Using the Future to Influence the Present w/ Brian Brinkmann, VP of Product at Logi Analytics

Hey everybody, this is Mike Belsito, co-founder of Product Collective, and we are back for another industry interview. Um, I am excited about today's chat because I think this is a topic that's definitely relevant to folks, relevant to me as a product person. Um, but for those of you that haven't joined before, actually, if you could do me a favor there, I will say we're kind of experiencing some technical issues here. Um, our guest Brian Brinkman, um, he can't see or hear me, so I have him on a phone. I know there's going to a little bit of an echo while I'm talking. As soon as I'm done talking, I'm going to mute my phone so you shouldn't hear that. But if y'all could in the chat, just somebody let me know if you can hear or at least see Brian, just so I know that you can hear both of us.

That would be good. But a little bit of a—

all right, great. People can see Brian, so that is great. So if you don't mind, we're going to kind of go through— like I said, there is this echo at first, so sorry about that. When we cut to Brian, which is going to be the meat of it, you shouldn't hear that. And really, these industry interviews, if you're here for the first time, these are fun. I mean, this is really our chance to get to speak with product leaders from all throughout the world and dive in on certain topics. And so I have a few questions for Brian, but I'm sure you all do too. And so if you do, you'll notice there's a chat, but there's also a section called Ask a Question. And I'd ask that you put your question at any time. You can start doing it right now. Load up your questions and ask a question. And so throughout the interview, I'll actually go to that Ask a Question section as well. So I'd love to— I have questions, but I'd love to really go with what you all have as well, because I'm sure you're curious to pick Brian's brain. So, um, the only other thing I want to mention is we do these often. Usually it's the first and third Wednesday of the month. Um, November looks to be one because of the Thanksgiving holiday, but we have some awesome chats coming up with Jason Fried of Basecamp, Scott Belsky, um, author of The Messy Middle. So those will be a lot of fun. Definitely check those out. Um, so with all of that out of the way, I I just wanted to first say thanks for being here, Brian. Sorry for these technical issues, but really glad to have you.

Thanks, Mike. Very happy to be here.

You know, Brian, just to really start, maybe you could just give a little bit of a sense of, you know, who you are and your role and, you know, kind of why we might be honing in on this topic on predictive analytics when it comes to product people.

Sure. Great. I'm Brian Brinkman. I'm the Vice President of Product at Logi Analytics. I've got a great role. I get to work on product strategy, product management, and product marketing. I've been here for about 6 years. Prior to that, I spent 15 years at MicroStrategy. There I was largely responsible for launching new products and bringing new products to market. So my last 20 years have been in the analytics industry, either as a consultant, a practitioner, a product manager, and product marketing. And really, as I've made my own evolution in my career to now be embedded analytics, I'm just a firm believer that the best way for people to get information is in the apps they use every day. And embedded analytics is exactly the way to do that. And we do that every day. We do that with our phones. We don't even consciously realize we're doing that. And I think that's the key because we're all more efficient and productive and frankly happier.

That's awesome. Well, and maybe we could just, just to start, this is sort of like the, the 101, if you will, or at least the 101 question. But what is predictive analytics? I feel like it's something that sometimes people talk about, but I don't even know if people are talking about it the right way. You know, when we think of predictive analytics, what should we be thinking about?

Yes, that's a great question because right now there's a lot of buzz and noise in the marketplace. I think the simple way to think about predictive analytics is to kind of think about where we started from. So if you look at typical or traditional analytics, it really takes historical data and it tries to understand what happened and why. So it uses descriptive analysis, things we do every day. It uses ratios, aggregations, normalizations, averages, and it helps us do root cause analysis. What happened? Why did it happen? The difference with predictive analytics is it still uses historical data, but it uses more sophisticated algorithms, machine learning algorithms, and that really helps determine what will happen in the future. So not what did happen, but what can happen. And it gives you the ability to say, well, that's going to happen, what can I do about it so that it doesn't have to happen? You know, there's simple examples that really affect probably all of us. Will a customer return? If so, what do I do about it? Is the machine going to fail? So what should I do about it? Is a payment going to not come through or am I going to default on the loan? Great. What should I do about it? So it's really about looking at that historical data and figure out what's going to happen.

