Subscribe
Share
Share
Embed
In this episode, Russ Wilcox the CEO of Quantum Analytica chats with Brett about the idea of democratizing data science.
During the Covid-19 pandemic, many businesses were faced with a decision, close or evolve. Did businesses close because of the pandemic, or was the real reason that they didn't have the opportunity to evolve?
What does it take to found a globally important company in these times? We’re interested in what happens before universally-acknowledged success.
Join Brett Kistler as he engages in deep conversations with business leaders from emerging markets, being vulnerable about their experience in the early- to median-stage moments of their founding journey.
Intro: Given a devastating event like with Corona, you have two options, either close or evolve. Unfortunately, what we saw during the Coronavirus pandemic was a lot of these small businesses just didn’t have the opportunity to evolve, and that’s where we got this idea of democratizing data.
Brett: All right, everybody. Welcome back to the ClearView podcast. Today I am with Russ Wilcox, CEO of Quantum Analytica. How are you doing today, Russ?
Russ: I am great. How are you?
Brett: I am doing great. Tell us a little bit about Quantum Analytica.
Russ: Sure. Quantum Analytica, we really focus on what we call democratizing data science, meaning that we live in a society, in an era where pretty much everyone expects a certain level of personalization whether you are talking about Amazon or Netflix. We have grown accustomed to that. We are creatures of habit. We love when artificial intelligence and machine learning is able to make inferences about us to make our lives easier.
Unfortunately, that level of machine learning is isolated to a very, very small sliver of our economy, leaving out small businesses. What we really focus on at Quantum Analytica is two things, one bringing democratization of data science to Main Street where we really try to work with Main Street and local businesses to create machine learning models to drive innovation and drive success at the local level. The second is the democratization of data. One of our projects called Cush Analytica focused on recreation and medical cannabis, and we have noticed in this field that there is no true ground truth of data. There are lots of bifurcating sources of data that are often contradictory. What we do is we have custom machine learning models that not only provide hyper personalization inside the cannabis field, which at times can be thought of as small businesses if you think about a local dispensary and especially since this is a rather new economic domain, even these larger dispensaries still function kind of like small businesses.
What we do with that is we create a unified playing field with data where we become the ground truth, use machine learning to aggregate raw unstructured data from a variety of cannabis sites and retailers. We use machine learning to unify that into a democratized data format, meaning that all of this unstructured data is now structured that we can provide to our users, those data points directly or feed that into a larger pipeline. Really at Quantum Analytica, you can think of us as a machine learning agency that is hyper focused on the 21st century economy, ensuring that Main Street exists as time progresses forward and ensuring that there is relative ground truth for new domains, such as cannabis or cryptocurrency where there is a single unified data set.
Brett: That’s fascinating. There is this problem with this moat that larger companies that have access to the data and access to the data science teams are able to just swiftly outcompete the little guys because of their access to that talent and their access to bigger data, and so it sounded like you have two prongs here where smaller companies can have access to those algorithms and access to data by democratizing and pooling both sort of the algorithm side as well as the actual data.
What would incentivize a larger player to participate in this and make their data available to you and their potential smaller competitors?
Russ: It is about the economy, the local economy. I am not talking about the economy of money, but the social economy, the social good. If we lift up Main Street, we all succeed. Where this came about was the fact that these larger companies don’t want to do that. Facebook, in particular, I think back in the early spring promoted that they are now providing customer segmentation models within Facebook and are really being small business friendly. But that’s not true. That’s a way for them to get more users.
What we try to do is we look at where these small businesses succeed. Now, they don’t have to have a 90% accurate model like these large retailers do for a number of reasons. First, they don’t have the customer bandwidth that Amazon does, the customer reach that Amazon does. Secondly, they don’t need it because they have recurring customers that trust them. Where this came out of was my co-founder, Josh McDonald, and I, we have been good friends for about 10 years. We went to college together, and I was down in Virginia Beach visiting him. He was working for a small business, and I said to Josh, let’s get going. Let’s go skydiving, I think we were going that day. He said I can’t. I have all of this work I have to do. I am like what do you have to do. He is like I have these spreadsheets, doing this and that, trying to track orders. He was just having a nightmare of a time because small businesses don’t have the infrastructure to automate tasks.
