Podcast Transcript: Actionable Spend Analysis Through Applied Data Science
The following is the transcript from a recent BMP Radio interview with Brian Seipel and James Patounas, both from Source One. If you would like to listen to the podcast, please click here.
Kelly: Hello, and thank you for joining us today. This Kelly Barner, Editor at Buyers Meeting Point. Today, I would like to welcome two guests to BMP Radio: Brian Seipel and James Patounas, both from Source One. Source One recently announced the launch of Spend Consultant, a spend analysis as a service web platform, and both Brian and James played a key role in its development. So that's what we're here to discuss today.
Brian focused on helping companies implement innovative solutions to drive revenue and expand market share, through procurement and strategic sourcing best practices. His combined background in business analysis, information technology, and marketing allowed him to play an instrumental role in developing a proprietary spend classification taxonomy and user experience for Spend Consultant.
Now, on the other side of things, James Patounas brings a slightly different perspective to the project. He is Source One's Senior Data Scientist. His background has focused heavily on the practical application of data mining, acquisition, normalization, and visualization, to facilitate both procurement and strategic sourcing. He is the team lead and primary architect of Spend Consultant.
So Brian and James, thank you both so much for joining us this morning.
Brian: Thank you, Kelly, for having us.
Kelly: So, Brian, if I could start with you, would you mind starting us off by providing a little bit of the background on what Spend Consultant is, and why Source One made the decision to invest in and launch it?
Brian: I certainly can, and really to start the conversation off, I'll back up a bit and make a pretty wide statement that spend classification can be a very complex thing. When we're working with clients they often have multiple data sources, potentially very dirty data, in terms of quality, and if you're working through a merger or an acquisition, there are a lot of moving parts that procurement pros may need help with. So, what Spend Consultant really is, is a procurement-based tool that helps them work through spend by first collecting it, then cleansing it, and finally, categorizing it in a way that's useful to them.
So, there are a lot of tools out there who do something very similar, so a good question is, why Spend Consultant over anybody else? And what we found is that a lot of tools that are available can really be just as complex as the process that they're trying to help. So, a lot of times, procurement's trying to solve one complexity by introducing another, and that's why we wanted to develop a tool that cut through that. I can give you a few examples of that.
When we're talking about a software solution, there's a lot of time and energy that has to go into the implementation and tweaking to meet procurement's needs, and that's even before getting the system up and running to really allow procurement to develop a go-forward strategy for strategic sourcing. There's a lot of opportunity-missed cost that we wanted to address, and that's what Spend Consultant is. It's a very highly targeted, very fast turnaround, turnkey solution for analyzing your spend.
Kelly: Now, one of the things that I had mentioned in the introduction is the fact that you were instrumental in creating the taxonomy that is used within Spend Consultant. So, what's the basis for this taxonomy? A lot of us are familiar with SIC, UNSPSC, NAICS, some of those different taxonomies or categorization systems. What's the basis for the one that's being used in Spend Consultant, and what were you able to do to ensure that the taxonomy would apply across different kinds of companies and different industry sectors?
Brian: Sure. The classification systems you mentioned are definitely very detailed, and very descriptive of what a company does, or the process that a company uses to create its product. But for our interpretation, where we really wanted to go...And, again, this is based on the idea that Spend Consultant is developed, really, by procurement for procurement. So, when we're looking at spend, it's important to know what a company does, which any classification system can do for you, but it's also important to look at that taxonomy in terms of how you would approach it in the market.
So, for example, we want to be able to look at spend and give our clients a good feel for not just what their providers are giving them, in other words, what are you spending your money on, but if you're going to go to market, how would you group them in terms of your strategy moving forward? And for that reason, 99% of the time, it's going to be a taxonomy that works across industries. There's always the chance that a sourcing approach may differ from one of our clients to another, and the system certainly does take that into account, but what we found over the many years of developing this taxonomy is it's a system that works for the clients that we'd worked with.
Kelly: Now James, let me come to you now, because as I said in the introduction, you're the data scientist, and I think there are not a ton of procurement teams out there that are either used to working with a data scientist or have a data scientist among their ranks, although admittedly, maybe more and more, we should. So for people who are sort of new to this idea, some of the things that I have mentioned from your background - data mining, normalization, visualization - can you just give us a super-high level explanation of what exactly data science is and how some of the things you have listed out relate to the world of procurement today?
