Using data: technology, culture and organisation

Frode Stenstrøm: A great many companies want to become data-driven in their decision-making. But getting there is not so easy. I have therefore wondered what a company should consider in order to become data-driven. What does it mean to become data-driven, and why is this relevant? What is required in terms of organization and culture? And what role does technology play in the journey toward becoming a data-driven company? I hope Øyvind and Eivind can provide answers to this. My name is Frode Stenstrøm, and this is ProsessPodden. First, I would like to welcome you, Øyvind, to the studio. Øyvind Nilsen: Hi, thank you. My name is Øyvind, and I have worked with analytics for about ten years. Both as a consultant previously, but also in-house. Now I am a consultant again. Eivind Lofthus: My name is not Øyvind, but Eivind. I have also worked with analytics for many years. I started in the 90s. Working with data warehouses and data science and all sorts of different things in between. Frode Stenstrøm: Welcome, this will be exciting. I thought we should first clarify some concepts. Øyvind, what exactly is a data-driven company? Øyvind Nilsen: It is a term that is very much in vogue, or at least has been very much in vogue over the past year. But to me, it is really about how a company should use its data, not just having it presented in the form of tables or other types of visualizations, but being able to make use of the data it has Frode Stenstrøm: What does a data-driven company mean to you, Eivind? Eivind Lofthus: It depends somewhat on the situation, I would say, but actually I am not particularly fond of that term, really, because it sort of indicates that the data should run the business for you, that you should make all the choices, which is not the case. But for most people it means, if we interpret it as good insight, that you make informed choices based on findings in the data, among other things, findings in the data. So that is what it means to be data-driven to me. Frode Stenstrøm: If we continue clarifying these concepts, and we were talking about data-driven companies, why is it relevant today? Øyvind Nilsen: That concept is very relevant because companies today have a great deal of data that they do not make use of. When you collect and structure your data, you also naturally get a much better overview of it. You also gain insight, and you can share that information with the other employees, and in that way try to create value from it. Frode Stenstrøm: But what is your view on the importance of data for the company, Eivind? Eivind Lofthus: Well, I think that is changing a little... It is largely because companies are becoming increasingly digitalized. That essentially means that the systems are recording more and more data. So the potential for using that data is only getting greater and greater. What may be new in recent years is that traditional BI has focused very heavily on the result of a process in a company, whereas now with modern systems you can also record many of the input factors. This opens up new analytical methods. I think that is something that increases the importance, at least, of looking at the data, considering it, and including it in the decisions you make. You could also say that if you work in a very customer-oriented B2C company, you also have the opportunity to let the data actually make decisions, such as in marketing, where instead of subjective opinions, probability models can be built that form the basis for customer journeys. Frode Stenstrøm: But do you think all companies should aim to become more data-driven? Øyvind Nilsen: If we just continue a little with what Eivind is talking about, then companies also need a little help with... That is, what data can they, what data do they have, that is, what data can they make use of as well, and that is an important point to include, I think. Eivind Lofthus: I think if you run a business in a competitive part of the business sector, then if you stop perfecting what you are doing, you lose the competition. So if you feel that you are exposed to competition and you do not use your data, then you have, in a way, an area where you can improve Frode Stenstrøm: Øyvind, how should a company build a good culture for collaboration around analytics and data-driven decisions? Øyvind Nilsen: I think that, again, everything starts with a need, but if we can assume that a company has decided to make better use of its data and have it presented, that is often not quite enough, at least according to my experience; this needs to be anchored. First, you need to establish that there is a need and a goal for where we are going. Otherwise, you lose the rest of the organization very quickly, because when you start doing analytics, perhaps you are going to load a data warehouse, and you are going to present it in some kind of analytics tool, and you are, in a way, going to work your way through it—there are many approaches, but you can take a "top-down approach." And the faster you move down, the sooner you encounter challenges. You encounter challenges related to ownership of things. You encounter challenges regarding how the new way, if there even is