How manufacturers can use descriptive, predictive, and prescriptive analytics to plan for the future
For the latest episode of our manufacturing podcast, Data In Depth, we sat down with Kyley Darby from Mountain Point and Skye Reymond with Terbium Labs. Kyley and Skye dig into the ways manufacturers can leverage descriptive, predictive, and prescriptive analytics to maximize business outcomes. They also dig into the various ways companies can use Salesforce’s Einstein Analytics tool to optimize their data and better plan for the future.
Check out this episode below and head over to Data in Depth to listen to our other episodes!
Have a topic you'd love to hear more about? Interested in being our next guest? Let us know in the comments below!
Announcer: Hi and welcome to "Data In Depth", podcast where we delve into advanced analytics, business intelligence and machine learning and how they're revolutionizing the manufacturing sector. Each episode we share new ideas and best practices to help you put your business data to work, from the shop floor to the back office, from optimizing supply chains to customer experience. The factory of the future runs on data.
Andrew Rieser: Welcome, thanks for joining us for season two of "Data in Depth", the podcast exploring data and its role in the manufacturing industry. I'm your host, Andrew Riser. Today we are gonna be doing things a little bit differently We've got two rockstar guests joining us, Skye Reymond, a data scientist and Kyley Darby, a Salesforce Solutions architect. Skye coincidentally was our first guest on the first season of "Data In Depth", so it's great to have her back and Kylie is relatively new to the Mountain Point team but has been well on her way into her journey to becoming an Einstein expert in the Salesforce ecosystem. So, welcome to the show.
Skye Reymond: Thanks for having me, Andrew. I'm excited to be talking with you all again.
Kyley Darby: Yeah, thanks for having us.
Andrew: So before we get started, it'd be good to have you guys each share a little bit more about your background, just so that we can kinda set the stage for not only your experience in the ecosystem and with data and analytics but also what led you to this and what got you passionate about this career track.
Skye: Yeah, so my background has primarily consisted of data scientists roles in the past. I've worked in both the manufacturing and cybersecurity industries as a data scientist and I also have about five years of experience as a Salesforce administrator. What really drove my passion around data science was seeing the significant business impacts it can make and having the ability to help companies run more efficiently.
Andrew: Yeah, that's great. Obviously, you hear the catchphrases of, data's the new oil and things like that and so I think this is an area that's gonna continue to grow and expand. So, excited to have you share more of some of these examples and how you approach data with customers. Kyley, what about you?
Kyley: So, yeah, I've been in the Salesforce ecosystem for about four years now but I actually worked within the manufacturing industry for about five years. In my previous role I helped to implement several different Salesforce platforms within the supply chain area of my company, including Einstein Analytics. So, I would have to say that I always had a passion for puzzles and problem solving. So, in my role as having the ability to kinda see firsthand how the Salesforce tools can solve problems within a company has definitely fueled that passion for me. And now as the importance of data and data quality are becoming so much more prevalent within the manufacturing industry, I think being able to use analytics to actually help solve those problems has been extremely exciting for me.
Andrew: So, manufacturing has always been a legacy industry, if you will, or an antiquated industry that has grown over time through acquisitions. So when you think about a lot of data that these manufacturers have accumulated over the years, and now with things like connected devices and sensors and things that are on the shop floor and everything kind of being connected to the internet and to the cloud, it obviously brings a lot of data into the fold and so I think what gets lost is how to make sense of that data and how to kind of sift through all the noise to then find what is really relevant and what are the things that the business should care about. So there's a lot of different angles and areas that we can dive deeper into but Skye, I think you do a great job of summarizing, just at the 30,000 foot view, what should manufacturers care about and how do they create a data framework internally so that they can start putting the stuff that we're talking about into action from a process standpoint?
