Richard Jacobs: Hello. This is Richard Jacobs with the Future Tech podcast, I guess is Adrien Schmidt, CEO of bouquet.AI like a bouquet of flowers. That’s the website for bouquet.AI. So Adrien, thanks for coming. How are you doing?
Adrien Schmidt: Hey Rich! Thanks for having me. Because they’re like a bouquet of data applications actually.
Richard Jacobs: Oh, that’s true! What about a Cornucopia or was that too long of a name?
Adrien Schmidt: That could be something, does it even fit into Twitter in a tweet? Well, actually yes you come out with related products.
Richard Jacobs: Maybe you’ll call one the cornea copy of version maybe that’s the professional or industrial version of bouquet just to finish up a bad joke.
Adrien Schmidt: But yeah well actually we’re talking to more and more about the naming for a product which is Aristotle, maybe LinkedIn itself.
It’s quite a mouthful.
Richard Jacobs: Okay! All right. Well, tell me about the bouquet.
What’s the premise of the company?
Adrien Schmidt: Yeah! So we work in natural language processing and we use AI algorithms and NLP to help people get answers to questions they have. So people in companies we work in are in the Meta B space, and basically can you just ask your data a question and get an answer? Imagine it kind of Siri for data so you have for databases and Apps and just ask the question and get an answer.
Richard Jacobs: Oh yeah. I mean right now you got to take the data and put it the spreadsheets and pivot tables. I’ve had the analysis statistics. So what kind of questions are you able to answer or do you want to be able to answer right now?
Adrien Schmidt: So we’re focusing on databases, right? So exactly the kind of questions that you’ll have in spreadsheets, in reports and dashboards.
So imagine you’re a salesperson or a sales manager. You want to explore opportunities for your team, how much they’re worth, how a specific salesperson is doing, if you’re more on the financial side, if you want to collect and round up all these numbers on your division part, but we’re also looking at some really cool operational use cases. So imagine you’re in a store and you want to serve its customers better. Can you look up inventory? Can you look up product descriptions simply by using voice command? And that’s really what we’re trying to make, all this data accessible super easily and in the fastest, simplest way by using a voice which is probably our most natural way to interact.
Richard Jacobs: So this is B2B for a back office to have a customer service people or warehouse people or it would be the specific users of it?
Adrien Schmidt: Yeah, it’s definitely B2B right now and so customer service reps, we have a use case there when they need to lookup information very quickly on a specific customer and our use cases, people who need to know how a customer has been using a product when they have them on the line. So analyzing usage data on the flights and saying, hey, how much of this has been consumed? Can you split that by-product? What was the growth year to year? How about for this specific product or this specific region? And you know, really being able to answer those questions on the fly, imagine that without Aristotle, which you have to do is browse through a bunch of dashboards or reports and if you don’t want to do that, you have to call someone up to do it for you, which is not the most cost-efficient way or even the fastest way possible and do that.
Richard Jacobs: So we try to do it in real-time?
Adrien Schmidt: Yeah! Exactly right. Unless you know it by heart, which would be awesome, right? If we just knew all these numbers on the top of my mind but if it’s not the case, then imagine if you could just ask someone or virtual someone, which would be Aristotle and hey,I just need this number and Aristotle knows it on the fly and just gives it back to you instantly 24/7 wherever you are and through any kind of channel. So I talked about the voice you can do that through Alexa, but if you’re in an open space and you feel kind of uncomfortable talking to your computer, then you know, you could just use slack or Microsoft teams or you know, whatever instant messaging system your company uses to get the answer. We even support.
Richard Jacobs: Oh, that’s consumer channels.
Adrien Schmidt: Yeah.
Richard Jacobs: So how far along are you with this? Is it in use? Is there any specific use cases that you can talk about?
Adrien Schmidt: Yes, so we’re in and what we’ve done so far is built out the technology and started to run it out as a better product with a bunch of companies. So mostly pretty big ones and we’ve seen a pretty wide variety of use cases. So in one case it’s really for an internal use of a big data warehouse of stuff and you have a team of analysts who answer questions for marketing from engineering from quality assurance and they get questions all the time on their data and the fact is that a high percentage of those questions, and I think the commonly agreed upon measures about 60% of those questions are pretty basic right?