Thanks, Brian. And let me ask you this. Why, why should product people care? You know, this is where we're in this world where we learn from our customers, we're iterating. And so if even if things pop up on our end that change, you know, we as product people change course. But I imagine predictive analytics can help us. I'm just curious on like why is it relevant to product people?

Yeah, so great question. The simple answer is that for product folks who are building applications, predictive analytics gives you an avenue into insights that you wouldn't otherwise have. And if you can add that to your applications, then you can have the ability to solve business problems you couldn't solve before. So if you're a product manager and you're speaking with your customers frequently and you're asking them about their requirements, you can ask a different level of questions that can go into more depth and probe into some of the challenges they're having. And there's an opportunity for predictive to take a look at that historical data and provide insights that weren't there before. Now from a product perspective, what that means is if I can add some insights that weren't there before, my application suddenly becomes far more valuable, right? I can now answer questions I couldn't answer before. And if I have a commercially available software product, right, I sell it, now I have more differentiation from my competition. I may even have an opportunity to increase revenue or create new revenue opportunities for my product that I didn't have before. So those are obviously, you know, pretty good things as product managers as we're thinking about, you know, ways that we can add value to our products.

And I'm curious, you know, now with the work that you do, you come into contact with clients of all kinds that are using predictive analytics in their work now. I'm just curious if you could maybe talk about a couple different cases. You don't need to name the specific customers, of course, but If you could talk about cases where people weren't using it, they started using it and actually saw benefits. I'm just— for those of us that are on the call that maybe we haven't done a whole lot with predictive analytics in the future, I'm just curious about that because that might help us envision how it could help us in our work.

Yeah, that's a great question. We do have a number. One we work with is a customer. GroupFIO, and they offer cloud-based applications for customer management, marketing, ordering, and planning. It's really for mid-sized retailers, and the reason these folks need it is obviously they're competing with much larger, much better funded retailers, you know, the Walmarts and Amazons of this world. What their basic challenge was is when I have a guest visit my store, whether online or in person, you know, how do I maximize that visit in terms of, you know, they call it the basket size, but really how much money can we get out of them. What they found is by using predictive analytics, they were able to look at the customer data and determine based on those customers and their past behavior what kind of marketing campaigns could we provide them with that would provide an overall uplift. They were hoping to get an uplift around 10%, and they found in their initial campaigns that they were able to get a 22% a 50% sales uplift by creating very targeted marketing campaigns. Now, obviously, that's a pretty substantial benefit, and that's all done by really taking a look at that data and actually probably increasing the experience that the guests have at their stores while still, while actually maximizing more money. So there's a great benefit all the way around. That's the first example that comes We've had other examples where we've had customers who are suffering from a problem that frankly almost every customer has, which is customer churn. And by working through the customer data and looking again at their behaviors, they're able to predict, are they exhibiting certain types of behaviors? Are they putting, placing more calls? Are they issuing more cases? And what might we do if someone's likely to defect and what kind of, again, campaigns or promotions can we offer them that will turn them around and make sure that they'll help us stay as a customer. So as we all know, it's much harder to get a customer than it is to keep a customer. And so there's a lot of value there as well. Those are just two examples, but there are many more.

Yeah, I bet. I'm sure we could talk all day about that. No, but thanks for bringing those up. That is helpful to get a sense. Now, for when we're just getting started, like I imagine, again, folks on this video chat right now, maybe this isn't something we've taken the first step with at our company. But when we do take that first step, what are the common challenges you usually see when, when people start going? So I imagine this is something that, you know, it— you want to get started in the right way. If you don't start off the right way, you know, might, might taint our view of the whole process of, you know, hey, is this even helpful? Putting a lot of effort into this, but we're not seeing the results. Sometimes it's just a matter though of, you know, we didn't start off the right way. So what are some of those initial challenges that you see people run into?

