SAP BI Roundtable - Thorny Questions Edition with James Taylor of Decision Management Solutions

podcastlogo_jonerp.gif"Discussing the Thorny SAP BI Questions - including business rules integration, predictive analytics, and unstructured information - ERP Lounge #19)
Podcast Interview Date: May 8, 2011
Podcast: Listen Now!

For this pre-Sapphire return of our semi-regular SAP BI podcast series, fellow SAP Mentor and co-host Vijay Vijayasankar and I talked with special guest James Taylor of Decision Management Solutions about some of the tougher BI questions SAP - or any other vendor for that matter - must tackle. Elsewhere, there's plenty of great content on SAP's important BI 4.0 release (I recommend the DSLayered podcast series). But this "thorny questions" edition of the ERP Lounge looks at the challenges SAP faces as we re-evaluate new SAP BI catch phrases like the "Semantic Layer" and "Predictive Process Designer". 

During the 50 minute podcast, we move from a look at 4.0 to a discussion of the importance of business rules and the issues SAP needs to address with separate rules engines for operations and analytics (including two for operations, BRF Plus and NetWeaver BRM). We talk about why predictive analytics matter to companies, and why vendors struggle to deliver meaningful predictive analytics solutions. A lively discussion of the importance of unstructured data catches the three of us jousting a bit on the relative importance of text analytics versus other BI priorities. We veer into a HANA discussion before wisely saving most of that one for a different podcast down the road.

Editor's note- podcast links: Vijay's SCN blog has many topics pertianing to this podcast, as does James Taylor' own blog. Vijay's personal blog, "Vijay's Thoughts on All Things Big and Small," also has some relevant BI content. Also note: I had a minor static issue on the taping during the first 11 minutes of the podcast. That has largely been fixed in production, but the sound quality improves further at the 11 minute mark.

Note: to comment on this podcast series, or send in a question for us to answer in the next one, be sure to join our ERP Lounge Group on Linkedin. If you want to subscribe to the series, get the The JonERP Master Blog and Podcast Feed. Or find Jon on his @jonerp Twitter feed. The ERP Lounge podcasts are also included in the JonERP iTunes podcast feed.  

Podcast Highlights

0:00 - Jon back in the podcast saddle (with semi-regular co-host Vijay Vijayasankar) - Jon's been busy with, but glad to be back on the podcast airwaves. Special guest James Taylor of Decision Management Solutions joins us. We brought James in to bring a broader perspective into the issues of SAP BI. SAP BI isn't the only BI that matters, and a bigger view can help us to make sense of what SAP is doing. 

2:10 Jon to Vijay: We're waiting on BI 4.0, we expect GA announcement soon, what can we expect? Vijay: I'm excited - we've been playing with the rampup version of SAP BI 4.0 for a while, and I'm excited. There is better integration, and SAP has pulled some much-requested features from 4.1 right into 4.0. SAP has been particularly helpful during rampup. VP level folks and higher have been responsive on Twitter to issues and helped to get the questions answered. Jon: the SAP BI team has been particularly engaging with influencers, so kudos to them for that.

4:00 Jon: There's a lot of features in BI 4.0 to talk about, but SAP in particular has been pushing the Semantic Layer as being a very important milestone development, including the self-service user, empowering business users to build their own dashboards, freeing up the BI IT bottleneck. Quick training from Webi-experienced users might be enough to get a business user going on designing Dashboards in 4.0. Vijay: The Semantic Layer has gotten a major upgrade from prior versions of "Universes". Having a common data definition is important. The technology matters, but the data definitions of common terms are important. If you say "net profit" in a large company, you might get 25 different answers. It's not just the technology. Jon to James: Semantic Layer - BS or important evolution. James: I think it is important. There is a lot of BS around semantics, but I do think semantics matter. From a business rules perspective, my question would be, "Well this is great, but when will there a semantic layer where you can use all the development tools SAP has against it, not just the BI tools?"