Not like super sophisticated stuff that you need to spend three weeks analyzing but actually pretty basic stuff and things that are totally automatable with a question and answer system. Like Aristotle, it’s just that the users don’t know where to look for the data. When they do know how to look for the data or where to go. They don’t necessarily know how to query the data. And there’s not, maybe even if they have dashboards and reports, they don’t even know which ones to find or to get or another explanation is that a lot of them are just kind of lazy. No one can do it. Right, so to answer that Aristotle provides the helium that users can have talking to someone, a real person because they just asked basically their question and Aristotle will engage in a conversation to try to understand what they’re seeing exactly and provide an answer.
Richard Jacobs: Interesting! You got to the point where it’s listening to let’s say the customer service interaction or the salesperson interaction and it’s prompting and suggesting metrics and data based on the context of the conversation. Like customers are asking Oh the book, well this pile be able to carry a hundred PSI of whatever’s, 600 pounds steam and okay the AI looks from the database and says our piles are rated to blah, blah. This, this size that prompts the salesperson though the customer service person with the right data, without them even asking,
Adrien Schmidt: Oh man, wouldn’t that be cool? Exactly. And I think this is we’d love to do that and it’s our roadmap. I think you just nailed it. Exactly. Basically what you want to do our end game right, is what we call casual data usage.
So having data being used and being accessible at any time of day whenever you need it right. And anything we can do to remove the steps that take to get from you to the data you need to support the conversation you’re having or the presentation you’re making, review email that you’re writing or even the thought process that you’re having. Anything that you could do to remove those steps goes in the direction of what we want to do. So what you just suggested there is obviously awesome because you don’t even have to query to ask a question. You get the answer just from the conversation, you get the prompts just from the conversation. So I think you really got the idea there.
Richard Jacobs: I think you’d be able to figure that out actually pretty quick because if you listened in on a thousand CSR calls, there’ll be an 80, 20 there all right. These same issues seem to be coming up over and over and CSRS best answered them in this way. So you could probably figure this out pretty quick by doing that. Do you know what I’m saying? They like knowledge tuning the script.
Adrien Schmidt: Yeah absolutely! What you’re seeing here is a, I mean you’re describing this casual data usage because then CSRs could get this insight when they needed and have a much more meaningful conversation. Even from a customer perspective, you’re talking to someone who’s immensely knowledgeable about what you’re, what you’re telling this person and can provide the right feedback and provide the right insights and advice and how to sort your personal situation. So, so I think I love it, I think it’s great.
Richard Jacobs: That’s really funny. You should say your Avengers’ end game goal is to eliminate half of all questions.
Adrien Schmidt: Yeah, yeah. Right, exactly! I’ll put that in.
Richard Jacobs: Yeah. Yeah. I don’t know if, uh, if they’ll like that as throw stuff at you and say, stop with the bad jokes, but you know, it’s there for you if you want it.
Adrien Schmidt: All right, I’ll take good notes.
Richard Jacobs: So how far along are you in this process? And I’m, you’d have, you have it in use live in certain test cases. You know, what’s been the experience of the person providing the service from the company side, you know, that CSR salesperson and what’s been the experience from the customer side? Do you have any data or any codes from that?
Adrien Schmidt: Okay! Yeah, well as I said we’re in late data, so we’ve been working with companies on actual rollouts. I would say what are some fun insights? Because a lot of the insights that we got is, how to work on this kind of thing in an enterprise environment with all the security gone through the Charity. How to make actually match on enterprise data which usually is not super clean can be pretty complex. So aside from the technical stuff a lot of people love the voice, right? They always want to have this use case of voice, but in practice when you talk about the rollout, we’re still using text as the input method.
It’s the same question that you can ask. It’s a natural language. It’s usually a short question and then the dialogue with a follow-up. But we’re still in the reality of things they’re still using texts. So even though we design our questions and our dialogues for voice, it still kind of texts. So that’s the kind down to earth kind of thing. The other thing is that we were a little bit hesitant to talk about simple simplicity, simple questions, and things like that. We always felt that we would be kind of like the company coming in and doing simplistic stuff and yeah, you can do voice, but it can be really simplistic, In fact, that was totally the wrong way to see it and simple is good. Right!