James: Well Kelly, unfortunately, data science as a whole, because it's a fairly new thing, I don't know that there's a uniform accepted definition. In fact, if you look at a lot of college programs, they're still developing them, and there's a significant variety in the courses taught. At the end of the day, though, I would say data science is the application of the scientific process to the manipulation of data. And as a whole, it's really the combination of three skill sets: it's practical business applications, it's math and statistics, and it's computer programming. Ultimately, for any company, the goal of any position is to be able to start off with some business-related question, and then use the information and the techniques as their resources in order to answer that question so that they can develop some type of useful information, and then implement it.
As a whole, what I would say is the goal of data science in procurement, it's really just to go through, take historical information, aggregate it, use some predictive element to it, and ultimately answer business questions. In the case of procurement, that's ‘how do we reduce costs while still taking in consideration all of the things that we as a company value as part of the profitability process?’
Kelly: Now, one of the challenges that I suspect is part of this whole process for procurement, and it's interesting that you talk a little bit about the scientific method and bringing that to bear on data, because the idea of the scientific method is that you start with a theory, or you start with a thesis, and you then do a series of experiments, attempting to confirm or prove that the theory is incorrect, right, but that you're focused on finding that evidence, so that that becomes the backing for whatever your theory or thesis was. And yet, when we're looking at data, it is always possible that there are opportunities that exist in that data that would surprise us. Let's say someone could magically come to us and say, "This is your ideal, three- or four-quarter set of sourcing waves," there might be things in there that we would never have suspected or known to look for. When we come to data and want to go through a rigorous process, how can we prevent either past experience or preconceived notions or expectations that are already on the table from clouding the results that we end up with? How do we keep the data itself right at the heart of what we're doing, and allowing that, instead, to lead us to what the best conclusion should be?
James: I would say that it's actually a very hard practice. At the end of the day, the issue you run into with any type of scientific experiment is, typically, we're conducting the experiment because we have some preconceived notion, i.e. we're coming up with our hypothesis. And it's very hard as an actual practitioner to not let that hypothesis influence your experiments, as well as your results. And, ultimately, that is the skill of a good professional. "I have a question, it needs to be answered, and ultimately, I just want to be able to take the data, ask this question, explore the data, and see if there is a answer that is resolute." More importantly, if I wanted to take this process and reuse it, that's really the computer aspect of it, is we need it to be something that can be replicated. We need it to be something that, if we say, "Okay, we're going to start with this set of information, we're going to analyze it, and then we're going to reach out and actually test it and make sure that what we've done is correct," that's where it really comes into play that you need to be able to replicate it, and you need to have other people be able to manipulate it and go through. The process itself is mostly cleaning...they're ultimately decisions about cleaning of the data, and they're ultimately decisions you're going to have to make that go into that. You're going to have to decide if a single data point is an outlier, if it's going to have a major influence in your dataset, and I know one of the common examples of that is Bill Gates' income as part of an income-related dataset. Would you remove that? Well, that's a decision the analyst is going to have to make, and they're also going to have to be able to argue it.
And that's something that we have to deal with as procurement. We have to analyze the dataset, and ask, "Is this poor quality data because it was entered wrong? Was there a typo? Is there something wrong with the system itself?" And then we have to go through and try to figure out, again, possibly, say the total cost of ownership, which is ultimately defined as figuring out, say, the manufacturing cost of something, the sales cost, the inventory cost, the customer support cost, and the warranty cost, and we would have to go through and make a decision that would allow a business to make the best decision for them, based on their own business practices, and come up with an optimized model that would help them save money, while also taking into consideration things that they care about, such as supplier perspective, or even possibly consumer perspective.
I know one example would be Target. During the recession, I believe about 2009, they changed their suppliers, and they really focused on cost cutting. Unfortunately, the market eventually got better, and they found that a lot of consumers were very unhappy with their products, because they had reduced the quality, and once the market got better, people were purchasing with the expectation that they would be buying products with good quality. So there's a lot of things you have to consider when you're doing the process, and making sure that it's something that can be replicated and that it's correct.
Kelly: The example of Target is a great one, and it also makes the point that, we can talk about all of this in theory, but at the end of the day, it's what it enables, what the process enables, or what the discipline enables that actually has meaning for the business as a whole. So, Brian, if we come back to Spend Consultant from a process and application standpoint, it's my understanding, and feel free to either clarify or add to this, it's my understanding that Spend Consultant is really more about providing visibility that quickly motivates action than it is tracking every single penny for every single thing to every little PO that was issued anywhere in the company over an enormous span of time. So, is that right, first of all, and if it is, why should we, in fact, emphasize speed of action and ability to respond to what the data says, versus just making sure every single penny is in the right stack on paper?