This started as an automating of processes for one specific small business, Austin Custom Brass in Kansas City, Missouri, and then it grew to this hypothesis that we have this thing that everyone might have heard of gone on in 2020 called Coronavirus. What we saw was there was a massive number of small businesses that went out of business, and we automatically say that Corona caused that. We did studies. We did research into user behavior, user queries on Google, relative need for personalization. We actually found the opposite, especially in local communities, especially seasonal communities such as seaside towns that maybe have a summer season as in Atlantic City, New Jersey or Cape Code, Massachusetts, you are getting queries for these businesses starting in March. They were skyrocketing as compared to previous years when that would happen in May or June.
We found that there was a need for e-commerce personalization and e-commerce solutions, but the small businesses were not able to keep up with these large competitors, specifically large grocery stores or large restaurants, because they don’t have the infrastructure available, first off. Second off, they don’t have the literacy. Oftentimes, these small businesses think I can do everything by myself, and that’s a very, very good thing to be self-motivated as a small business owner. However, when you start talking about data, it is going to be a requirement to succeed in the 21st century that you have a data strategy. These guys think I don’t need Amazon. My customers will come. I don’t need predictions, or I don’t need insights. I can do all that.
However, what is being done is the users are starting to expect that, and if you don’t have the ability to meet that expectation, given a devastating event like with Corona, you have two options, either close or evolve. Unfortunately, what we saw during the Coronavirus pandemic was a lot of these small businesses just didn’t have the opportunity to evolve. That’s where we got this idea of democratizing data, meaning these big companies don’t necessarily want that data to go out to their users. They would gladly collect it to increase their own profits, but what we try to do is we try to promote data literacy. We try to promote machine learning at the small business level to drive local impact.
Brett: That’s interesting. I mean the idea of COVID just having accelerated a lot of processes that were already occurring, a lot of these businesses had already been falling behind in their data strategy, including even knowing what one was, and they needed one. That became very apparent very quickly with COVID. It is interesting to tease apart how much of this was COVID and how much of this was already going to be happening and what we can do about that.
Russ: What’s really interesting is we have a partner, Austin Custom Brass, that has been great to work with. They sell phenomenal trumpets and brass instruments. We did a test back in November, and we said let’s take a look at their Black Friday sales. They had their internal marketing team do whatever they did, and we said let’s take a random split of 50% of your customers and let’s do an A/B test where we use machine learning to drive product recommendations during this time. They were totally open to it, and said they were excited to see what we did with the customer segments. Let’s give this a shot. Within the first 24 hours, their internal marketing team generated around 647 dollars of revenue, which was better than they had done the previous year. That’s pretty good. Using our machine learning model for personalized recommendations, they were able to generate over 13,000 dollars in that same 24-hour period.
It is not because this model was the best, greatest out there, but you are essentially going from the industrial revolution, doing this manually, to the computer age, allowing machine learning to highlight the features that were important for your customers and provide recommendations that they would buy. That was one of the most astounding impacts we saw. We knew this was going to have an impact, but to have that substantive of an impact within the first 24 hours was astounding for us.
Brett: It sounds basically that rather than these teams having to take on the burden of figuring out how to use machine learning and acquire and clean their data, instead you are just bringing this tool in, and it just shines a light.
Russ: Perfect, because what we try to do at Quantum Analytica is drive insights. We don’t build machine building models just for the sake of building machine building models. Everything we do has to have a correlated business outcome. We say what the business’ outcome is. In this case, it was to create more sales and to free up the marketers to do what they do best, which is Instagram posting and active marketing, not sitting behind these computer screens doing Excel that no one likes to do. What we said was let’s use machine learning to help collaborate with the human environment of a small business, to help that succeed rather than thinking about doing this just to replace a worker. No, no, no, that’s not what machine learning is meant to do.
The machine learning often succeeds the most when you have this collaborative, human-machine environment to help facilitate business and outcomes.
Brett: It is ideally an extension of our own thinking and sensemaking that can pre-process and filter information, so we are making better decisions with better information not necessarily, so it is making all the decisions for us.
Russ: A hundred percent.
Brett: Ultimately, running a business and sales, there is always going to be some strategy that you want to be able to take your time to be working on and not chasing down all the detailed information that a model could theoretically put together for you.
Russ: At Quantum Analytica, like I said, what we really focus is the data, not the model building. My background is actually in Theoretical Physics, and oftentimes we want to look at the dynamics of a system before we try to model it, whereas most computer scientists today will say let’s put this into a neural network. We will clean the data a little bit. We will organize it so it doesn’t come up with crappy results, and we will let the model do the rest. That’s where we flipped the problem on its head, and said no, let’s look at how businesses are using data, especially unstructured data. How do we normalize that, so they don’t have to hire a data analyst or a data engineer or their IT guy to help figure out API endpoints? How can we create this unified structure to help promote small businesses?
