Enterprise Decision Management - a Weblog

Fair Isaac's big event - InterACT - is just three weeks away. InterACT has more than 100 sessions on analytics, collections, customer centricity, fraud, collections amp; recovery and more. There are sessions targeting financial services, insurance, retail, telecom and healthcare. Not only canyou hear a whole bunch of very smart Fair Isaac people (including me and Ian) you can hear from a gaggle of outside presenters from DriveCam, Spira de Mexico, HBOS, Stanford University, GE Money, Best Buy, Fingerhut, Dell Financial Services, Air Products, Avnet, IDC, Canadian Tire, WaMu, TowerGroup, CIBC, JPMorgan Chase,ICICI, Microsoft, IBM, Santander Banespa, Westpac, RBC, VFMBBVA Bancomer, Healthways, Tenet Healthcare, Pro-Change Behavior Systems, Celent, Chubb, Millennium Information Services, Property Casualty Insurers Association of America, Westfield Insurance, and the American Insurance Association!

The show also has some great keynotes - Don Tapscott, author of Wikinomics: Competing in the Age of Collaboration; Chip Heath, author of Made to Stick: Why Some Ideas Survive and Others Die ; and Michael Lewis, author of The Blind Side: Evolution of a Game, Moneyball: The Art of Winning an Unfair Game and Liar's Poker: Rising Through the Wreckage of Wall Street

Book now!and remember the blog discount code - CRJT90!

Shameless Commerce ends.

Technorati Tags: InterACT

(Posted by the Guest Blogger, and Host of the Decisions Podcast, Ian Turvill.)

InterACT, the Smarter Decisions Conference, is coming!

One of the most promising areas in analytics involves using transaction-level data to improve decisions. Transaction analytics are already a key part of fraud solutions like Fair Isaac’s Falcon Fraud Manager. Now the question is, how much value can transaction analytics add for lenders who are trying to manage rising bankruptcy and bad debt?

That’s the issue Brad Jolson will discuss at InterACT in San Francisco. And we have a preview recorded in February, when Brad presented on Predicting Bankruptcy using Transaction Analytics.

To sign up to attend InterACT, please click here. Don't forget the exclusive EDM Blog discount, using code CRJT90 when registering.

Download: DecisionsPodcast.No.17.mp3 (8.9MB, mp3)

Over the last couple of days I have been pointed towards a couple of items that made me think some more about personalization and the fit of enterprise decision management, EDM, with the kind of targeting and focusing that personalization requires. Firstly a fellow author (hi Ann) sent me this link Epic 2015 - The Future of Media? It is quite long, but definitely worth watching to the end. The piece basically predicts (though that may be too strong a word) that media will become more and more personalized over time. I am not going to get into the general rights and wrongs of this - the video does a good job of portraying them - but the video talks a lot about how technology can be used to deliver targeted and personalized content. Secondly Seth Godin also had an interesting piece on personalization - Ego - that my old buddy Ken sent me. Lastly I had found an interesting piece by Jack Vinson about relying on your network. All of these came together to make me wonder about personalization.

EDM is ideal for personalization because it allows you to combine rules (for preferences, simple control of interfaces and content as Seth suggests, explicit recommendations) with analytics (for predicting interests from past behavior or for grouping people by like interests). Furthermore it assumes that the right personalization is not static but must be evolved and changed, challenged and reviewed using an adaptive control approach. By focusing on the personalization decision as a separate opportunity for targeting customers it also supports a multi-channel world, bringing data from all channels to bear on a decision that can then be intelligently applied to all channels. Because personalization decisions are typically micro-decisions, this is critical. There are lots of examples of this:

• Allowing customers to specify rules for what kind of offers or content interests them
• Allowing marketing to specify rules for cross-promotion based on analysis
• Predicting likely future behavior to personalize a retention offer
• Predicting which menu items are most likely to be interesting to take best advantage of a small screen
• and so on

The good news about personalization, as Seth noted, is that you can engage people more readily and make content, offers, products seem more relevant to them. People like to be recognized and acknowledged. There is a downside though, as noted by the Epic video. The downside is that you pre-filter what someone sees. Even if you do this using people's network as Jack noted or by using the power of networks to find those things likely to be interesting to you (as suggested by Chris Anderson), you still start to narrow down how people see the world. If you recommend products based on what you know about someone, or based on what they tell you they are interested in, they won't see the full range of products you have. If someone sees book recommendations based on what people like them read, you will have your interests reinforced and not broadened.