So our concern about being simplistic was like, we want to do basic stuff because you want that experience that looks like search, right? When you search for you use the instant you don’t think about it, you know exactly how it works. You don’t have to ask yourself a bunch of questions. It just works. Right!
Richard Jacobs: So when you need some plastic, you mean, okay, be bold. Wouldn’t lie to be forced to ask the question in a simple way. They rather just ask it in a technical wobbly book, and you know their internal company.
Adrien Schmidt: Well, I mean we’re in the space which is analyzing data, right? So you have simple questions. Some questions are really simple, like how many items did we sell last week or did we ship last week? Right? That’s a really simplistic question or a simple question compared to what is the forecasted variance of our shipments and stuff like that, and we were a little bit you know saying well let’s start with the simple level questions and hope for the best. There’s a lot of questions like that going around.
Richard Jacobs: Do know why that’s better, because if you’re requiring people to talk in that way, then when they turn around and talk to customers or even internally, it’s going to be clearer communication. If you forced them to do it in the simple layman’s way, it that way.
There’ll be no miscommunication and again if people used to talk really technically, then they get on the phone with a customer or whatever it is that they come back and come at them with all this internal acronym laden, you’re getting Gobbledygook. The person is not going to understand and they’re not going to do well. And so this was kind of forced the whole organization to go in the right direction anyway by doing this.
Adrien Schmidt: Yeah! And let me give you an example here of a question. Even if it’s not gobbledygook. How about this question? So you’re talking to an executive and executive says, why are people leaving? So it’s like talking about his staff. Why are people leaving, well, that’s a tough question, right? I mean, why are people leaving and you imagine asking me a system, why are people leaving?
Right! Well the first question, follow up question is, are people leaving? Right? So how many people left last quarter last year? Is it growing or done? Okay, so 312 people left. How many left last year? Well, 280. Okay, so actually fewer people or more people are leaving. Right. In which department and then you have this broad question of it seems untakeable but in fact, you could sub-questions that are themselves really simple, not simplistic, almost simplistic, but I mean really simple. But that gives a tremendous amount of insight to answer the really tough question, which is why are people leaving? You see, that’s the kind of thing that we discovered is that a really past problem could be kind of addressed with simple questions. As long as you make it easy, fast enough to get the answer and that’s what we call casual data usage.
You know, why are people leaving? Let me just for a sec, how many people left? Where are they left? Is it increasing? And then I can decide what to do about it.
Richard Jacobs: It sounds like that this really needs the component of an expert system to it. You can’t just take this and said I will use it for our company and we saw like, I don’t know carpeting or something. It really sounds like it needs some to collaborate unless the learning can be done on its own. I don’t know the AI, if it’s strong enough, but it sounds like you need to sit down with the people and go through the issues of the possible issues they run into and kind of tune it the AI so that it can prompt them with the right things to ask if they’re not asking it and take it to the next level beyond just allowing them to ask whatever question they want, you know?
Adrien Schmidt: Yeah! And so let me rephrase that a little bit because we are a that is a really important component of what we’re doing, but it’s not exactly, I think the way that you’re putting it there yes we need to sit down with people but it’s not necessarily because we need to tune the AI a lot. Okay. Our technology is basically to build an AI model and natural language understanding model on top of any proprietary database. So this is where we put a lot of effort and I think we reached a certain point where we can turn out a lot of models, right? And they’re pretty accurate and understanding questions related to a specific database. However we want to make people, we want to influence people’s behaviors to use data in a casual way, meaning increase their usage of this data.
If we make it easier for them to get it, then we want them to actually take the action of going and substantiating their conversations and what they’re doing with data. So that brings the concept of triggers. What kind of triggers do we have to remind people that this is something that they can use data to support, right? This is a question they’re dealing with either a situation or a time in the company or whatever they’re doing right now. Well, you know, if they’re on a call with a customer or whatever, they can use data to substantiate what they’re doing right now. And so understanding what those triggers can require a deep understanding of the company, and so it’s not really so much of a technical issue because triggers are pretty basic, but it’s really interesting, thing from a behavioral standpoint because triggers will induce actions, right! Which are to go and get a piece of data when you need it and until it becomes a habit.
