The NeoFusion Blog

Leveraging Business rules with Predictive Analytics on a Real Time Project

2011-05-27T06:08:00.000-07:00

Centralized decisioning and it is use on real-time micro-operational decisions has become a must on today’s enterprise strategy. As the core of this technology is the business rules management system (BRMS) that allow business users to create, modify, deploy and optimize their business logic without or with a minimum intervention of an IT department.

Business rules are basically made of the logical if…then…else declarative statement where, on an enterprise-wide project, the outcome might be a large set of decisions. Decisions over decisions. A series of attributes must be needed to filter out the final choice. An example of these attributes are the business performance goals as increasing product margin, reducing customer churn, reducing customer fraud, etc. where one, or a weighted combination of all, drive the final result. All these ‘goals’ are a series of business traditional dimensions, that not always provide the most optimal result. Setting the margin or price as the main measurement to prioritize an offer might result in a low acceptance rate, and setting the lowest cost offer as a goal might result in offer cannibalization.

So real-time smart decisions are here, but where is the intelligence to support them? Here is where Predictive Analytics steps in, not only to look at the past and respond the why, but to project into the future and respond where. Adding predictive analytics to business rules improve business results by adding fact supported and educated insights to any decision. Of course, the GIGO paradigm applies here as on all data related process. Modeling the data to predict an expected behavior about a customer, product or transaction is an important service to be used to support a decisioning. The predictions are made in the form of propensity or likelihood scores of a customer accepting an offer, a transaction being a fraud, a customer churning, or a customer defaulting a loan to name some.

A model score can be returned from different sources:

Offline scores: the models are executed in house or by a third-party provider and assigned as an static attribute that is fetched during an interaction.
Inline scores: the models are pre-defined as declarative statements and scores calculated on real-time. Examples of inline scores are scorecards and decision trees
Response-based adaptive models: The model score starts as a preset value and it is automatically fine-tuned overtime based on the historical recorded response at to the moment of the decision.

There is no prescriptive rule about which method to use during the decisioning process, and is more related about the enterprise sophistication and how are their resources prepared to handle Analytics as business as usual. Most of today’s analytics platforms have their business counterpart that take the complexity out from a basic analysis, and let the business user generate insights without a high degree knowledge of statistic. But again, the answers are always as good as the questions.

In summary, your real-time project can be optimized when the power of analytics is added as a support factor to other business goals, or as an standalone factor. In a future post, we will cover how all business attributes and goals can interact in real-time to drive a decision.

By Leo Guim, NeoFusion

Beyond “Go Live”: Making Your Real Time Marketing Implementation a Success with Continuous Improvement

2011-02-28T09:23:00.001-08:00

What defines success? Imagine that your organization has decided to implement a real time marketing system to support a cross sell/upsell business case by intelligently interacting with its customers. As this is a hefty business and IT investment, the organization wants to spend the time and manpower to get everything as perfect as possible before going live to ensure that the return on investment is immediate. Your organization spends months perfecting the business rules and building the initial predictive models to drive the intelligent decisions of your real time marketing solution. The system is successfully deployed on time and under budget. This sounds like a successful implementation, right? Initial reports indicate that the marketing offers in the system are not effective and have a low rate of acceptance. Employees using the system complain that the system does not make recommendations for most customers. The delivery of the return on the original investment is delayed from 6 months to 3 years. Success is not just about going live with your marketing system. Using Continuous Improvement techniques, your organization can ensure that your real time marketing implementation provides an immense return on investment in a short time frame and is a true, long term success.

Deploy Quickly

Many organizations make the mistake of spending effort trying to perfect the system before launching (usually called a “Big Bang” approach). What these organizations fail to realize is that the system always changes drastically based on (customer and employee) feedback and learning during the pilot window of 1 to 3 months after go live. All the time and effort that the organization has spent in the initial project is wasted. An organization may find that their users do not actually use a function of the system as intended or that the customers react unexpectedly to a marketing offer.

A quick deployment (for phase one) is paramount. Organizations should spend less time to get the system live but spend the time and effort ensuring that appropriate data collection and feedback loops are in place to improve the system. Deploying with a pilot group (a subset of users), a subset of marketing offers, and slowly rolling out to all users reduces time to launch.

Input!