As soon as we started doing that, we saw this opportunity within the cannabis industry to start applying that same principle of data aggregation and data unification to really establish this ground truth, and we are excited to be able to use that offering with small businesses as well.
Brett: How does that look technically? Let’s say I am a small dispensary, and I’ve got some of my own customer data, some of my own sales data. It is in its own format in some siloed database running on like FileMaker Pro or some random thing, and I want to participate in this aggregate data pool and learn insights from other dispensaries and their own sales activities. How do you technically connect those things?
Russ: That’s a really great question. This is what we really found revolutionary about this process. The way we do it is through a process we call natural language processing, and that’s just a fancy way of saying that you are going to create a machine learning model that learns to read or at least is able to recognize context of verbiage. Typically, where we started was we said there is a plethora of readily available information that’s unstructured. You can get this on Facebook. You can get it from Reddit. You can get it from scraping. The problem is that it is unorganized. What did we do?
We said let’s figure out a way to get the machine learning model to organize that data for us. To do that, we created a process that using or scraping, if we were to put in Fat Panda, which is a very popular brand of Cannabis in Washington. If we were to put in Fat Panda into our algorithm, we could generate up to 5.3 megabytes of associated unstructured text that is directly related to Fat Panda. Then, what we do is we have all of this unstructured text, let’s train a custom language model to be able to identify the distinctness of these articles or these reviews or these product descriptions. Once you get the computer to have a good enough model where it can contextually understand the difference between indica and sativa or Fat Panda and, I don’t know, Kira Leaf, at that point, you can use this not only for modeling but for data engineering.
Using this, we created a process to automate the extraction of information within raw text. One interesting insight we found is that a lot of cannabis users typically will either give zero or five stars to a dispensary. There is not much in between. If you look at just the zero to five starts, you are going to see really good or really bad. What our AI model is able to do is able to parse through the review and understand. This dispensary was great. It had this strain, but it gave me a headache. But I still love it. The user might have given it a five, but there is that slight connotation of headache in negative. The AI Is able to understand the sentiment of that and say maybe that’s really a 3.5 or a 4.0 review rather than a review.
Similarly, we can train these models to extract useful information. This is called topic modeling. Does it recognize the brand? Does it recognize the strain? If so, let’s pull that useful information out, put it into a structured database that we can then run machine learning models on. If we were going with these small dispensaries, what we are able to say is not only can we look at the product offering in our country but the entire state or the entire country and understand how you are pricing versus your competitors or how many views your competitors are getting through Google versus you. All of this we hand them on a silver platter even before they give us their data.
Once they give us their data, then we can really do true hyper personalization because what we do is we do not have access to transaction level data because we are doing this all by the book. We cannot scrape personal identifying information or transaction data. We don’t do any of that, but what we do is we rely on the content of the unstructured data around us to fuel our models. At that point, when a customer or dispensary comes to us and says I want a recommendation system that will make sure that I am giving an individual who comes in here for depression the right strain. They don’t have the data to do that. They just have what types of transactions are occurring. It’s a mutual benefit to us and the dispensary to merge these two data sets together to really drive that hyper personalization.
Brett: That makes sense. You are basically using NLP to extract this data and then cluster it into different semantic concepts and then associate them with one another. I’m sure there is a lot of error in that. I am imagining a review where somebody used the word sativa and headache, and there is a big difference between saying the checkout process was a headache while buying my sativa, and the sativa gave me a headache. But it is okay to have some error there if the associations you are making on a large scale are better than chance, then you are getting some value out of it. Then, incrementally from there, you are providing insights that people wouldn’t otherwise if they simply were not looking at this data at all.
Russ: Exactly. What we do here is one step further. I love your example because what you are highlighting there is something called context ambiguity. If we look at natural language processing in the early 2010s when you had wordtovec and doctovec come out, oftentimes it would be looking at these kind of keyword solutions or keyword relations between say sativa and sativa or indica and indica, headache and headache, and there is no context.
Brett: This reminds me of SEO in 2005.
Russ: However, Google in 2019 created a transformative methodology using what we call a transformer. A transformer is just a fancy neural network that looks at context. Our models use that same transformer technology that Google does within their search engine to recognize questions and what not.