Me, I think personalization and targeting are both valuable to people (who seem busier than ever) and essential given the rapid growth in information and product options. I think we need to be careful though as we consider how personalization affects mass media, news and more. you are interested in some of my previous thoughts on this topic, I gave a presentation on extreme personalization a little while ago and blogged on how to scale personalization. You might also check out one of my favorite posts on hits and niches, a post that resulted from reading The Long Tail.

Technorati Tags: adaptive control, business rules, EDM, enterprise decision management, Jack Vinson, micro decisions, personalization, predictive analytics, Seth Godin, Chris Anderson, the long tail

(Posted by the Guest Blogger, and Don LaFontaine Devotee, Ian Turvill.)

InterACT, the Smarter Decisions Conference, is coming!

One of the areas many financial services companies are exploring is customer-level management. There are a host of benefits to be gained by moving from managing at the account level to managing at the customer level. And one of the keys to customer-level management is developing customer-level scores.

At InterACT in San Francisco, Susie Upton, Head of Unsecured Risk and Analytics at Australia’s Westpac Banking Corporation will discuss her bank’s move to customer-level scoring: why they did it, how they did it, and what they’re getting out of it

Susie presented on this topic at our European InterACT conference in February, and her talk was one of the highlights of the conference. So we thought we’d give you a little taste of her talk, to whet your appetite for InterACT San Francisco.

To sign up to attend InterACT, please click here.

Download: DecisionsPodcast.No.16.mp3 (9.4MB, mp3)

Don't forget the exclusive EDM Blog discount, using code CRJT90 when registering.

DM Review just published a shorter version of my book review for Competing on Analytics: The New Science of Winning by Tom Davenport. You can read the original reviewhere, buy the book hereand comment about the review here!

Technorati Tags: analytic application, competing on analytics, predictive analytics, DM Review, Tom Davenport

Diego Lo Giudice of Forrester published The State of Model-Driven Developmentrecently. This is a well-written paper and does a nice job covering Model Driven Development of which it says "Forrester expects model-driven development (MDD) to play a key role in the future of software development; it is a promising technique for helping application development managers address growing business complexity and demand.". Readingit i was struck, though, with the lack of attention paid to business rules in this context.

If one of the key drivers for MDD is the need to build systems that can cope with faster business change and to do so by improving business/IT alignment then business rules would seem to be ideal. Not only are they a great way to approach
"change-time" they can also really help with the kind of alignment described in Forrester's Concurrent Business Engineering. This made me wonder what is a model, exactly? Should it include non-graphical elements? Certainly Diego felt it could include "a declarative language for assertions". I would go further and say it could include a declarative language for rules in general for, while improved business/IT cooperation requires a shared language, I don't believe that UML is it. Business rules allow for IT departments and business users to truly work together.

The paper also talked about the reuse of models in MDD, whichseems like a real advantage, and about tracability improvement - better linkage ofrequirements to code. While this sounds great, I fear thatwe are often talking about requirements that are really rules- rules are not requirements and better traceability or better requirements documentation will not help.

Lastly there was some talk of Domain Specific Languages. Instead of using Domain Specific Languages as part of MDD, why not use business rules? There's been some good discussion about this (particular this post) so I won't add anything.

I believe that business rules are a valid component of MDD and that MDD without business rules is flawed. You can check out a previous post with some of my thoughts on MDD and business rules.

Technorati Tags: business rules, change time, concurrent business engineering, dsl, forrester, MDA, MDD, model-driven, model-driven architecture, model-driven design, domain-specific language

Continuing the series on Adaptive Control:

• Why do you need Adaptive Control?
• What's a basic technology approach for Adaptive Control?
• What is Champion/Challenger?
• What is Experimental Design? - This Post
• How do I do Decision Analysis?