Yeah, absolutely, that's a great point, and we do see a number of common ones that really everyone runs into. The first one has nothing to do with technology or product, it's really about the business problem. When we consult and help folks and customers, we ask them, what is it you're trying to solve? You know, product people, I'm certainly guilty of this, we get fascinated with technology and we're very excited about technology. We forget that technology is an enabler. The very first thing to do is make sure that you've got that business problem, because any sort of predictive analytics is not valuable in and of itself, but it's valuable how it's used. And so that's really where we start. So we can help. I recommend everybody start with a problem you have. Once you identify that problem, and frankly, I'd probably try and find 2 or 3 or 4, because the next stage is also difficult, which is, do you have the data to actually solve that problem? Sometimes we think we do, sometimes we think we don't, or sometimes we think we do, we realize we don't actually have the data. You have to look into the data and look for what we call predict indicators. You know, really, is there something in that data that will identify something that's predictive. A lot of times you have to go get more data. You usually have it somewhere, but you have to get it into a spot so that the algorithms can work on it. And then once you get that data, you're going to find out almost invariably, you're going to find out your data has issues. You have missing values, you have dirty data, you have incorrect values, and you really need to address that because the machine learning algorithms need to see that data in a cleanly fashion. So once you get there, and those are really the two biggest ones, the next problem is really around expertise. You figure all this out and you want to start creating models, but you don't always have those folks around, the data scientists around to help you. And frankly, data scientists are in short supply. They have a lot of work typically they've got, they've got to do. And one reason we created LogicPredict in particular It's to help the product teams out and frankly to help the data scientists out too, because the product will then take that information and help you build a model in a wizard-driven, no-code approach. Not only can you build it, you can tune it. And that way the dev teams themselves can build the models, actually helps the data scientists too, because they can build models quicker and helps them work out their backlogs. So those are the challenges. Then the last challenge is the one we often forget about. Which is we're getting really excited, we get these insights, and then we forget that we got to get those insights out to the people who need them in the front line. And so getting that insight out, and again, we recommend you do it in an embedded fashion or through your existing application to the front line folks, is really what makes this powerful because now they can make decisions based on information, they can take action all in one place, and now you're starting to see the benefits of those insights. It's not just the insights, it's taking those insights all the way through to action. So there's really going to be some value added.

That's super helpful. And I, you know, along those same lines, I have another sort of similar question. But I guess before I do that, just another reminder to everybody, if you have a question for Brian, I would much rather ask your detailed questions, especially if you have something that's been on your mind about predictive analytics. Maybe that's something we could jump into with Brian. So just, you could do it right in the ask a question section. I'm keeping an eye out on the chat as well. So if you put it there, that's fine. But ask a question section, it'll be even more helpful. So yes, please do that. But along the lines of my last question, you know, those common challenges to look out for, I guess once you as a product person decided, okay, you know, our organization, we're going to take the first step. It's good to know what these challenges are, you know, and do we have the data? Do we have the right people? How else can we prepare our organizations for taking that first step? Like, what, what do we do? Like, how do we get started? Obviously there's platforms like Logi and others as well, but, um, are there other things that you feel like a company should do to prepare themselves to take that first step?

Well, I absolutely— so there's— and that could come in lots of facets. One thing I thought I heard you asking there was, you know, how do I help build the business case for the investment in the product? And you want to do that as well as you can and as quickly as you can. You don't want to overburden yourself, but that really helps the organization get aligned behind the requirements and priorities. So there's many people typically in the decision process that may not be familiar with, of course, predictive algorithms and the technical necessities to make this happen. So just getting to organizational alignment is actually a big hurdle. After that, there is actually, frankly, probably too much information out there for the product team to start looking and learning. But I would say start somewhat basic. There's good books, there's good articles, there's good podcasts and videos that really start getting you thinking about, uh, what can I do? What are the categories of algorithms that will answer various questions, right? There's classification algorithms, there's forecasting algorithms, there's clustering algorithms, there's outlier algorithms, there's a whole set of algorithms. And what business questions can you ask? Once you get into that, there are a lot of freely available tools online. There are products you can try out. There's certainly education, online tutorials where you can start tinkering with what's possible. And that's really where I recommend getting started. And frankly, it's a journey for most of us. Many of us have engineering, mathematical, STEM-like backgrounds. Some of us don't, but all of us are probably a little bit rusty anyway. So it's good to get some refreshers because There are some things we need to remind ourselves as we're walking down this path, because we do also need to be careful not to just blindly trust what the algorithms are going to do for us, right? We need to sanity check them and check them to make sure they're not providing false positive results that aren't actually true. So you can absolutely get started. Again, I would start with the basics and move forward. But I would encourage everyone, don't be overwhelmed. Like, don't be afraid to get started. I know it can be a little daunting, but it's absolutely worth the trip.