Brett: You have OpenAIs GPT3. That’s another example of a transformer for our listeners.
Russ: Perfect, that’s an excellent example. We are on that same level of contextual learning, meaning that our model should be able to distinguish the sativa that gave me a headache versus the store that gave me a headache. We don’t just look at those keywords. We look at the broader context within the NLP.
Brett: It’s fascinating. This sounds like a really interesting product. I am also curious in this conversation to dive into some of your journey as a founder. There’s a lot of AI startups these days, and there are a lot of AI startups that are marketing to small businesses. I imagine small business owners are just as bewildered as they ever have been, especially now that COVID and everything else has been throwing so much else at their plate. What is it like to reach this market right now?
Russ: It’s been an amazing journey. I think some of the strongest impact that I take away is seeing that small business succeed. When they are teetering on success or failure, to know that machine learning really, really helped and helped put us over the finish line. That’s a really good feeling. A lot of the small business work we actually do, and we are in the process of actually spinning up a non-profit, is open to any small business, meaning that we will work with you no matter your budget because we want you to succeed. We don’t bring a lot of our revenue in from small businesses.
We are very good at machine learning. We have enterprise clients, and we have Cush Analytica. What we think of this is a societal need. If you are a small business that doesn’t know about data or doesn’t have a data strategy, that’s okay. We will provide this pro bono for you in order to help you succeed in the 21st century economy.
Brett: Tell me a little bit about some of your challenges personally in growing this business.
Russ: It has been interesting. My background is actually not in artificial intelligence or computer science. My background is actually in computational physics. I was doing great there, full ride to a Ph.D. program, and then my dad got ill. I had to transition out of that Ph.D. program, and I needed a flexible work schedule. You only get through consulting. I started Quantum Analytica as a sole consulting, and that had its own interesting challenges.
I actually got my first consulting gig through Reddit. It was with an app company called Pizziz. It was really doing some really cool things with Gans and music generation, so that was my first gig working with Pizziz and their CEO, Rockwell Shah. It was a great experience. I loved consulting. It got me to be able to work on problems I was interested in, and then it became a snowball effect. You have one good success, and you refer to another client. This time it was for a client that was working for the U.S.’s largest eye care provider, and I really got a totally different problem space. You go from novel machine learning to how we improve marketing efficacy.
Then the next client was a Norwegian data science firm I was partnering with doing product recommendations in AI, and that’s really how I got into natural language processing. We really sat down and established Quantum as an entity about a year ago as opposed to a sole proprietorship as a consultant. That transition was amazing in so many ways.
First off, it forced me to grow the team. You had to realize that in a startup you can’t just do one thing. As much as you want to, you just can’t. What you need to do is find co-founders you can trust that you can lean on, that you know will always be there, having your back and having the same goal. I happened to be very blessed to have two very strong co-founders, Josh McDonald, who I was close with in undergraduate, and then my chief operations officer, Bernie Wright, who does a lot of the work managing the operation, organizing and also we were both physics majors at Boston University. It helps to have a very strong co-founding team. We are all going to have good days and bad days and positives and negatives, and you really lift each other up.
That’s one thing I would say for most startups is make sure you have a good executive team that blends well. The second is realize that you are in for a lot of work, meaning that it is not a 9 to 5 job. You have to do the billing. You have to do the invoices. You have to do the marketing. You have to do the AI. You have to do the data science. You have to do all this type of stuff. If you think about that, there is no way you are ever going to get anything done.
What you have to do is silo and focus on what is essentially needed for this product. A lot of new tech CEOs will say I want this feature and that feature, and this feature and that feature, and you get into what’s called feature creep. That will indefinitely stall your startup. Really sitting back and thinking of this as what I need to do to succeed, accomplishing that in almost like an agile methodology, and moving on to the next task.
The third is don’t be afraid to pivot. Yes, it is an issue if you are pivoting every week this project or that project, look, there’s a squirrel over there, almost like that kind of ADD kind of comedy. Oh my God, I can’t focus. Look at the squirrel over there. What’s really helpful is to recognize that pivots happen.
We started as a data science agency for large scale enterprise clients, and we still do some of that. We found something really, really cool that we can do with small businesses. We tried that, and we realized we are not doing this for the money. We are doing this to help societal good. How can we take that and still be financially viable? That’s how we landed into the cannabis space. This cannabis space is an interesting blend of the two. That’s kind of where we got a huge motivation to start creating products and developing product lines and all of this type of stuff other than moving from the sole consultancy.