Experimental Design is a mature and extremely successful science dating back to the pioneering work of R. A. Fisher in the 1930s. It is designed to generate efficient experiments that will yield results suitable for accurate analysis including understanding causes of variation, predicting how changes in operating conditions would influence the outcome and the possibility of optimization to achieve a desired outcome. To apply this in Adaptive Control you must design Challengers in a systematic way.

Remember, by adding (say) a pair of Challenger strategies, you gain experience in a wider range of approaches. However, you might not pick the best areas for the Challengers - you might choose to vary aspects of the rules or models that are not the most important. Even so, you are likely to move towards the "optimal" approach in fewer steps than if you had only a single approach. Good design for the Challengers is critical, however. Consider the diagram below, which shows a danger with poorly defined Challengers. The best Challenger (Challenger 2) becomes the new Champion but there is no guarantee that it represents a move directly towards the optimal approach. In this case, it implies improvement in a direction that will result in New Challengers that are less optimal. Good Challenger design should involve a constant movement toward the optimal approach because, although an adaptive control infrastructure will correct and move you towards the optimal approach over time, you may not want to wait as long as multiple champion/challenger review cycles might take. If, for instance, you must wait a year to see how well a particular retention approach worked then you cannot afford to drift off the direct path to the best answer. Similarly if the optimal approach changes over time, and most do, then the delay imposed by a "bad" set of challengers may well be unacceptable.

In contrast, if you do a good job of designing your Challenger strategies (by using a good experimental design approach), you can expand your Challengers around your Champion in a way that maximizes the likelihood that you can get close to an optimal approach quickly. This requires both fine grained control over your Challengers and careful design of them so that you can infer likely results “between” Challengers and cover a wide range of possible variations. Below you can see how a range of Challengers can be modeled, using experimental design, around the current Champion to ensure that the new Champion is moving move directly and rapidly towards the optimal approach.

Experimental design requires a general idea of the predictive models being used in the decision and plausible ranges for the control factors. Data are then collected in the most efficient manner to provide sufficient coverage throughout the operating range of the decision, such that the model will yield accurate predictions and optimization results. Prior experience and theoretical insight into a problem help with the task to design the best experiments to allow for efficient, systematic acquisition of data. When designed appropriately, the number of Challengers or experiments required will be minimized.

I realize that this post is more of a call to action ("You should do this") and less of a how-to but it is a large subject and I don't feel qualified to go on about it at any length. There's more on the general approach to designing experiments, and a bunch of useful links, on Wikipedia.

Continuing the series on Adaptive Control:

• Why do you need Adaptive Control?
• What's a basic technology approach for Adaptive Control? - This Post
• What is Champion/Challenger?
• What is Experimental Design?
• How do I do Decision Analysis?

To support adaptive control, your production environment needs to support the deployment of multiple Challenger approaches to the existing Champion while business users need access to simulation tools.

The production environment must allow a Decision Service to have not just its current or “Champion” implementation but also a number of potentially better strategies � “Challengers”. It must be possible to randomly take some of the transactions and run them not through the rules and analytics that are currently defined as the Champion but through one of the Challenger strategies. You need to capture the results, both immediate and longer term, in a way that allows you to know which decisions were taken in which way (Champion v Challenger) when you come to do analysis. Performance management dashboards and Key Performance Indicators need to show both the overall average across all the approaches, the Champion’s results and the Challenger results when the measures differ across them. Thus if a Challenger strategy sacrifices retention for profitability by more aggressively dropping unprofitable customers, measurement reports and dashboards showing retention need to show the results appropriately so as not to mislead those managing the decision. In general, only a small percentage of decisions will be taken with one of the Challenger strategies, the vast majority (more than 90%) are likely to be taken with the Champion.