You know, you had mentioned resources like books, podcasts. Is it one of those things where it's like, hey, search for what's out there, or are there any that come to mind for you that maybe people should keep in mind?

There are, all of them are escaping my mind. There's some nice books. I tend to default to places like, the normal places like Codecademy and Khan Academy and other places that you can go find things, but there's resources everywhere.

That's great. I'm sorry, I know putting you on the spot there. Yeah, but hey, if there were that came to mind, I thought it'd be good for everybody to know. Um, you know, somebody did ask a question, um, and it was how does Logy in particular help with our data if that's the biggest problem? Or, and if the platform doesn't necessarily help with that, I guess what advice would you give to somebody where they feel like, hey, we do have data, but you know, I don't know what kind of problem they're mentioning, but maybe it's It's either we're not pulling everything or the data is— maybe it's polluted or I don't know. I guess for people where data is the issue, what advice do you have for them before they get started?

Yeah, that's a great question. So there are two things. The first problem with the data is again what I mentioned earlier is do you have the right, the right pieces, right? So sometimes data is sitting in different places and different systems. Sometimes you have to scan through it to find it. So no, no amount of automation is going to take that right now. But we do do things inside our product to help with that data problem. For example, we ran— we've been running extensive programs with our customers, and what we found is we needed to automate the cleanup of a number of data issues. And the most common of which is missing data values. So if you're missing data values, the algorithms really struggle with that. So we have a smart data cleanup that uses a number of algorithms to actually look at the data and clean up the data, make suggestions, provide some interpolations to fill in missing values. And that really helps smooth out the process. And that's one of the benefits that Logic provides so that you're making those models, you don't kind of get stuck right in the middle. So that's definitely one of the ways that we can help you through the process. And all that is tunable and controllable. We turn it on for you by default. But if you look into it and you need to turn it off, which sometimes you might need to, you can absolutely do that.

Great. Well, that's awesome. Question as far as for people that are, you know, they're interested in predictive analytics, think it might be something that helps their organization. Is this one of those things though that you have to go all in on? It's like we need to get the entire organization in, it's going to be a big effort, or I mean, is there a way that you can experiment in the beginning to kind of feel out if it is the right thing for you?

Yeah, absolutely. I recommend you start small. So I actually almost recommend with any project you start small. Constrain your challenge, and you can figure out if you can just work that one problem. You have to remember too is you don't have to do this in production, right? You can do this, you know, obviously in your development environment just to prove this out. I would— you don't need to there's no requirement that says you have to suddenly re-architect all your systems, all your policies, reorganize. You absolutely don't need to do that. In fact, it's much simpler to build a business case if you run a pilot project with probably just a couple few people, identify that problem, show what's possible, and then by actually showing a prototype or what's possible, it's much easier to get larger funding for more initiatives. But even then, I would say if your pilot project is successful, and I'll say when it's successful, and you can roll that out to a few customers, then you can do a lot of learning and you'll find, you'll find things you missed for sure. And then you can enhance that and refine that project before you roll it out to the broader audience. And I suppose that's general good advice for all new capabilities you're rolling out. But you definitely do not need to go in, you need to go all in, you don't need to spend millions of dollars on investment. If anything, I'd do the reverse. I'd start small and make sure it works right, because you will find— you're going to find issues that are probably unique to your product, as well as some of the common ones we talked about before.

Awesome, awesome. And by the way, I have a couple more questions, but again, if anybody else has questions that they want to ask Brian, I will put your questions is above mine, so let me know. We have a few minutes left. But Brian, you know, you had mentioned, you know, you don't need to be all in, and I'm wondering if some people might, you know, they might think, well, gosh, we have to be. What are some other sort of common mistakes or misconceptions or traps that people, when they're first starting to get into predictive analytics, that you see people make often? And like, just knowing that might help us stay away from those mistakes?

Right, I think, I think the biggest mistake people make is the one we talked about earlier, which is I'm ready to go. So I think lots of times the biggest mistake is I have the data and I'm ready to go, and when you start to inspect that data, you know, you realize that you're not as ready as you thought. And that, and that is typically why we ask people to back up and make sure you've got your you have your business problem ready, because if you think about the business problem holistically, you'll start to identify, you know, what data you have and what data you might need. But that, that problem translates usually into a schedule-based problem, because once you start down a project, you'll have a project plan, whether it's formalized or informalized with some timeline. And when you start to hit those hurdles early on, those project plans start to move out of time scope, and that's never a good thing for our schedules and our roadmaps. So that's really where I would start. Just be realistic and upfront to make sure you don't skip that first step, because if you skip that first step, you're going to constantly be paying for it further on down the way. And as we know, that always talks, that always costs far more than if you're going to get it for upfront.