Time and effort should be spent on putting mechanisms in place to collect information and feedback once the system is live. This is then used to refine and improve the system. Too often reporting and simulation are not fully thought out and implemented. A proper reporting mechanism is the single most important tool in knowing how effective the marketing system is (example: tracking offer presentation through to fulfillment). Simulation is the only way to safely investigate performance and tune marketing offer performance without actually involving live customer interaction. Other techniques can be used to collect information such as rules monitoring (what rules are firing and when), user surveys, customer surveys, etc.

Remember that the goal is to:

- provide a feedback loop that collects information

- create a BAU (business as usual) process to apply the feedback as updates (during a reoccurring interval like monthly for example) to the real time marketing system

A Champion Rises

Before an organization deploys a real time marketing system for the first time, your team may notice that one member of the project team (direct member or indirect) really “gets” the advantages and identifies the wasted potential of a real time, inbound marketing system. They start asking tough questions as to why certain customer segments are being excluded (as they understand that these represent massive potential revenue streams) or question the justification of a retention strategy over single vs. multiple channels. The organization should recognize the value in this person and allow them to lead the effort to move the real time system from a “project” during the first release to become a “business as usual” process for all future releases. They become the champion that helps to fulfill all the promises (and create new opportunities) as envisioned in the original business case.

Organizations should spend as much or more effort ensuring continuous improvement than implementing their real time marketing system for the first time. This is the only way to ensure that the system is a success in the long run rather than just a single, successful deployment that was on time and within budget.

Benefits of Virtualization During the Build Phase

2011-02-08T12:25:00.000-08:00

Benefits of Virtualization During the Build Phase
by David Russell, NeoFusion

Virtualization is the process of working with a virtual development environment that includes a self-contained operating system, application server, and database among other components, which can be opened and manipulated within a running operating system. Commonly known as an image, it represents a snapshot of a complex development environment that can be managed on a developer’s laptop without having to go through the hassle of manually installing and configuring all of the necessary software. The developer simply opens the image, and can start working in the environment immediately, completely bypassing the actual client environment.

Images can be extremely powerful, as it distances the developer from many of the time wasting interactions that take place when using an environment that is shared by potentially dozens of people. During the build phase, it is standard to expect this lost time to be absorbed by developers, making tight deadlines even tighter. The ability to regain any time during the build phase will contribute directly to the quality of the design and build, as well as reduce the additional hours of overtime needed to meet challenging deadlines. We will cover in the next sections specific advantages that an image can provide.

Removal of System and Network Dependencies
Working in an image that contains a fully functioning environment avoids the need to ever interact with a shared server, network, and network database and all the problems that come with that. This means no more delays due to the server being bogged down, networks that have slowed to a crawl, or databases that have gone down. For anyone who has worked on development environments at 3am when the server goes down, this is monumental.

Work Anywhere, Anytime
Virtualization means that productivity is no longer tied to physically being on-site, or being tied to a client’s VPN while working remotely. VPN access alone can be a serious issue, as many companies are reserved in providing VPN access to contractors. Even when it is provided, it can be slow or provide limited access due to network issues, significantly impacting the ability to be fully productive while being remote. Virtualization allows a consultant to work completely independently anywhere, including in a different time-zone, on an airplane, or in an airport while waiting for a delayed flight, giving the chance to claw back more time that was previously unavailable for productive work.

No Blocks from IT to Fix Environment Problems
With a confirmed working environment in the initial image, any environment issues that occur that cannot be quickly resolved through basic troubleshooting can be simply resolved by extracting the latest code base from the problematic environment and importing it back into a fresh copy of the initial image. The alternatives are time consuming: either spend hours troubleshooting an environment that you may not be an expert in, or create a dependency on IT to make time to troubleshoot the environment. The second scenario is even more problematic if the developer is remote, and even worse if they are working remotely in a non-optimal time zone. Being able to revert to a stable image removes the need for this dependency on IT and the time and effort to arrange for their time.

No Blocks from Other Developers Integrating Changes
One of the key benefits of working in an image is that you are the only person working in that environment. As a result, there is no concern over other developers making changes that would de-stabilize your environment. Put another way, there is no worry of another developer putting in bad code, then going home for the night, leaving you to spend time to either debug code that you don’t understand, backing out the changes, or trying to track down the person who made the changes in order to figure out how to fix it. In the same vein, some of the most frustrating issues to troubleshoot are when another developer modifies database data that you aren’t even aware of. This loss of countless hours can be avoided as no other developers can modify your image while you are in it, allowing you to completely focus on hard development.