Richard Jacobs: How hard is it to do what you’re doing and then how difficult is it to understand questions in their context and the relevance and the person’s really asking for?
Adrien Schmidt: well I’d lie if I’d say it was super easy because there’s just a lot of components that need to work together. Right? So first of what is Amanda, what kind of questions can we answer? Right? So what’s the language structure? Then there’s identifying what is inside of the user’s utterance. Basically what you said what is inside of what you said? So that’s a lot of semantic stuff and so there’s this whole first part which I need to understand what you’re doing, what you’re saying.
Right! Then there is a big second chunk, which is to say okay, now that I understood what you said, where will I find the data that basically answer your question and how will I query the data source to actually get back to number? Right? So that’s then the query part. And then when I once get the result, I need to package it in a way that is understandable. So we basically provide you an answer that should be able to transform into the text to speech. So traditionally you get answers in the format of charts like a pie chart or whatever chart or even a report or table you’d give answers in the form of an answer, right? So packaging that making all this work together is well it took us a lot of tweaking and then making that work all together in an enterprise environment with security, confidentiality was pretty tricky.
Richard Jacobs: And I would think you’d have to do a review of the database that you’re pulling everything off of like one of the different fields are mislabeled or missing. Can you make any suggestions early on, hey, this, its two data fields are really ambiguous? I think it’s going to hurt you in your ability to get one full outta here. We don’t have this on the customers, you know, it’s missing and you should go get that.
Adrien Schmidt: Yeah! So understanding what we call the Neta data of the source databases, basically the data that described the data. So I don’t know the names of the tables, the names, the columns, and the means of the veterans that’s a really good point. So how do we map that to what users say we have what we call synonyms, right?
So let’s say there is a field called revenue. Maybe people use different terms for revenue. Maybe they have acronyms that they use internally and that they feel comfortable with. So we found that the most efficient way to add vocabulary on top of a database was to pretty much crowdsource it, right? So when someone asks a question using a term that you’ve never heard before, we usually say, well either we can suggest something with you know, but with a sort of semantic match which you know we’re still working on or we say well, can you search within the stuff that we have do the match manually and then we’ll learn from that. Right? So with this crowdsource way of building our vocabulary we’re able to automate some of that, some of that without putting too much of a toll on users because once they done it tries to learn pretty quickly.
So if you ask the same thing again it’ll last time you said this or do you mean that? Or you want to look for something else. So that’s the way crowdsourcing is a great way to make a system like this really work, which means that it’ll be a better and better experience over time. Right! You shouldn’t expect to just download it, install it and it works right out of the box. I know you downloaded, I mean, you don’t download it. Basically, you start using it and then it has to learn and gradually it delivers more and more value
Richard Jacobs: How do you deal mechanistically with people’s voices? Soft resonant, different accents, Pronunciations. Yeah. Like for instance, for some reason, I have a very hard time with like Google assistant or Alexa or even dragon naturally speaking and they just constantly put the wrong words. I had to kind of speak like a robot, you know, do you have this thing here and how sophisticated is your system? Have you been able to listen to pretty much anyone or is there certain speaking styles that are difficult?
Adrien Schmidt: Well, first of all, I have a recommendation for you is that we tend to speak in a robotic way to an Alexa or Google home but in fact, you should go back to speaking naturally because they’re pretty good actually at understanding your common speech. Maybe when you started off using it had a hard time but you should try it. Sometimes when I speak to Alexa in too much of either a slow pace or I tried to sort out the words a little bit too much, it doesn’t understand. Whereas where I just try to speak naturally gives its better. So we do tend to speak to a robot and like a robot, right? So it’s Kinda like when you talk to her,
You talk to the child and you start speaking in this specific kind of childlike voice and sometimes I don’t know if you’ve ever had a kid look at you and say why are you talking to me like an idiot? You know like, no, I’m not.
Richard Jacobs: My daughter actually said to me, she said dad why do you talk like a robot to Google? I said because it seems like that’s the only way it understands me. She noticed that she told me about it two weeks ago.