Skye: Sure, yeah, so a lot of the time and I know a little framework, the way you set it up, it really needs to build on each other. So, the way that we tend to frame it is in three different areas, descriptive, predictive, and prescriptive analytics and those three different areas, if you have the descriptive part right, then the predictive part becomes easier and you can't have the prescriptive part without the predictive part. So to start, every manufacturer business should really start with descriptive analytics and a lot of them actually do this. Descriptive analytics answers the question, what has happened, and a lot of times this will explore historical data in your business and look at what's been previously collected to help you understand what has been happening in your business. So a lot of times this will be found in places like reports and dashboards that show trends over time, KPIs and we'll summarize past events that have happened in your company. And once you have an understanding of what the historical data says, then you can start to build on that and that's where predictive analytics comes in. So at that point, predictive analytics are more advanced and they usually require a little bit more technical modeling methods. These methods answer the question, what could happen? Techniques that are used often with this step are things like anticipating likely events, assessing future opportunities, mitigating risks, forecasting parts, forecasting for inventory and things like that, looking into the future and seeing if you can predict what's going to happen. And then once you complete that piece, which is fairly complex and requires a lot of time and effort, the final part of that is the prescriptive analytics and that answers the question, what should happen? So once you have your predictive models, you can do things like optimize your business outcomes. You're basically tweaking variables in your predictive models to help you make better decisions for your company. You can determine the best course of action, find process improvements but this step does require dependable predictive models and a really good set of historical data that's clean and where your processes are good. So, all of these three areas kinda play together and if you can complete all three of these areas, then that's really what helps manufacturer become an analytical competitor.
Andrew: Yeah, I think that makes perfect sense. I think the, like anything, everybody wants a framework or some kind of guidelines of where they wanna go and I think that where a lot of manufacturers get lost is thinking beyond that descriptive phase. So, the descriptive phase is pretty straightforward, like you said, it's leveraging existing data that exists and analyzing that data. We see a lot of manufacturers are pretty comfortable with standard dashboards and reports and using Excel. So Kyley, what's your point of view around that? So in working with these manufacturers, how do you see them leveraging data and what do you think prohibits them from moving beyond just that descriptive phase?
Kyley: Yeah, I think, like you said, a lot of manufacturers are still stuck using just static reports, dashboards, Excel and I think a big reason for this is a lot of manufacturers have a lot of different ERP systems, whether it be from acquisitions and then not consolidating. So their data's kind of all over the place and it's not really stored centrally. So a lot of times what we see is kind of piecemealing these reports together but I think for them to kind of move forward and start to look beyond the, what has happened and stop focusing on the what has happened, they need to start pulling their data together in a centralized manner so they can start saying, okay, we see what's happened but now what could happen? What could we change? And I think just the data being kind of all over the place is really holding them back.
Skye: I'll add to that, Kyley. Some of the stuff that I've seen in the past that's becoming less of an issue is that in the past, these methods have been really technical and a lot of the technical talent is expensive. If you don't have the technical talent that's necessary, you can find yourself following a predictive model that's incorrect, which can cost the business a lot of money, a lot of time, a lot of effort. That has historically been hard to find, that technical talent that can actually help you utilize predictive and prescriptive analytics but unfortunately with Salesforce, things like Einstein are making this skill more accessible to everybody. So I think in the future you're gonna see more of that, more of where you don't really need a full data science team to implement these methods. You just need to have a good understanding of Einstein if you're a Salesforce user and what those results are gonna mean for your business.
Andrew: Yeah, I always like to revert back to the golden days of the late 90s and early 2000s of using websites as an example. So I think that when the internet was first coming out, websites required somebody that had technical chops and understanding of how to build those and how to make those changes and updates and so those types of skilled resources were required. Now, that skillset is very commoditized and you have point and click options and drag and drop options that really make it accessible for the everyday user. I think the same as evolving with these types of technologies with Einstein and with AI and data science in general. To your point though, obviously you're still gonna need strong technical talent to understand those models and understand the data but the actual execution of that I think is becoming more and more user friendly and easier within these systems.
Skye: Right, yep. That's exactly right.
Andrw: So a big question that always typically comes up is we're paying for tools, we've got all this data, we've got these priorities of a master data management plan to kind of consolidate all this data into one place. We've got dashboards and reports that we're building to try to match up to our key performance indicators that we've defined internally. So what's next? How do we really put a plan in place to really start exposing the predictive phase?
Skye: So, I usually suggest when manufacturers and businesses are starting to implement predictive analytics that they start with an area of their business that they consider to be their competitive advantage. Usually those areas are very good to implement predictive analytics because your business already has an emphasis in those areas. You probably have people on staff that are experts in that area. You probably have clean data. It's probably something you already do very well and have good processes for. So usually the best place to start with a predictive strategy is going to be the place that you already have some sort of competitive advantage.
Andrew: So maybe you can give us a real world example of that. So, what's a company you've worked with in the past in supporting and helping to take that competitive advantage to the next level?