a new way, how things should be done. Then you need to have management behind you when you are going to do things like that. You very quickly get a great many questions when you actually sit down at the drawing board and do things. Eivind Lofthus: If an analytics or BI project is anchored only in IT, it almost never goes particularly well. And if analytics is performed by people with no IT knowledge at all, it almost never goes well either. So it is, in a way, what shall we say, an undertaking that requires three types of expertise. One is methodology and expertise and understanding of numbers, and the second is an understanding of IT. And the third, which is the most important, is really an understanding of the business. Øyvind Nilsen: Yes, what you are saying there is very important, that we must by all means avoid a BI system being IT-driven; yes, that is terrible, right, as a consultant right now, I do not know what or how the daily work of a sales manager or a category manager is; I do not necessarily know his day-to-day needs; he or she has to tell me, and then I can discuss things a little with them to figure out how we can reach that goal together. And that is where that comes in, in a way. If this is to be implemented successfully, then we need, in a way, to have spokespeople throughout the organization. Especially if it is fairly large, that is, if 20, 50, or 100 people work in the sales department, and the same in four other departments, then it is clear that you cannot receive all those messages on your own. You need a key person. A key user to be able to communicate and ensure that everyone is pulling in the same direction. Eivind Lofthus: I have been in this situation, where I was responsible for the development of data warehouses and BI in a larger company; what you notice is that requirements come from an entire organization and are of a very different nature, yes, and for an IT person they are very difficult to prioritize correctly. So putting a governing body in place, or reaching consensus on which direction the development should take and how you should prioritize it, is absolutely crucial. It also means that. You bind the people who sit on that body to the mast. You avoid all these repeated battles over priorities and resources that you otherwise often get. Frode Stenstrøm: doesn't the data warehouse team or analytics team then quickly become a bottleneck in the system? If you are going to serve such a broad range of the organization, there are a great many people you actually have to satisfy? Eivind Lofthus: That is where modern tools come in. Because it used to be that the data warehouse group was always a bottleneck. Yes. Whereas now, with the self-service system, you have the opportunity to create much more flexible solutions that reach many more people and are much easier to set up, where the user does much of the work that the data warehouse team used to do. Frode Stenstrøm: Yes, because I myself have been on the ordering side with you two, that is, I want to get a report on certain things, because I have to sit down, describe what I want now, you have to create it in a display format and some form or other, and then I realize that it was not what I wanted. And then I go through the whole cycle, whereas if I could just sit down with the self-service solution and ask the question and get an answer, almost like a conversation, then I would probably have obtained that basis much faster. Øyvind Nilsen: When I was responsible for data warehouses at one of Norway's largest companies, so it was a bit like, should this be a self-service solution? Where users are allowed to create the content they want, based on the data we make available. Or should it be the case that this goes through me, then, and IT? Frode Stenstrøm: What did you do? Øyvind Nilsen: No, there were a lot of discussions about that. We concluded that this had to be encapsulated a little bit, without taking... The freedom away from our users, because they had been allowed to do whatever they wanted, until I started there, and the result was an incredible amount of poor content, a lot of duplicates, I mean, there were 30 reports of the same type, but you didn't know that. And I didn't know either, because we each had them somewhat separately. So how did we sort this out? It was about encapsulating it a little bit, but I quickly realized that this can't manage on its own. I somehow have to move a little toward that mouthpiece I mentioned instead. And then it was a matter of finding strong personalities in... Sales, strong personalities in logistics, category management and procurement, and so on, because they would be able to help me prioritize, but also communicate to the organization that something is on the way, you'll get this, you won't get this now, and so on, and as we gradually established that group, and it became part of everyday operations, it perhaps took six months. Then we got more or less where we wanted to be, with everyone actually pulling in the same direction, so we didn't get so much pestering, and the organization was satisfied with what we delivered. Eivind Lofthus: Clearly, it's a balancing act, among other things, we require that, I mean, if you're going