6:42 Jon to James: You focus heavily on business rules - why is this area worth blogging about? James: My focus is on decision making, which brings me into contact with a lot of BI people. But I look at high volume, transactional decision making, where it's not always practical to put information in front of a person and expect them to make a compliant and legal decision. So how do you help an organization take control of their decision making process? Business rules are a great tool for that, because they are much more transparent and easy to change than code. They become more of a managed base of business logic and as a result, they really help decision making. As soon as you get into call centers and knowledge worker decisions, there's a tension between BI people and rules people in terms of whether you let individuals make the decisions or embed them into systems with business rules.

8:50 Jon to James: So we're trying to take the human element out where we feel we can take it out? James: To some extent, yes, but I prefer to say, "We're trying to help the right person make the decision." Do you want a new call center person to make the decision on how to retain your best customer? Business rules allow you to separate who decides how the decision can be made from who delivers the decision to the customer. In classic BI systems, those two things are merged.

10:00 Vijay to James: I'm really curious to hear how you see SAP's business rules functionality. SAP launched BRM a few years ago, but we don't hear too much about it anymore. James: when most people talk about business rules in SAP, they're talking about NetWeaver Business Rules Management (BRM), which is integrated with NetWeaver Business Process Management (BPM). In my mind, the more interesting of the two products is BRF Plus, which is an ABAP-based rules framework that allows you to manage your business rules in a repository and then deploy it as ABAP code. For a lot of SAP customers, that's crucial. You can go to TechEd and see the popularity of ABAP over Java for SAP customer.

So what I'm interested in is that product, which is starting to be used by the core enterprise applications team. I think that's a tremendously good decision on SAP's part, as it allows customers to rapidly configure these rules while using a standard data set. I'm a bit concerned they have two different rules engines, and then the SAP BI data quality rules look like a third kind of rule. There doesn't seem to be an easy way to access the Semantic Layer from either rules engine, even though that would give rules designers a huge advantage. SAP is doing a lot of the right things, but what is lacking is a recognition that there is a whole cluster of decision management technologies that are separate from process execution and also separate from data management.

13:20 Vijay to James: I completely agree. How to you see NLP (Natural Language Processing) playing into this? If a system can figure out a rule, can it learn a new rule on the fly without a human maintaining a decision tree? Do you see a need for that? James: I'm a bit of a cynic on NLP, because of the question of whose natural language you are talking about. What's more interesting to me is the application of text analytics in a broader effort to use analytics to say, "Here's what the data says your rules should be." Taking both the unstructured text and data fields, these appear to be rules that might allow the creation of new and effective rules from these disparate data sources. The interaction between analytics and rules-based output has tremendous potential and we're pretty much scratching the surface.

15:30 Jon to Viyay: That's a pretty challenging area for SAP to deal with in terms of three different rules areas. Is this ok to have different engines or do they need to be united? Vijay: For a company the size of SAP, it's not surprising to have more than one set of rules technologies, and there might be technical reasons as well. But eventually, convergence is the right thing to do. The users don't care. To maintain SAP, you tend to need Java, Web Dynpro, BSP - no matter where you have multiple technologies it's harder to maintain systems and train teams. It's not an easy situation but I understand why it is.

17:30 Jon to James: Let's venture into predictive analytics. SAP is sick of me asking them about it, so let me ask you guys instead. Why does predictive analytics matter to enterprises? James: It matters for a couple reasons. First, without it, you are driving without a rearview mirror. Any action you take doesn't have an impact immediately, it has an impact in the future. If you can't determine what the future will look like, then you don't know what action to take. Predictive analytics turns the uncertainty of the future into a useable probability. You can now act in light of that probability, and over time, that gives you a much greater chance of success. It particularly matters in front line disciplines. In insurance, you know there are some false claims.