Adrien Schmidt: Oh, there you picked up on that? Yeah. So well cause that just confirmed it, so I try to be natural and try to speak to them like a move.
I do that too, so right now for this part, we don’t directly address this because we actually use Google Assistant or Alexa, right. Or Siri as our voice to text a translator basically.
Richard Jacobs: There’s such sophistication in what you’re doing.
Adrien Schmidt: So, first of all, let’s leverage technology that other people have built and probably done a better job at it than we would do. The other thing is that Aristotle is supposed to be kind of this virtual assistance to help you get access to data, right? Or virtual data analysts, you know, at your side. It’ll be available through the channels that you commonly use. So if you want to chat with Aristotle, it’ll be in slack or tapes. If you want to talk to Aristotle on your mobile phone, then you’ll be talking through the phone. And the phone’s interface is, well, if it’s an iPhone, it’ll be Siri. If it’s an android phone and it’ll be probably the Google Assistant. Right. So, yeah it’s also deliberate to say we don’t sell you an app. We still use this virtual thing that sits inside of your existing communication.
Richard Jacobs: Yeah! People prefer to communicate in different ways. Some texts, some talk, some, you know such and so, yeah, whatever you prefer to communication method it is, that’s to keep you in that mode because that’s what you like, you know?
Adrien Schmidt: Yes exactly! And if you do whatever preferred communication method and also whatever your preferred communication channel. So if you have Alexa for your car you’re not going to add this other assistant for Aristotle. Right. And you’d rather just have everything the same in the same device. So we’re like, let’s not add more apps, more devices to our users who already have too many.
Why don’t we just integrate where they already are? It’s also great for support and you know, deployment and all that. We don’t have to, you know, handle all that we can deploy to thousands of users without having to worry about the compatibility of our app with their phone, screen size, whatever, you know, texts, input message, voice translator. It’s all handled.
Richard Jacobs: That makes sense, so what’s in store for the next year or what are some of your like metrics or milestones you’re shooting for there?
Adrien Schmidt: Yeah, so we’re a tech startup, so we’re all about milestones, achieving levels of growth and funding and all that but the fruit from a product standpoint and a technology standpoint. I think we’re really focusing on two things, First thing is all about understanding and answering broader questions through this semantic analysis that I alluded to previously. So if we want customers to be able to say or users to be able to say, well hey I have a problem in Germany and identifying that problem in the case of your industry refers to potentially a field in your database called air rates, right? So to be able to say, well you have a problem, would you like to see air rates in Germany? And so okay yeah sure and how start digging into air rates maybe by product category and stuff like that. And then gradually identify and answer your initial question but by mapping your question to a database. So that’s what we call fuzzy querying. The other thing is to fulfill, we have this vision, right, and that our technology could voice activate the world’s databases, right? So our technology is completely decentralized, so we can basically give people or the tools to train technology on their database and create these kinds of skills, just like Alexa skills Or Google’s actions. And so we call the library of skills or the marketplace of skills or the skill store. We call it Alexandria, just like the Big Library of the past.
And so we’re really going to be working on letting our users subscribe to the data points that they want to subscribe to whether internal in their company or external from either paid or open sources. So you really have a way to ask questions across a very wide range of databases and have this experience that’s almost like search right. You know if you Google something and it just goes across all websites well here, if you could just ask Aristotle question and it spans across all sorts of databases and it helps you find the answer that you’re looking for.
Richard Jacobs: Well, very good. So what’s the best way for people to learn more and to ask questions of you guys?
Adrien Schmidt: Yeah! Well, thanks for asking sir. Our website is bouquet.ai. We have various Twitter handles. We just launched Alex Aristotle underscore AI on Twitter to be informed about Aristotle specifically. So that’s pretty new you could also reach out to me directly ad_schm on Twitter. We love to hear what questions you have and hopefully be like Aristotle and answer.
Richard Jacobs: That’s great! Okay, well thanks for coming on the podcast. I appreciate it.
Adrien Schmidt: I really appreciate you having me. Thank you so much.
Subscribe to Our Newsletter
Get The Latest Finding Genius Podcast News Delivered To Your Inbox