to release raw data to end users, it requires Much more thorough preparation. It requires, in a way, preparation of the data that makes it suitable for anyone. But the other thing I want to say is that this is very dependent on the business and the industry, really. Some industries are by nature much more analytical than others. They require many more "ad hoc analyses" than other industries where things are more stable. So there is, in a way, a spectrum of correct answers, but the key is to organize intelligently, regardless. Øyvind Nilsen: Yes, and that is part of the challenge, at least with this, and probably many other systems or solutions as well. It's a bit like, you meet users who are technically proficient, and some who aren't as proficient, and some who are in between. But you also meet users who are interested in what is being shown, while some perhaps don't see much value in it. How can we help the organization become more data-driven, as we discussed at the beginning? It's precisely about... Selling things effectively, but then you need buy-in, and people see that, not least, they know that they will actually be allowed to improve at this or be allowed to become good at it. Frode Stenstrøm: But there is presumably an acknowledgment in what you're saying that this is not a project where you build something and then you're done, but this is more a part of line operations, something you work on continuously, and by seeing that management uses data for communication, you will, in that way, build a cultural understanding that, «Okay, there is a point in using our business system, which the data comes from». Because there are a great many companies that do what I call management by mail, or management by Excel, where they either manage tasks through emails to each other, or they simply create complicated Excel spreadsheets that no one has any visibility into. I strongly believe that if you open up the dataset and share it with everyone, and make the whole mess of input factors visible to everyone, then over time you also realize that yes, we have control here, we know what is happening, and why it is happening. Eivind Lofthus: Companies are constantly changing. Unfortunately, BI reports don't. Frode Stenstrøm: No. Eivind Lofthus: A fairly common phenomenon is that data warehouses report on a whole host of things that are no longer used. An old trick of the trade is actually to turn off all reports. And then you turn back on the ones that are requested. Frode Stenstrøm: Yes, so you turn them on once people complain. Eivind Lofthus: Yes, exactly. But that's rather drastic, though it's probably the right way to think, at least traditionally, that it's a factory producing figures, and it's probably worthwhile to have a clear-out once in a while and see what is actually being used. Øyvind Nilsen: Yes, absolutely. That exact trick you mentioned there, we actually had to do that. I mean, there were such large numbers of reports, and there wasn't very good control over who actually had access to this, and not least who used it, and how often, right? Then there was a "performance issue" with this as well, where part of the blame... Could be placed on there being so many duplicates and so much poor content. How could we make the content relevant and more focused? I mean, what should it look like when the logistics manager or some other employee opens it? What do they actually want to see? What can they do something about? Frode Stenstrøm: You're touching on a topic I find quite interesting, because it concerns communication. Because when you look at reporting, for example, I might understand figures in one way, while an HR manager might understand the same figures in the exact opposite way. My father, who is a journalist, interprets the figures in a completely different way again. So isn't it actually important to think about communication? Eivind Lofthus: It's very easy to create figures that are liable to be misunderstood. My first job was at Statistics Norway, and that was the most, most, most important thing. It was that the figure you published could not be misunderstood. Now, it was misunderstood anyway, you could say, but it's an important point. Øyvind Nilsen: But this is where... A single source of truth, reliability, and uptime come in. Frode Stenstrøm: Availability of the figures. Øyvind Nilsen: Yes, exactly. This is something that goes without saying, and you've heard it before, but if you don't have a single source of truth, and good uptime, then it becomes, well, then you don't get any reliability around this. You don't get any... Momentum in this type of solution. Frode Stenstrøm: But I'm wondering how we can ensure, and use technology to ensure, that we can trust what we actually see? Eivind Lofthus: What you can say is that you don't get very many attempts, because poor data quality or simply incorrect figures destroy credibility, and then it's no use coming back with the correct figures later. When you create a solution like this that produces statistics and figures for a company, it should be done properly. From the start, yes. Quality assurance? Yes, that's the most important thing, in my view. Otherwise, you'll have a problem. Frode Stenstrøm: But isn't it very demanding to quality-assure that data? Eivind Lofthus: Ideally, the data should be correct when it is entered. A traditional BI project comes afterward. After you've invested large sums in operational systems. Operational systems are designed to ensure proper processes, not necessarily to enter high-quality data. So much of the work in a BI project is finding and correcting the errors and deficiencies in the data from operational systems. If you think a little more holistically, get involved early enough, and preferably use products from one vendor, then you can ensure that the operational system delivers high-quality data from day 1, which is critical. So I think, given how the market has evolved with the three major cloud providers, that this is a new opportunity. Øyvind Nilsen: Many of them have businesses with a lot of legacy. They have a system or a bunch of processes that have always worked. But often they depend on a small number of people who maintain this system, or these processes, right, which makes it very vulnerable. Then, when you come in with an analytics platform, you can of course check against it, if that's the case. But what we often like to do, or what we should do, is retrieve the raw data. As a data warehouse developer, you always want to go to the source, inspect it, and mirror it. Verify that you have the same thing; that is step one, before you then process this data, continue the process, and ultimately present it. Frode Stenstrøm: Is it possible to find a company that doesn't have one data quality challenge or another in some operational system? And yet there is still someone sitting at the top who wants some numbers. Can you find the gold and remove the noise, or does the company have to clean up the operational systems in order to get any benefit at all from and enjoyment out of that investment in data warehousing and reporting? Eivind Lofthus: That is a very good question. I would say that the best thing is to fix the data where it comes in. That is, you design a good operational system, also with a view to the data being used in analyses. Frode Stenstrøm: They have been in operation for ten years, and then they start reporting now. So there are ten years of production data. Eivind Lofthus: Think about analytics from day one. Then you avoid that problem. But if you now have the problem that you have a lot of... Among other things, partial history, for example, in operational systems, then a good analytics platform and good people there can do incredible things with that data. Øyvind Nilsen: An old and well-worn expression is "shit in, shit out," and I have experienced that many times too, and then my answer is yes, in a data warehouse or an analytics solution, we can always manipulate the data, either remove it or change it, add some logic, and so on, but we do not want to do that. Part of the point for me when you make the whole journey and get to the point where you are going to visualize something in, for example, Power BI, is that you can use this—and people forget this—you can use it to highlight errors. To find out where the errors are. Frode Stenstrøm: As a tool, then, also to sort of clean up the operational systems? Øyvind Nilsen: Yes, just as much a tool for finding errors, or to find errors, as for creating another type of insight. It is also insight when you manage to find errors, not least where they come from, perhaps even explain them, so that you can then go back with a plan to the source or how it should be, and then try to correct it. Eivind Lofthus: That is very, very important. It is much more common for people to think that they have good data, and then the data is not of sufficient quality to draw the conclusions they draw. The reverse, sort of. When people think they have bad data, the data is good enough. That is quite rare. So it is very important that those who are going to make decisions based on information from data know the data quality. It is extremely important Øyvind Nilsen: to have ownership of it Frode Stenstrøm: ownership of the figures, yes Øyvind Nilsen: and that is also, just to take a situation that you quite quickly encounter in many places. That is, you need to supplement the data from the source with some other sources, or something entirely external. Let us say some kind of grouping that is needed. Someone needs to use it, but it does not exist in the source. Okay, then someone has to own that grouping, someone has to take responsibility for it, someone has to update it, because what we do is read it in, and on the other side there are people thinking that this is correct. Which it usually is. But what if that person leaves, or forgets, or does not do their job, right? That is why ownership of the data is so important, but also that this ownership is anchored, that the users also feel the responsibility. Then it comes back a little to culture again. Frode Stenstrøm: I am picking up on what you are saying about having to think of analytics as a foundation in the operational systems if the data is going to be used for anything; it should not just be entered and logged. Eivind Lofthus: Yes, that is important, but when it comes to culture, there are many places where they