A simulation/testing environment must be available. This needs to have access both to historical data and to randomly generated data. If analysis of customers rejected in the past can be done to infer what they might have looked like as customers, this kind of data can usefully be added. Then, if you change your customer acquisition or origination strategy, you will be able to do some analysis to see how this might have affected your customer base. Within this simulation environment, you need to be able to do some testing of how different rules and analytics might affect results in various “what-if” scenarios. This might be as simple as running a test Decision Service against the data and seeing what results you get to as complex as running formal simulation and optimization technologies.

Business users, those who understand how the business operates and what its measures and objectives are, must be able to interact with both environments. They must be able to analyze the results both of production Challengers and of simulations. They should be able to design and run simulations for new scenarios and they should be able to design new Challengers and push them to the IT department for final testing and production deployment. This requires a combination of reporting and dashboard technology with rule management applications.

Continuing the series on Adaptive Control:

• Why do you need Adaptive Control?
• What's a basic technology approach for Adaptive Control?
• What is Champion/Challenger? - This Post
• What is Experimental Design?
• How do I do Decision Analysis?

Champion/Challenger is an important concept. The idea is that you identify your current approach as a “Champion” � documenting the business rules and analytic models that together represent your best approach to a given decision. "Challenger" approaches are then developed. Each Challenger differs from the Champion in some measurable and defined way. Perhaps it has different business rules, perhaps it uses a different risk model, perhaps it is more aggressive about retaining customers. Each Challenger will therefore deliver different results from the Champion.

These results may be better or worse but only testing the approaches with real transactions, in a live environment, can really show. A Decision Service is therefore configured to push a small percentage of the transactions through each of the Challenger approaches while pushing the majority through the Champion. Results from the different approaches can be compared and measured over time. If a Challenger does better than the Champion, it can be made the new Champion and the process of identifying and testing new Challengers repeated to continually improve the decision.

The advantage to considering multiple approaches in parallel is shown below. If we have a current strategy and a target, but unknown, optimal strategy then our objective is to move towards it over time. With only a single current approach, we gain experience only in a limited area and so can only move towards the optimal approach within that area of experience. This means potentially many attempts at an approach before we reach the optimal one. When it may take months or years to see the full impact of an action, this may be too slow to be acceptable. In contrast, if a number of Challenger approaches are also defined you dramatically increase the decision space considered and so can move toward the optimal approach more quickly. This increased speed of improvement is enhanced by the more systematic approach to comparing approaches offered by Champion/Challenger.

The first issue to consider with Adaptive Control is why to invest in the necessary infrastructure to improve your decisions over time. Adaptive Control requires software assets and staffing to be applied not to building new Decision Services, but to improving old ones. At first sight this goes against the whole "EDM reduces maintenance" pitch (that I have made here, for instance). It does not. You should not consider improvement of your decisions over time "maintenance". Adaptive Control is about continuously improving the way you make decisions. Some of these changes will come from changing business conditions that force a change in the approach being taken to the decision. Mostly, however, it is a case of making a decision better and better over time to boost profits, reduce losses, or improve retention.

After all, you constantly learn more about your customers and gather more information about their behavior. New insights and market trends come from you, your competitors and from third parties. A process for continual review and improvement of how you take a decision allows you to detect and respond to changes in the behavior of your customers without having to start a special project and helps you show an ROI for the data you collect and analyze. This is complicated by an interesting fact about business decisions - at the point of decision it is not known what the long-term outcome of that decision will be. Thus not only does the best decision change over time but you may not know what the best decision will be when you must make the decision. For instance, the graph below shows how various actions taken today influence your profitability over time in different ways. One causes a steady growth in profitability, one has good profitability but only after some time and the third has a rapid growth in profit followed by a falling off.

This illustrates two truisms about decisions. Firstly if you use a single approach for every decision then you will only plot one of these curves and will have no data about how other actions might have resulted in better (or worse) results. Secondly it shows how important it is to track your results and know what kind of impact you are looking for - short term or long term? Low risk or high risk? Only by building an Adaptive Control environment into your Decision Services can you create an environment where you can manage and improve these decisions over time.

Tomorrow, an overview of the technology infrastructure for this.