Skye: Sure, yeah. I worked with a company one time that really put an emphasis on their service and it was in their slogan and it was something that they really pushed for their customers that they provided great service. And so for that company, when we started looking at where to implement predictive analytics, we found that the repair shop was the best place. They did things like service calls, they repaired products and what we did was we looked at their repair process and looked at ways to optimize it and because we did that, they were able to increase their turnaround time on repairs, predictively forecast when customers would need help with their products, anything that they could do to increase their service and their turnaround times and have parts on the shelf when they needed them. That was really an area where they focused on and they did see an increase in turnaround times and were able to provide even better service than they previously were.
Andrew: That's great. So Kyley, I know you're in the midst of exploring Einstein with a customer right now, so can you maybe share some of your experience around that? What are the challenges that they're facing and how's Einstein gonna help solve some of those challenges?
Kyley: Yeah, so one of the customers that I'm currently working with has data again spread across multiple systems. So they have some data in an external ERP system and then they have some of their data in Salesforce and essentially what they've wanted to do was overlay that external data with Salesforce to get a really clear picture of their pipeline. So Einstein analytics was brought in and essentially what we're able to do is pull in that third party data, merge it with the Salesforce data and really give them that big picture of here's what your pipeline looks like in a centralized location. Not only can they see what their pipeline is but now they can actually start looking at different metrics and KPIs that they weren't even aware of to begin with. So it's really allowed us to merge their data and show it to 'em in a way that they hadn't seen it before that they were really struggling with. They tried to use some of the standard Salesforce reports and then they had another reporting tool outside of that that they were using that they found that having to toggle between two reports just to get that one view really wasn't working.
Andrew: Yeah, that makes sense. Skye, you're also in the midst of supporting a customer that is exploring Einstein as well as some other metrics and forecasting. The way I like to describe it is to eliminate the finger pointing and so when things get disruptive on the manufacturing side and lead times and changes to production and things like that, get thrown curve balls, then it's a typical finger pointing back to sales and sales is being disruptive and then vice versa. Sales is pointing down to manufacturing and saying, why isn't it shipped yet? Nothing's changed on my end, everything's great. These products need to get out the door. So maybe you can share a little bit about that experience and kind of the problems that we're looking to solve there.
Skye: Sure, yeah. So data science techniques and advanced analytics will often increase visibility, which is why sometimes there's pushback from users and that maybe don't necessarily feel comfortable with that much visibility into the processes. I think one of the important things as you're implementing these techniques is to make the end user feel like they're part of the process. So make sure that as you're implementing the predictive analytics and whatever else it is that you're implementing, whether it's reports or some sort of modeling techniques, that you're involving them every step of the way so they can have some ownership of it themselves. That'll do things like increase user adoption, increase trust in the model or the system. Some of the stuff that we're doing at this particular customer, we are looking at customer volatility rates and so there are times where a customer will call in and change their order and when they do that, it messes up the lead times on the parts and it messes with their scheduling and all sorts of things. So we're looking at creating a pretty basic model that will help the end user see what that customer volatility rate typically is. So they can kind of assess and plan and hopefully get out ahead of any changes that are being made but these are all good areas to implement, predictive analytics areas where when there is a disruption, you've got some finger pointing and sometimes you're not really sure what's causing the disruption and that's usually an ideal place to start implementing these techniques.
Andrew: Yeah, data definitely helps shine a spotlight in areas where you can ask more informed questions and rely on that data to help reveal some of these problems. Just to kinda pull on that string a little bit more, the customer volatility I think happens everywhere and so especially in manufacturing, you've got your a steady as she goes run rate business, where these customers are just consistently ordering month after month and setting up the schedule and production is pretty smooth for that series of customers. And then below that you have the customers that are consistently inconsistent and so they have a plan, they have a forecast, they submit that plan, whether it be through a sales agreement or through some other forecast that they're supplying to the OEM only to then change that because obviously a plan, once it gets submitted, things change and things cause things to evolve and so that's the real area where I think causes so much pain and heartache and disruption. And so this type of customer volatility model allows you to really shine a light on the customers and products and the changes that are happening. So people may perceive that customer A is a, the bad apple that's causing all this disruption but if you don't have the data to back it up, it's hard for you to go have that conversation with that customer.