Predictive analytics helps you to predict which claims might be fraudulent. Using those rules, you can predict possible outcomes. Predictive analytics are embedded in a way that data visualizations are not. Machines can't look at visualizations, they can only look at mass, so predictive analytics can be embedded - that is tremendously powerful. It's a change from what can we put on the glass to what can we embed in operational systems. The power of predictive analytics to bridge the gap between analytics and operational systems is critically important.

20:50 Vijay: I couldn't agree more. It's not even a new technology itself - the statistical methods on which predictive analytics are based have been around for many years. But: only a small percentage of SAP shops are using this technology. The companies that are using it are not shouting it from the rooftops because they don't necessarily want to clue in others on what they are doing, for example in fraud detection. James: one customer at an IBM event said, "I think you should not adopt predictive analytics anytime soon, because I really like the competitive edge they are giving us." Vijay: Even for common use cases like the storage of sales orders, there are mathematical models than can be used to predict possible outcomes. Call center individuals can be clued into up-sell and cross-sell opportunities by using these models, so there are use cases.

23:40 Jon: I'm going to try to explain SAP's predictive analytics strategy and you guys can comment on it: SAP has said they have two major activity threads. One is they are working on some HANA-based scenarios (though they will not be limited to HANA), where they will use the R open source statistical language, and they are also looking at using fuzzy logic and IQ to expose predictive analytics via in-memory. The other approach is one using a process-driven approach with a new tool called the Predictive Process Designer that they should be showing off at Sapphire Now this year. They are targeting Predictive Process Designer for release at the end of the year, but they are looking at embedding it into SAP BI rather than using it as a standalone workbench, though that scenario may be possible also.

25:05 James: Embedding R into HANA is a perfectly acceptable approach. There are companies out there who have embedded R for a while, and it's not a small amount of work to do it. Fuzzy logic is something Sybase has been used for a while. The challenge is that neither of these approaches is presented clearly as an approach that can be embedded in the BusinessObjects stack that can take advantage of the Semantic Layer, nor are either of them integrated in an obvious way into the Business Suite. The danger is producing predictive analytic models that don't go anywhere. There can be a huge struggle to get models out there that are embedded properly and that get adopted by actual user. Predictive analytics don't tend to impact the business until they are embedded into decision making.

I still don't quite understand who enterprise application customers are going to be able to consume those predictions in a useful way to make better business decisions. As far as the other products are concerned, I suspect that will be about using analytic techniques on processes to improve process outcomes. Vijay: combining predictive analytics with HANA is a neat thing to do. In terms of use cases, I haven't heard a lot from SAP personally on this yet. HANA can understand any data as long as you can make a model out of it. You have to integrate these tools in an intuitive way, and there's where a few focused use cases can come in handy. SAP has a huge amount of process knowledge, and combined with SAP's new "design thinking" approaches, there is a lot of potential. But: I'm hoping to learn more at Sapphire from specific use cases.

30:00 Jon to James: You're not going to be at Sapphire this year, but let's say that you were: what would you want to ask SAP BI executives and product leads? James: The first one is clearly, "What are you going to do about integrating BusinessObjects and the rules-based infrastructures on the BRF Plus and BRM side?" I would also want to hear, "What are the use cases for R in HANA and IQ? What use cases do you have? How are you making it easy for companies to use this technology?" All the big analytics players are all focused on the deployability of their models. You have to make it easy for companies to improve and change their business. If it can't do that, it doesn't matter how good the model is.

32:10 Jon to Vijay: Without stealing your Sapphire thunder, what kinds of SAP BI questions are on your mind this year? Vijay: in the context of predictive analytics, I think this would be a good thing to do with customers who have been asking me, "Where are the innovations from on-premise solutions?" A lot of good stuff is going on with on-demand and on-device, but what about on-premise?" Predictive analytics that are embedded, rules that are embedded - these would give big bang for the buck for SAP customers. They can move on this more quickly because they are already deeply invested in these on-premise systems - this is where the money has already been sunk. James: I would agree. All the big vendors have had a whole separate analytics stack, and separately, an operational stack. We have two stacks because thirty years ago the mainframes couldn't handle both at the same time. There was never a coherent reason. So we have a massive BPM and operational market on the one hand, and an analytical market on the other hand, and there was never a reason to separate them - so where's the analytical innovation in the operational stack? That's where the money is - in these massive transactional systems.