The biggest take homes I can say for any startup founders is have a good exec team, have folks you can trust. Second, it is a lot of work, but don’t try to take it on all at once. Take a single bite at a time. Focus on one project, getting that done and getting that out to production. The third is pivoting is okay. Pivoting is going to happen. It could happen two times. It could happen five times within three years. As long as you have that final goal in your mind to say that you want to succeed, you want to accomplish something, you will find your niche, and just don’t give up.
Brett: That’s all great advice. All of that sounds like it came from a number of mistakes or near mistakes as most of the startup process is. It is a process of constantly making mistakes of various types or narrowly avoiding really big ones and being like woah. I am curious as we close. Of all the lessons that you have learned the hard way, what is the biggest one that touched you the most personally?
Russ: That’s a really good question. I think I can answer this from two perspectives. The first is going to be more of a technical for those CTOs out there. Validate, validate, validate. I remember it was my first time dealing with Pi Spark and working with a client and doing database calls. I wasn’t trained as a computer scientist, and I had a very, very good mentor. We had a couple projects that were fairly rough. This was early in my career. By fairly rough, I mean rather rocky. It came down to a great lesson of always validate your data. Have a different way of checking it two or three times. Before you put something in front of a customer, you are sure it is going to work. You can apply that to software development or whatnot, but I think oftentimes in data science that process often gets forgotten, meaning that you are so focused on the algorithm, and you are so focused on the product, that you don’t care about the data and what insights can be made from it. Always look at the data and validate it.
Brett: That also makes sense in software in general. If you are rebuilding an app in some new technology or just building something new that you are going to be migrating data from the client’s previous database, it can be easy to design exactly what you want to build and then realize that the data that you are actually migrating over doesn’t have what you need to actually do this. You have to account for that.
Russ: One hundred percent. Data validation is key. Then, secondly, failing is okay. This is for more of the CEO side. One of the biggest things we struggled with was how you connect to small businesses, meaning that they are probably on average about 10 years behind technologically. They are still using WordPress. As good as these platforms like Wix and SquareSpace and Weebly are, they are almost crutches. They get you into the digital domain, but they don’t allow you any control over your data.
One of the things we started pitching to companies was data science, Bring data science. They are like what’s data science. I was like let’s scratch our head and start over again. How do we market this? The second thing we did was small businesses know websites. Let’s build websites and we will make them data driven. That went as well as the Hindenburg, I would say, because these types of do a website as the service, like SquareSpace and Wix and Weebly and to some extent WordPress, if not done properly, gives them the feeling that they control things when they don’t need to.
One of the things that a lot of small businesses want is the ability to update textual content on relatively static pages. We asked them. We are going through the feature. We get done with the website, and they say I want to be able to add this text box in. We are like we didn’t provide you with a CRM. They are like Wix does that. Let’s get down to this. Over the past 10 years you have used your website, how many times have you updated the about me section? Zero. Regardless, that sense of security, that sense of ability to control what’s on the website is lost when at times you go through a custom site. While the custom sites have a lot of benefits, we offered things like analytics packagers and Google analytics done right and all of this stuff like that, it wasn’t enough to get past that literacy hump.
We got into website projects that went okay, but it was very challenging ones. Other ones that no matter how hard we tried, we couldn’t keep up with Wix or Weebly because that’s not what we are trying to do. We are not trying to be a web development agency. We are trying to be a data science company bringing data science to Main Street. That’s when we said let’s take a step back and think about this almost as like an NPO and say how can we promote this literacy. Then we found a niche that we succeeded in, and we were able to navigate through and do some tests, and then stumbled through cannabis.
The moral of the story is you are going to succeed sometimes when you are a startup, but you are going to fail a lot of times. That’s okay because as long as you are learning from your failures, you are able to make those necessary pivots to accomplish your end goal. Yes, failures happen but don’t warn over that. Look to the future and say what lessons can I learn from this situation and use that on my next project. That’s what I really like about the startup environment because it really fosters this idea of growth mindset, meaning the only stopping you is you. Once you get through that kind of mental hurdle, that world is your oyster.
Brett: Ultimately, you can run your startup in a way that even if the whole startup fails, you have still learned a lot and you have built connections. You have built resources, and you are in a better place than you started. I have seen that happen a number of times. I mean this has been a really great conversation, and I really appreciate you joining. Thank you, Russ.
Russ: My pleasure. Thank you so much for having me.