Great, thank you. We did see one question just come in now, and it was referring back to, you know, you're talking about actions. So, you know, you, you, we get all this information, you know, once we start using, again, could be Logi, could be some, some other sort of analytic system. We need to take action. How do you decide which action to take? Like, the data is all there, but I guess, like, how do we know what to do with it? Is that something that platforms help us with, or is that where the data scientists come into play? Or I'm just curious your take on that.

Yeah, the answer— that is a fantastic question. The answer is both, and we see that a lot when we're working with customers. So let me take that in two questions. The first question is, can the platform help with it? And second question is, can the predictive help with it? So from an analytics platform perspective, you want to make sure that you have the ability to initiate, generically speaking, a workflow. I need to initiate a process that involves some logic that gives my end users the ability to take an action, right? I might need to initiate a campaign. Or offer somebody a special service or extended talk time if it's a telco customer. So your platform itself needs to be able to take that action, initiate that action, communicate that information to other systems so that it will either trigger another system or set another system up to make sure that those actions can take place. So that is part one, and that's very, very important. The other part is how you choose which action to take. And that's where the predictive algorithms come in. So if I set up, let's just stay on the example we talked about earlier with customer churn. If I have a customer that's going to churn and I can identify that customer who's going to churn, I can then use predictive algorithms to determine which campaign they might respond best to. It's likely that if you have a company, you have 3, 4, 5 campaigns that are running at any given time. So based on that customer, you can use your predictive algorithms predictive analytics to say, well, they're going to respond to a lifetime value campaign better. So when you get to the point in your application where you need to take an action, the predictive analytics can actually narrow down that action for you. So instead of 5 campaigns, say you can narrow it down to 2 because the customer service rep on the other end might get some last-minute information that allow you to decide, and that helps the decision process the prescriptive process along the way as well. Yeah, we see that a lot. That is a very good question.

Well, awesome. And I, you know, I know we have time for really one more question, but I want to take one from the audience, if that's okay. And I thought this was a good one too. You know, if somebody, say they move forward and they start to, you know, embed predictive analytics model into an existing application, how long does it take to really deploy it, how long does it take to start to see actionable results?

Well, in the model and training the model, right? Because that's where the machine learning algorithms are doing all the computational intensive work. Once the model is created to actually get your outcomes running your data through the model is not so bad. Actually, it's pretty— it's very, very fast. You can actually do it in real time as well. So you just request and you can get an answer. Deploying that model is really straightforward and fast. That's not hard at all. The second part of it is interesting, which is how do you determine how well it's working? So what we recommend that people do is with any amount of, with any of these predictive algorithms, is as you're taking action, you also measure the results of those actions. So you have a predictive model, you expect a certain outcome with a certain probability. That doesn't mean you're going to get it, but you need to track that. And when you track that, what you can do is take that information back to your original model and use that information to refine the model. So now you have a very virtuous cycle. You're getting information, you're creating a predictive result, then you're using the result, you're testing the result, and then you can go back to that base model and improve it with the information you're actually getting. And we're seeing people do that as well. So that's taking the model prediction and helping increase the accuracy and tuning it to what's actually happening in real life. Fantastic question. Great.

Well, Brian, I know time kind of flew by. I, I, we're at the end of our time now, but I appreciate you joining us. Um, for folks that want to keep tabs on you, what you're doing, what your organization's doing, where, where's the best place for them to, uh, keep tabs on you all?

LogiAnalytics.com. There you go.

Okay, sounds great. And I know that there's, aside from just, you know, getting to learn about the platform, I know there's a lot of resources there too. Which is a very cool thing. So Brian, thank you for joining us. Everybody else, thanks for joining us as well. Again, we're gonna have more of these planned. Our next one coming up is in early November with Jason Fried of Basecamp. So I'm excited for that one. And we will see you all very soon. Thanks so much, everybody.

Thanks, buddy.