35:10 Vijay: I have a similar question on the HANA side - most SAP shops were sold R/3 with a promise that you never have to worry about table and field level information because it's all shielded from you. With HANA, you do have to know about tables and fields, and you have to know that at the database level, not the Data Dictionary level, so you have practically no abstraction. This is not the long term vision, it will change over time. But wouldn't it have been more value-added if HANA already supported ABAP? SAP-centric shops have moved away from SQL, as you don't need a lot of SQL for ABAP. With HANA, you do need more SQL know-how. This means SAP shops won't be able to use the ABAP skills they have for HANA in the early releases.

37:56: Jon: While we're on the thorny questions themes, what about the struggle with unstructured information? My theory on unstructured information is that there is a huge amount of valuable information that is not transactional. It might be intellectual knowledge, manifested in email threads. The socialization of the enterprise makes these problems more difficult, with more open channels where potentially valuable information is posted but then buried. I've helped smaller companies leverage this kind of information but it's much harder to scale that. The last time I asked SAP about this issue, they explained that they had tackled a fair amount of the text analytics challenge but that they hadn't done a good job of promoting their existing solutions. Unstructured data is the holy grail of ERP systems.

Vijay: If SAP has solved this issue, the world needs to know about it. There's a lot more to explore there. The unstructured world is not easy to grapple with, it's much more they keywords, you need to understand context, multiple languages - even puns. Then there are the rules that need to be inferred from this unstructured information, it must be an integrated solution. I don't think any vendor has done a great job on this, though I think there is a lot of good research going on in this area, but I'm not sure companies have solved this issue yet.

41:04 James: I'm a bit fed up with this issue because I'm sick of hearing that 80 percent of the value of a company's information is tied up in unstructured data. If that was true, a database would have been created for it. The vast majority of companies are not taking advantage of the structured data they already have. Looking at what prospects actually DO versus what they say they are going to do is one example of why I prefer an emphasis on structured data. Solving unstructured data issues doesn't make a lot of sense when companies haven't done nearly enough to take analytical advantage of their structured data. Most companies are "doing analytics" in the form of reporting. They aren't typically doing predictive analytics, data mining, or tying data into decision making. If they aren't doing any of these things, why are they asking about text analytics when they aren't even at a baseline competency with predictive analytics? Vijay: I have a similar view of using Twitter as a sentiment analytics channel. What happens after that? Twitter is not an easy place to do customer service or solve the problem on Twitter. Which raises the question of how much money should be put into social media analytics if they can't close loop it.

Jon makes his case for the value of unstructured information - he doesn't want to let the vendors off the hook on this issue. James: I do see some use cases for it, but you need to be at a level of sophisticated analytics to get there, and customers are getting distracted by social media and mobile analytics. Helping companies make better decisions with their existing enterprise reporting tools is much harder.

48:00 James: Companies like SAP should invest in this problem of how to help operational systems integrate more effectively with analytics systems. A typical CRM system is as dumb today as when you put it in. Vijay: SAP is probably doing the right thing about how to solve LOB (Line of Business) problems like the large enterprise on-demand team that was under Wookey is doing. If you ask a business user the right questions, in terms of what happens with the reporting result down the line - I'm not sure any vendor is asking that question. James: I'm not sure the new cloud vendors are asking that question either. Vijay: It has to be a complete business process, it can't be a point analytics solution. Compartmentalizing these things is not the best solution. James takes his exit from the podcast, Jon and Vijay wrap.

52:00 Jon and Vijay wrap: that was an entertaining convo and a fresh perspective. We spent a lot of time talking about difficult issues, but they are meaty and important for both customers and vendors.