Skye: Right, and it could be that, you're assuming that customer A is the one who has the issue but when you dig in deeper, you find that that same product across all customers are having an issue, they just order more of it or whatever the case may be. So there's a lot of times where sometimes your initial assumptions are correct and that's great but there are other times where you may make assumptions where the data points to some other culprit and that's what's really important with these customer volatility models. There's a couple of different ways that you can do them and that's the cool thing about data science is that there's often more than one way to approach a problem. It's just finding the best one. So a couple of different ways that we've looked at is looking at weighted tables and giving weights to values. When there are changes, it gets a specific weight based on the product that's being changed, the date that's being changed, the customer that's changing it. Some of the things you can do are time series forecasting methods and what I mean by that is you could create like a time series forecasting model and put the predicted values up against the current values that are in your system and in places where there's a difference between those two things, you can maybe get out ahead of a customer making some sort of change. So that's kind of a good touch point if you see that what's being forecasted is different than what's currently on order, maybe that's a way to go ahead and get in touch with the customer and be proactive and kinda take control of those changes before they even happen.
Andrew: Yeah, and what I really appreciate, your ability to do with these customers is kind of, all of this sounds super complicated. It's very hard problems to solve and you're talking about weighted tables, you're talking about all these different variables but really when you put it into the context and leverage some of their data and then simplify it in the shape of a simple spreadsheet to kind of talk them through like, this is what we really mean by this, then I think it helps them visualize, oh yes, data science and solving for these problems sounds extremely hard but when you take a step back and think about the data that you have and the variables that cause some of these disruptions and then package it up into a model like you're describing, then I think it helps kind of ease the pain of taking that next step and getting something in place that's a, whether it be a minimum viable product or an actual solution for them, for the business to get value out of.
Skye: Right, yeah. With this stuff, the more simple that you can keep the model, the better it's gonna be and the more intuitive and things like that. If it's intuitive, the users are gonna be more likely to use it, they're gonna be more likely to trust it. Usually the data science life cycle takes place over several different iterations. So maybe you start with a basic model and then once you implement it, you get it rolling. You can find places where you can make improvements to that model, whether it's bringing in data sources that you haven't previously used or even trying new modeling methods. It's usually a lot of trial and error and even after you have kinda rolled out your solution, whether you're using weighted tables or whether you're using a time series forecasting model or something even more complex, following up those results and continuing to see ways that you can improve upon the model is gonna be important.
Andrew: Cool, so let's switch gears here a little bit and a point of conversation that also often comes up is within the Salesforce platform and the tools that they have available. We always get the question of, what's the difference between standard dashboards and reports? What's the difference between Tableau? What's the difference between Einstein analytics? So you've got different ways of how you can view and manipulate and work with data. So Kyley, maybe you can shed a little bit more light on those solutions and how you think this stuff fits together.
Kyley: Yeah, so I think the most important thing to note here is that with the presence of all of these tools, Salesforce standard reports and dashboards are not meant to be replaced. So essentially they are utilized for your quick on-the-go reporting, your operational reports, quick daily, what does my pipeline look like? What does my current case count look like? These are questions that they're not changing frequently from day to day. With Einstein analytics, now you're starting to look at those more complex questions. You're starting to get into that prescriptive and predictive analytics. Not only are you doing that but Einstein analytics actually allows you to integrate, like I said, Salesforce data with external data whereas in your Salesforce reports and dashboards, you're very limited on what's actually in Salesforce. So if you don't have an integration, you're not getting that full big picture. analytics, you can import that data from an external system without having to have a Salesforce integration. So CSV files, connectors, et cetera. Einstein analytics also lets you take actions on Salesforce records. So as you're going through analytics and reviewing reports, if there's something of key importance, maybe you see an opportunity that seems kind of off to you, you can make a quick note directly on the Salesforce record, create a task, chatter posts so you can actually take actions in Einstein on those records. Einstein definitely is utilized more for internal company sharing. So if you have a lot of Salesforce licenses or you wanna share these dashboards and reports internally with Salesforce users or non Salesforce users, that's kind of what Einstein is gonna be utilized for. With the acquisition of Tableau, Tableau and Einstein are definitely very similar. Tableau still allows for Salesforce data, it allows for external data, it allows for that complex data manipulation, just like Einstein does. It does have that prescriptive and predictive, it's answering your why questions but the biggest difference between Tableau and Einstein analytics is that Tableau's gonna be utilized more for larger groups of people outside of your organization. So think of, obviously right now with COVID-19, a perfect example is the COVID-19 rapport that is being put out by Johns Hopkins that's being powered by Tableau that's shared for the masses. So in a situation like that, you would definitely wanna utilize Tableau over Einstein.
