call up your supplier and be like, Hey, I know I missed the order deadline. Like I really need this thing. Can you help me out? And if you don't have a relationship with them, like

Your agent's gonna say that, their agent's gonna say no, and you're just gonna be stuck. And so I think that, you know, for us, our belief is these are still human-run businesses. They'll always be human-run businesses, but there's so much of the work that is somewhat undifferentiated and is highly manual. And we just want to take all that off people's plates. Like we want, you know, these founders who are creatives and operational experts to focus on

Building and creating the best product and packaging it in the most compelling way and finding people who want to buy it. Now, can AI help you with more and more of that over time? Absolutely. Can we start to identify, you know, like one of the things we're working on is the ability to leverage your relationships to say, hey, I have these customers that I work with and you have these customers and we have similar products, and I could reach out to your customers.

But I might not get a response. But if you reach out to one of your customers and say, hey, this brand is great. You should check them out. They're probably going to respond. And so how do you identify who the right customers are, who in your network can help you get introductions and then reciprocate that so that it's mutually beneficial? Like there's that sort of body of work that I think it just takes so much effort today. And a lot of people either don't have time or don't even know how to approach it. And that's the stuff that I think AI can take off your plate.

And and that's the direction we want to go is just remove all the tedious work so that you can focus on the creative and the relationship stuff that's still inherently a human and and I believe will probably always be inherently human.

Jordan Buckner (19:34)
Very exciting, Ryan. I like that approach and I think that's going to really be beneficial as you continue to grow Peasy. Thanks so much for being on and talking all things AI and how it intersects with operations and supply chain. you know, Peasy I always recommend as the best platform to get started with and and transfer to to grow your business and manage your operations. Anyone listening to this, if you haven't checked out Peasy, we'll put the info in the show notes. Definitely do so to manage better operations for your business. Ryan, thanks for being on.

Ryan Conti (20:03)
Thank you, Jordan. Always a pleasure.