have the opposite culture, right? It is good to have an opinion, go out and express lots of opinions. Some things that we do not really have evidence for, yet have opinions about. Frode Stenstrøm: I think we have a political system that is based on that. Eivind Lofthus: Yes, they are sort of "cowboy operations," you might perhaps call them, but it is probably more common than "cowboy operations," really. Frode Stenstrøm: But we did not entirely avoid the topic of artificial intelligence and the Microsoft Copilot questions. I hear what the advertising says: you should be able to ask the chatbot questions about some things, and then it should be able to look them up in the underlying systems and give you an answer. But have you considered whether that is, in a way, a path forward given how things look now, or is it mostly a toy today? Eivind Lofthus: That is the big question. I think the answer actually lies somewhere in between. I am quite sure it is going to become a useful tool, but perhaps not as important as the hype suggests. Frode Stenstrøm: Those who hype it the most, those who really hype it up, they say that, okay, if I ask what this week's revenue figure was, I get a figure. And then I should be able to ask a follow-up question. Why is that? And then, in theory, the tools should be able to go and analyze the underlying figures and give me a proper answer as to why revenue fell by 10% from one week to the next. And that is where I wonder how much of it is actually real. Eivind Lofthus: I do not think you will ever get a model to state the underlying cause. It can probably break down the figures for you, so that you know which products are selling less Øyvind Nilsen: We will talk again in a few years about this. What already exists is very impressive, and I do not know everything that exists either, but what I have tried to test is very nice and good, really, but it may have become one of those buzzwords, like big data and those kinds of things a few years ago. I still have not—we have had some big data projects, but in a way, there has not been enough data to really benefit from it. Is AI interesting enough now, or good enough now, to really benefit from it, or am I just as well off sitting and thinking for myself and clicking around a little and asking my own questions? It is very difficult to answer, but I do think it will get much better and that we can use this for things I have not thought of yet Eivind Lofthus: What we do know, at least, is that two years ago no one could predict how good the language models are today. So it is probably likely that this will happen again, so it is very difficult to see how good those models will be a couple of years from now. Øyvind Nilsen: Development is moving incredibly fast. Eivind Lofthus: Yes, it really is moving fast. Øyvind Nilsen: Because these large companies have... Absolutely staggering amounts of information that feed these models, and that is also the explanation, or much of the explanation, for why it is getting so much better so quickly. Eivind Lofthus: What I think—and this is no surprise—is that things like programming languages are very well suited to that type of artificial intelligence because, after all, it is a fairly structured and rule-based way of having a language, quite simply. I think it is going to become very common that instead of typing code into a tool, you can speak that code if you prefer. Frode Stenstrøm: Yes, because you can see how the models—what is current now—is that the models for converting speech to text, or text to speech, have become quite good. So you can now speak to Copilot in Norwegian, and it will translate and transcribe the question with very high quality. Not perfectly, and there are of course dialects it struggles with, but we have come a long way. When you first ask what this week's revenue was for product category B, the language model manages to turn that into a query to the underlying systems and give me an answer. But that is just another way of looking up that sales report. So I am asking whether it is more toy than useful. For me, the benefit would be if I could ask a follow-up question: but why is it like that, and actually get an answer? But that is where I still think you need to have built a number of analyses and models and forecasts that perhaps explain the input factors. What I think will have the most impact is internal communication, that you can share information more easily without creating structured, predefined reports, as you touched on a little at the beginning—that is, self-service solutions for getting answers about figures. When you see what Microsoft is doing with Copilot in the operating system, where they are building it in, you can essentially ask about any of the data that you have made available to the system. If you were to start a completely new project for a new customer today, who has operational systems but wants to begin the journey, where should the customer begin that journey? Øyvind Nilsen: It depends somewhat on what we base it on, but often the customer may already have some other form of data warehouse. So then the question becomes why they should switch and what they should switch to, right? We are going to provide