Andrew: That's a great summary. So this is a question for you both as you are evolving your journey with learning Einstein and learning these new tools and data sets, what's been the most exciting aha moment and what's been the most frustrating trying to bang your head off the table, figuring out how to get something to work with the tools?
Skye: The most exciting moment for me is just, I've done a lot of these data science models manually in the past and when you do them manually, it's a process, it takes a long time. There's a lot of coding, a lot of base coding that has to be done before you can even start to work on a predictive model. The great thing about Salesforce Einstein is that the data's already formatted, it's clicks not code so it's very straightforward. It's very easy, you can have a model running within 10 minutes as opposed to previous way that I've had to do it in the past where, it can take hours before you even get the first model, or longer depending on the quality of your data. So that part's been really exciting for me. I think the frustrating part, just as somebody who has a data science background is there is less flexibility, there's a trade off there. I know that Einstein has had some updates in the past couple of months, like previously time series forecasting was not something that you could do and now they've got a time series forecasting module. So that's been nice and I'm sure that, as it continues to grow in popularity that there will be more and more things added to Einstein that'll probably continue to make me really excited.
Kyley: Yeah, definitely. So for me with Einstein, I think the coolest part is with the standard apps that the Einstein offers, the standard reports and dashboards that you can kind of spin up. It allows you to pull your data in and it gives you a subset of reports just kind of right off the bat and I think for me that's cool just because sometimes I think we beat our heads against the wall on what do we wanna report on. We do have a smaller subset of KPIs but then beyond that, I think you find a lot of people struggle with, well what else, what else do I wanna know? I don't know what I don't know so I don't really know what to ask and those out of the box apps that Einstein offers are really cool. It really gets those gears spinning. So that has been fascinating for me but on the flip side, the frustrating part for me has been trying to figure out the best way to get the data in and manipulate the data. There's a couple of different ways to do it in Einstein. So figuring out the best way has definitely been frustrating and I've definitely had some bang my head against the wall moments trying to get that data in and get it configured right so that it works best, especially if the data in Salesforce isn't following a more standardized approach.
Andrew: Yeah, that makes sense. So before we sign off today, I'll ask one more question of you both and that is, what would your advice be to a customer or manufacturer that is looking to evolve their data journey and get the most out of it? Where should they start? What should they consider?
Skye: Yeah, so I always suggest first making sure that you have an outline with the analytical framework. If you find that you start building a model with data that is not clean, you are going to cause a lot more problems than you're going to solve. So doing that is the most important thing you can possibly do. Just taking a data inventory, finding out the quality of your data, is it good enough? And if not, then changing your processes so that you're collecting the data that you know will be helpful in the future and then the second thing, which we touched on a little bit earlier, is just finding that competitive advantage that you currently have and finding a way to use data science to further that advantage.
Kyley: To add to that, I would say don't be afraid to challenge the business process and I think what I see a lot is when you're looking at the data, there's discrepancies. There's a lot of just kind of piecemealing or band-aiding data problems that really make the data quality worse and a lot of it is because there's not that push back on the business to say, why are we doing things the way that we're doing it? If we just changed our process, the data could be cleaner. So I think the biggest piece of advice is if you find that discrepancy in your data that comes back to a business process, don't be afraid to push back on that 'cause it does help out in the long run.
Andrew: That's great, well Skye and Kyley, really appreciate your time today and speaking with us on the podcast. I think there's a lot of good nuggets in there that our listeners will get out of this and I'm also super excited to continue to watch you both on your journeys of learning more about Einstein but also more importantly getting the success stories out there from these manufacturers that we're supporting. So thanks for joining us on the show today.
Skye: Thanks Andrew.
Kyley: Yeah, thanks for having us.
Andrew: For those of you listening that would like to learn more about Mountain Point and Sky and Kylie, I'd encourage you to visit the website, mountainpoint.com and if you'd also like to connect with them, we'll be sure to provide relevant links to their online profiles in the show notes. If you enjoyed this episode, please take a moment to rate the episode and subscribe to Data In Depth available on iTunes, Google, Spotify, Stitcher and pretty much anywhere else you might listen to your podcasts. Thanks for joining us today.
Announcer: "Data In Depth" is produced by Mountain Point, a digital transformation consulting firm, focusing on the manufacturing sector. You can find show notes, additional episodes and more by visiting, dataindepth.com. Thanks for listening and be sure to subscribe wherever you get your podcasts.