Brian: Sure. You're definitely correct, and I guess this is also a good time for me to go back and sing a few of James' praises, as well, or really the praises of data scientists everywhere. When we're talking about the complexities that James brought up, one of the big things these days is the notion of big data. Obviously there's analysis that procurement will do on any size dataset, but the farther we go in developing technologies, the more data we're generating. So, what we end up seeing is that there's a really good chance for paralysis by analysis. And you know, hey, I'm an analyst. I definitely get that, and it can be a complex thing to wade through a lot of data. So I'm very glad that Source One has James. I personally cannot see not having a data scientist involved in a process like this.
But, to go back to your point, what we wanted to do is really consider the idea of strategic sourcing sort of like a roadmap. It's not enough to say what your point A and point B are, in this case point A being where your spend is and point B being where you want to reduce your spend to, the savings you want to hit. You need something that draws the line between them as well. You need that actual road on that map. So, we wanted to focus less on the hyper details of spend. Certainly, it's a tool that does track and can tell you where your spend is, and potentially where your spend is going, but we wanted to create something that would actually help move you along, as well.
Earlier, I mentioned the lost opportunity cost, and that's a very real thing. If we're talking about potentially going to market to replace an incumbent who has a contract coming up within the next several months, for example, if you are very preoccupied with a very deep data analysis, you could miss the opportunity to really evoke change on that incumbent. So, towards that end, Spend Consultant isn't just a way to cleanse and analyze data, it's also a tool that lets you see really good strategies for how you can go to market, and how you can leverage the strengths that you have to evoke that change quickly.
Kelly: So, a quick follow-up for you on that, Brian, and then, actually, I'm going to talk to James for a similar question, because it was actually on my list of topics to hit, was to address this idea of analysis paralysis. But before we get to the data scientist's perspective on that, Brian, when you talk about procurement getting stuck in that moment of paralysis, it prevents us from taking action that's ultimately going to drive the results that we want, but I suspect that the mindset that it leads us to spend most of our time in while we're at work is also one that would make it very hard to then back ourselves up to a high enough perspective that we could discuss what we have learned and what we're working on with a more executive-level member of the company. You know, they don't want to know about pennies. They don't want to know about teeny tiny details. They want us to know them, because that's important in order for us to be effective, but that's not the level that they want us to communicate with them on. Any thoughts about, needing to pull ourselves out from that granular level, although we have to do the work on that scale, but then being able to translate that in our own minds to discuss the recommendations, discuss the observations, discuss strategies on a higher, more executive level?
Brian: Sure. I think...and I would agree with you completely. When you're in an analyst role, oftentimes, it's very easy to get really engrossed in the numbers and the detail, and personally, it's something that I do enjoy. But it is very easy to lose the forest for the trees, so to speak. And I think it's very important to be able to look at your initial goal. So, for example, if you know you're looking to either right-size your spend, or find more competitive alternates in the market, you know what your overall goal is. So, what are the big items that, really, the top brass would want to see? Obviously, return on investment, speed to savings. So, from the start of a sourcing initiative to the end of implementation, how long will that savings take to hit? Also, ease of implementation. How much impact is this process going to have on the day-to-day business? How many resources will it use up, and what disruptions will it cause? Those are really the big items that any analyst needs to be able to address to any higher-ups in management.
You know, it's one thing to know the fine intricacies of your dataset. It's another thing to be able to answer "How fast will this happen?” and “How much will it disrupt business?” and What kind of money are we making for our troubles?" As long as those three things are kept in mind, and really, not just shape the work you're doing, but guide your process, it should be a much easier discussion to have with higher management.
Kelly: So, James, coming back to this idea of analysis paralysis, obviously you don't become a data scientist because you would rather not work with data, right? So, it has to be something that you enjoy, to be in the field. What are the techniques that you have employed or conversations that you've had around ensuring that you never do more analysis just for the sake of analysis without moving on to actually accomplish the goal or the objective that that analysis is being performed to support?
James: Well, to be honest with you, I would be lying if I say that never happens. However, at the end of the day, the biggest thing you need to keep in mind, and this is why it's so important to have teams versus just an individual, is you're trying to answer a question. And that question is generally going to be delineated by someone that's knowledgeable about that specific area. So let's say that we're dealing with someone that happens to be a director, or some type of category manager. They're going to have some underlying knowledge of this theory. Your goal is to make sure that you're going through, looking at the data, and hopefully answering that exact question, and you need to ensure that you take into consideration the weights between the amount of effort you invest, as well as whether or not you're answering, or what you're answering. And also, to what level of confidence you need to get that answer to be.