Work Incrementally Without Breaking Others
Knowing that others will not be making changes to your environment, and that you will not be messing up anyone else by committing changes, allows you to develop code in a more incremental approach. This allows the developer to work and test at their optimal speed, without the worry that the changes made will de-stabilize the environment and prevent others from working. Being able to check in code incrementally allows testing at a more granular level, without the need to make all of the changes at once before checking them in.

Integrate with Real Data Instead of Simulating Data
An effective image should not only contain a fully functioning database, but also include a subset of customer data has been imported into this database allowing integration with real data. The alternative would be to either work with the on-site environment and all of the related dependencies and issues (as discussed earlier), or to build standalone and to simulate data while testing components. Data simulation or “dummying up” data, does not remove the need for testing with real data, but adds an unnecessary additional step. Testing directly with real data removes the need to perform any data simulation, while performing the important task of data integration early in the build process.

Listed above are a few of the many advantages of working in a virtualized environment. Not only does it provide an effective means to work remotely, but saves precious time during the build phase by eliminating key environment issues and the associated time needed to resolve them.

Utilizing Standards in Developing Business Rules (Decision Logic)

2010-10-26T22:01:00.000-07:00

Real time decisioning is so complex because it blends strategies and tactics from such a wide variety of functional business areas. The decisioning logic or business rules are essentially the product of the entire real time decisioning effort, the brain if you will. As such, all those strategies and tactics can add up to pages and pages of rules. Fusing them into one interface and language can be a daunting task. The real challenge is that the strategies and tactics themselves are constant in flux.

So, the question is how do you control (or at least contain) such a dynamic, complex and voluminous mix of business rules? This question is applicable to both the initial implementation as well as the ongoing “business as usual” maintenance.

The answer is standards. All organizations need a set of standards for their business rules. These standards would naturally supersede any language standards or frameworks that come with the enterprise platform of choice. Why? Because the enterprise platform tool was built to provide a framework of code to build any real time solution and is not likely to address your specific industry application of “best offer” or “intelligent work routing” or “customer retention” or whatever your scenario may be. The reality is that most enterprise decisioning platforms have a business user-friendly interface for developing business rules, which in many ways is great benefit. However, the power of that user-friendly interface comes with a specific responsibility to adhere to rules that (unfortunately) the user needs to create and enforce themselves.

So, what are some good standards to adhere to? Well, this list isn’t meant to be comprehensive but it is certainly covers the absolute basics to developing quality business rules (i.e. decision logic).

Separation of Rules & Data

The separation of rules of data is one of the best ways to streamline decision logic. This is because the methods used to perform these two actions are incredibly different and in many ways should be done sequentially rather than at the same time. Meaning, it makes good sense to prep the data before it is used. Further, if the data can be prepped in one space and referenced in another it usually cuts down on redundancy. Its easy to get caught in this trap as we see all too often organizations will choose to hard code data into the business rules because it seems like an inexpensive alternative to proper data integration. However, in the long term this approach usually ends up costing much more in maintenance and errors.

Expression Complexity

The complexity of expressions is sometimes unavoidable however in practice it is best to keep decision logic as simple and readable as possible. This measure is meant to draw attention to areas of logic in which the expressions are incredibly complex or voluminous because those areas would naturally be a pain point for the “business as usual” teams, even if they may be unavoidable. If an expression is unavoidably complex then ensure that it is isolated and commented well by the developer. In our experience there are only ever very few core expressions that are complex so keeping them isolated and well documented provides good value vs. effort.

Visual Organization

Many enterprise decisioning platforms support visual management of business rules or decisioning logic. At times the ability to group logic in different layers and place components freely on the screen can easily become a pain rather than a blessing. As a best practice the visual organization of the logic should be as easy to read as possible and be structured (whenever possible) in a sequential and logical flow. When this is not done with care a relatively simple bit of logic can seem incredibly complex simply because of a poor visual representation. When visually organizing business rules consider what it would take to demonstrate to an audience with no skill set in the tool what is actually going on in the logic.

Redundancy of Operation

Redundancy of operation is essentially doing the same thing again and again within the logic. Examples would be writing the same expression twice or more, using two rule components when one would suffice, etc. It can be extremely easy to allow this to happen in larger implementations when multiple individuals are working on different subsets of business rules. The tried and true fix for this sort of thing is to follow object oriented code principles. A practical way to enforce this is to always have one developer oversee the architecture of the entire business rule development effort.