some technology advice, we are going to do an analysis of what it is they want, what the needs are here. Before we can, in a way, start looking at what data they should have and so on. Eivind Lofthus: The best starting point is to begin with business needs. What is it you actually want? What are the challenges? In other words, what do they need information about? And then the data follows, and then the technology follows. That comes last. And when it comes specifically to technology, it has, in a way, I think then, become easier and easier to master, in a way; there are fewer and fewer technical or technological barriers to doing this. Frode Stenstrøm: Would it really have mattered if a customer had chosen technology from Microsoft or Amazon or Google, one of the big ones—have they become so comprehensive now that it really doesn't matter that much anymore? Øyvind Nilsen: Without going into detail, speaking somewhat generally, most tools can do largely the same thing. But there may be slightly different paths to the goal. The advantage of the big companies, such as Microsoft and Fabric now, is that they package everything together. That makes it a little easier and faster to get started. Frode Stenstrøm: Is it cheaper for customers to go that route? Eivind Lofthus: Good question, actually. Frode Stenstrøm: Because many of the customers we talk to every day are companies with perhaps 10-20-30 employees. And in the old days, they would not have been able to afford, for example, a data warehouse platform at 15-20,000 a month; it would have been completely out of reach. Eivind Lofthus: I think what Microsoft has done with Fabric, because I believe that right now, they have a competitive advantage in the market. They have a head start. That is due to two things, really. One is that Fabric has a really broad range of tools and services. Which they have "bundled" together, priced, and their pricing is actually quite aggressive. So the entry price for Fabric is acceptable for almost all companies. But there is a downside to doing it that way; the pricing is very non-transparent. Frode Stenstrøm: So it is difficult to know what it will cost over time? Eivind Lofthus: But at least you avoid the typical salami-slicing that many others engage in, where you have to... A little for this, a little for that, a little for this, a little for that, and then the total is quite large, at least. Frode Stenstrøm: I firmly believe that it is unwise to start evaluating and comparing their features, because these are services that are developing so incredibly fast that if you do such an analysis, that analysis is almost obsolete the next day. So instead, you have to choose a provider you trust and know can be there for the long term, so that the investment you make does not disappear. With the investment Microsoft has made in Copilot now, which I understand has not just happened in recent years, even though the market has become aware of it in recent years; they have been working on it for many years. Now that they have opened Copilot connectors for everyone who has worked with Power BI, 1,100 connectors in total, they say that this has been a vision they have had and worked on for 5, 6, 7 years. It is not just because we are aware of it now. Øyvind Nilsen: If such a large company as Microsoft makes up its mind, then things happen. One example is that I have worked with many different analytics tools, meaning the presentation part of it: Cognos, Tableau, Qlik, Power BI. We had Power BI in for a trial five years ago. No, we did not want it. It simply was not as good as Qlik and the other tools. But it was still quite new then. What has happened to Power BI compared with Qlik and Tableau and Cognos over the past five years? Power BI has become much more than a presentation tool. It has actually always been that, but the possibilities there now are... I mean, you can get lost, so there is a risk of getting lost, but it also provides the opportunity to be much more dynamic; you get a great deal as well, and it is the same with the other technologies they are introducing too; things move very fast, a great deal is happening, almost a little too much, and you get where you need to go. Frode Stenstrøm: So, to summarize, perhaps it is not the choice of technology that is the focus, but the reporting culture and the data foundation? Eivind Lofthus: I think technology is the least interesting part. Øyvind Nilsen: Personal preference, where do you have the most experience, right? What does it cost, of course? But partly personal preference. Frode Stenstrøm: I think you have given many good answers. I thought we would wrap up there. Thanks to Øyvind and Øyvind for your contributions to today's episode. This episode can be found everywhere you usually listen to podcasts, as well as on the website Radio.ProsessPilotene.no. This podcast is produced by ProsessPilotene. ProsessPilotene increases the value of your customers, uses market-leading technology and works with the business processes of marketing, sales and service. The person responsible for this episode has been me, Frode Stenstrøm. I welcome questions and feedback on LinkedIn. I wish better business processes for everyone, and until next time.