One thing that's commonly associated with statistics and math in general is there's what's called ‘spurious correlation’, and that's when you find an underlying relationship between two things that really have no relationship. An example of that might be that if you looked at data, you could find a relationship between the price of alcohol and the high salaries of CEOs. While they have no bearing or relationship whatsoever, if you're just looking at data, it's very possible that you could uncover that.
So, how do you go about ensuring that you don't fall into that trap of just looking at the data and finding relationships that really have no meaning or bearing? Well, one, it's to make sure that you understand the actual business practices and the implications of what you're finding. And two, it's to have a good system of checks and balances, to have a team where you're bouncing ideas off each other. And Brian was, and Jen also, were wonderful resources that we went through at Source One, we bounced these ideas off each other, and we had discussions. “Does this truly matter? Does it seem to matter? Is it something we need to explore further so that we can resolve the answer to that question?” Inherently it needs to be part of the business practice, and it needs to be a checks and balances system that you develop with the team you're working with.
Kelly: Now if there's anything embedded within Spend Consultant that might serve as something of a remedy to analysis paralysis, I would say it's the fact that there are market intel reports embedded directly within Spend Consultant. So this is taking each of the categories in the taxonomy, the one that Source One developed, and it aligns it with the category research that you have done. And speaking for me, I have a background in market intel. I could just completely geek out in there, and sit and read all of those reports, but Brian, why is it part of the design of the solution to actually embed that information in there as a PDF, as opposed to keeping it strictly data-focused, and making it just a traditional analytics solution?
Brian: Sure. It's really a matter of context, and I think it's important to recognize that procurement is a moving thing. When we look at certain key categories, I can pick one out as marketing, for example. Marketing's interpretation of procurement, and really working with procurement, is something that is very much evolving. Not too many years ago, it was a bit of a frosty relationship. Some might say it's still frosty today, but definitely, as marketing is heading more in the tech direction in terms of things like digital, they're becoming more numbers-oriented as well, so there's more of a convergence, with not just IT, but also with procurement.
So, we wanted to have this market intel available, to provide context around what we're seeing in the market, what it means to go to market using the strategies that we're suggesting in Spend Consultant, and giving our audience a bit more information on what they can expect, in terms of either road bumps, or opportunities to leverage moving forward. So, it's definitely important to have that analytics, and of course we do, but without the context of seeing how the strategies that Source One employs, and that are built into Spend Consultant work, it can be a very one-sided story. So, it's very critical to have that background, as well, which is why we wanted to put in that market intel.
Kelly: And James, it actually makes me think of a follow-up question, because one of the words that Brian just said, and I wrote down and underlined twice, because it's so important, is "context." So, data science, I think of as being a very literal...Someone puts a dataset in front of you, there is an objective or a hypothesis, and you work towards looking for these relationships or correlations in the data. Talk to me about the role of context. Because how does that, in fact, get rolled into that process of analysis, and how do you allow it to steer your efforts working with the data?
James: Unfortunately, I would say that it's fairly difficult, because you need to ensure that you're keeping the question as simple or direct or as rudimentary as possible. One very simple example would be, when we were attempting to come up with a way that we could potentially compute and define speed-to-savings and complexity, we ran into the issue of, okay, are we considering this as the complexity and speed-to-savings if Source One were to implement it, or are we considering it as if any company that would happen to come to us would implement it? At which point, you start rolling into questions of, "Well, what can we make the assumptions, if we are talking about someone coming to us, what assumptions can we make about their ability to implement things?" And then, that leads to further questions of what technologies do they have access to, what databases do they have access to.
So, unfortunately, when you're dealing with context, it very much varies, so you have to make the sacrifice of, okay, we're going to standardize this for what we define as average, and then we're going to modify it on a per-client basis, based on the information available to us. And keeping that context in consideration, at the end of the day, the context itself is how can we help the client, and more particular, the way to address it is to say, "Well, in the past, this is what we've done as a standardized approach," or, "This is what we're going to define as our standardized approach moving forward," and then modify it on a per-client basis from there. So, the context does vary to some degree, but your goal to make a business process is to say, "I'm going to keep it as standardized as possible, up until the point where I have to modify it."
Kelly: Well, it certainly sounds like the advantage of where we are now is that, yes, as much as analysis is getting faster and more accurate and more automated, there is still so very clearly a role for human beings in this whole process, which is something that I'm positive everybody listening will be very glad to hear.
So, James and Brian, thank you both so much for sharing your time and insight and experience today.
Brian: Certainly, Kelly, thank you again for having us today.
Kelly: And thank you, as well, to everyone who has joined us for this BMP Radio Podcast. Please feel free to share your comments and feedback on this discussion, as well as to share it with anyone from your network that you feel would benefit from the information.