Diversity of Enterprise Platform Frameworks

The enterprise decisioning platforms available today, generally speaking, are extremely robust. They typically have a wide variety of ways to create an IF/THEN statement, predictive model and combination of the two. Developers should ensure that they embrace a wide variety of these components within the greater framework. All too often we see that the early developers may have not fully understood the entire toolset and thus struggled to develop elegant business rules. The point here is don’t develop three pages of business rules when they can be written another way in only one page.

These set of standards are so basic and fundamental they are applicable to virtually any set of strategies or tactics that require translation into business rules (decisioning logic). Though simple in concept, achieving these standards can be extremely challenging. That said, we’ll cover what skill sets we think are critical to a competent enterprise decisioning architect in a separate post…

Simulation is a must have capability

2010-10-21T06:29:00.000-07:00

If you are starting up a real time decisioning implementation do yourself, the project and your organization the justice of including an entire workstream around simulation. Why? Well, not having the ability to simulate your enterprise decisions is (in many ways) analogous to not having a window to see out of, an altitude gauge, radar, etc. while flying say a Boeing 747. If your flying 50 tons of aircraft you probably want to know (or have an idea) what is going to happen when you when push a button, turn the wheel, etc.

To say simulation in the real time decisioning space is a big topic is probably a huge understatement. However, for the purposes of this post we believe there are several areas of focus that are simply must haves for any real time decisioning implementation.

Micro & Macro results. The solution must provide users the ability to analyze results for individual customers (Micro) as well as large groups of customer aggregations (Macro). They each have very different purposes. Micro generally is used for basic understanding of the fundamental operations of your decisions i.e. how a customer is treated and why. Whereas, on the other hand, Macro results allow the users to understand the impact the entire decisioning program has on the organization across relevant dimensions e.g. financial, inventory, utilization, etc.

Integration with Business Planning. Whether you’re talking retention, cross sell, work routing or whatever… the realty is that the business logic that governs your enterprise decisions impacts your organizations’ assets. Naturally, this means that before your system goes live everyone involved should have a good idea on the magnitude of that impact. This sounds like common sense but I’ve not seen it done properly to date. For example, if you’re dealing with a retention implementation in which your organizations is literally investing money into targeted customers during the interaction then the system is literally giving away money. In this cases the business planning integration would be with Marketing Budgets & Finance. The system must be aware of budget allocation and burn rate in a proactive and controlled manner.

Accuracy. This is largely about the data that is used within the simulations. The main challenge here, especially in large organizations, is to have simulations run on environments that are identical to the production (or live) environment. This can be an enormous challenge (and thus expensive) due to the complexity of the technical infrastructure in a production real time environment. HOWEVER, back to the Boeing 747 analogy; would you pilot a plane like that if you couldn’t see, didn’t know your altitude etc. even though those instruments may be costly? Of course not.

Proactive Business Process. Ensure that simulation is embedded in the change management process of your enterprise decision management. Tactically, this means that no change (e.g. to fundamental rules/logic) should ever be put into production without a view from a simulation activity. This will ensure that the organization has even a chance to understand what changes were responsible for positive vs. negative results. All too often simulation is a reactive activity that is invoked after a disaster or severe issue to try to understand the cause. Specifically the result of doing this bit correctly would likely include things like regression and unit testing on a standard set of data.

Build a Control Group. Lastly but certainly not least, ensure that the project creates a sustainable capability for a control group. Without proper experimental design principles (such as a statistically relevant control) the project will struggle to understand and report on the ROI of the system. Additionally, control groups are fundamental to the organization’s ability to improve customer interactions over time.

The above points are not intended to be exhaustive. However, these areas are key in having an effective simulation capability. That said, all to often simulation is an after thought because, from a pure functional perspective, you don’t have to have simulation capability to go live. Similarly, you don’t need to be able to see to fly but it can be fairly dangerous to do so…

Welcome

2010-08-19T21:02:00.000-07:00

Welcome to the NeoFusion Blog!

NeoFusion is a premier strategic marketing consultation and technology implementation services company. We specialize in offering a global center of experience, excellence, and fostering an effective implementation of marketing strategies and solutions. We provide unique value to companies that want to solve complex marketing challenges. We understand what it takes to successfully deliver valuable solutions.