Making Your Business Intelligence Team Agile

So, you think you can’t make Agile work within your business intelligence team?  Think again.

Agile business intelligence isn’t new; many people have written about it.  It’s mature enough now to have a smattering of books lining the shelves. What are these books based on? Experience. People are executing it and have been for some time.

At Daugherty Business Solutions, we have a group of those experienced professionals on our team. In fact, we have worked with a large telecommunications client for fourteen years using parallel Agile sprints to support enterprise financial performance reporting. The group in place has evolved from large teams to manageable teams of eight, and members have cross trained between ETL and data visualization tools.

The evolution of these teams was not an accident, but instead an orchestration of the Agile methodology. Some people may think of the term Agile and wonder if that applies to business intelligence in the same way that it does application development, and there is no doubt that there is a difference between the two. But, the specific factors differentiating these doesn’t need to be standing in the way of organizations realizing the benefits of using Agile principles to bring more value to their customers, more quickly, and with less overhead.

So, how do you do it?  Before making that determination, there are two key things that need to be identified.

First, data projects are different. You have to acknowledge it.  Sing it from the rooftops! Just like we cannot use exactly the same team and methods to successfully execute a Waterfall application development project and a Waterfall data project, we certainly can’t use the same team and methods to execute an Agile application development project and an Agile data project.

For example, Team Daugherty is working with a cable company’s business intelligence group to build Agile capabilities into their current teams. As we launch projects with them, we’re working through our planning sprints and introducing an additional sprint to allow for some of the coordination and planning required when implementing projects. This is important when using full enterprise data stacks and fully mature data warehouses. By planning more up front, we’re able to fully implement Agile practices, and we are seeing an extremely positive response from our business partners.

Second, because information management projects are different and have different players, characteristics and challenges, we’re going to have to let go of some of our purist notions of Agile.  Think for a minute about this question: what is Agile at its core?  Is it ceremonies?  Is it a product backlog? Is it cadence? Agile is not one of these things. These are all things we do within specific Agile methodologies. To be Agile is to generate value for the business in a consistent, timely manner. The rest are just rules.

Does this mean that we want to throw all of these guidelines out the window? No! What it does mean is if a particular rule is preventing us from achieving our goal of realizing the benefits of Agile, we should evaluate it for validity within the information management discipline and weigh that against how that particular rule helps teams realize specific Agile benefits.

This is true for a number of the projects we are working on right now. One that is top of mind is a project that has Daugherty working with the business intelligence group at a retail client to define how Agile principles will be applied on their teams. As we know, one size does not fit all. In this IM project example, unlike application development projects, it does have some stories that must be executed in a specific order. Working together with the client’s teams and product owners, we’re redefining how the stakeholders prioritize features and the value produced in each sprint, knowing that some of our work items must be executed in a specific order. It may be a specific Agile rule that work items are designed to be interchangeable and can be re-ordered at any time, but we know that, in this situation, that is not what is going to lead us to be the most successful.

Just remember, we’re going to have to let go of “purist” rules and methods, weighing them carefully against the value they bring. Through our work we know that it has to be okay to sequence some work items in a specific order to deliver value to the business.

When working as a part of an Agile team at Daugherty Business Solutions, I have learned that we value “value” more than rules.

Driving Customer Loyalty Strategy Through Data Analytics

With the ease of online shopping, the pop-up ads sharing news of sales and low prices, and the help (or harm) of online reviews, many companies are feeling pressure from their competitors. The challenge of gaining return customers is at the forefront of the battle for survival and many organizations are turning to loyalty programs as a way of incentivizing their products and services. There are a number of different loyalty programs out there, however, the most commonly used ones measure the amount of customer engagements (number of visits, money spent, products purchased, services used, etc.) and allow customers to accumulate points for being a loyal customer. As an Information Architect, I spend my days helping clients to understand analytics and insights resulting from these programs. Let’s take a closer look at some examples and their affects.

Customer Loyalty Reward Types

Not all customers are profitable; similarly, not all rewards lead to loyalty. When measuring the success of loyalty programs, it is important to take a hard look at various rewards and measure which ones lead to the best level of engagement.

Some loyalty programs have simple rewards. For example, Kroger, the grocery chain, provides instant rewards to customers in the form of a discount for being a Kroger Plus customer. There is no process of accumulating points and redeeming them.

On the more complex end of the loyalty spectrum are the organizations that utilize programs in which customers accumulate points that they are then able to redeem for different types of rewards. You may be familiar with some of these programs. One example is the program in place for American Express customers. These users accumulate points simply by using their American Express credit card. The points they earn can be used for travel expenses (flight, hotel, etc.), gift cards, merchandise, or the customer can choose to receive cash back. Similarly, airlines such as Delta and Southwest, give you a host of redemption options to choose from just for choosing their airline. Another example of this type of loyalty rewards program is Dunkin Donuts offer of free beverages for the customers that buy their products most often.

Evaluating Rewards Using Data Analytics Techniques

Assuming your company offers a variety of redemption options, the most accurate way to measure the effectiveness of the different options is by conducting a pre-post analysis using matched pair design, a special case of the randomized block design statistical experimentation technique. More simply, individuals redeeming their rewards are divided into pairs of similar customers. Well, what does that mean? The way you define “similar” is based on your specific definition. Generally speaking, you will match customers using attributes such as age, gender, loyalty class (basic, premium, etc.), tenure, market/location or previous revenue spend. When creating these matched pairs, you will ensure that while the customers are similar on all other attributes, one of them has redeemed the reward you are evaluating while the other has not. This essentially gives you a control group of customers. The next step is to measure the customer’s behavior before and after the redemption activity. The measurement can be revenue, number of visits, number of trips/miles, etc. If you have succeeded in creating these pairs, you should find that the pre behavior of the control and target groups will be similar. You will then be able to observe the behavior post redemption and measure the lift the reward/redemption generates.

Potential Pitfalls

Although this form of evaluation does give you measurable results, there are many factors that you need to be aware of. For example, you could introduce sampling bias if you start by choosing the wrong group of customers. Many large corporations have tens of millions of customers in their loyalty program and only the active ones are maintained in a readily available database. This makes sense for different types of analysis as it improves the speed and performance of queries. However, if you start with this set of active customers, it could introduce sampling bias as you will probably miss customers that moved toward other vendors/loyalty programs after using a certain reward. In this instance, my recommendation would be to use a random sample from all customers, both active and inactive.

Other factors to keep in mind are the enrollment date of customers and the period for the analysis. You want to avoid customers that have only been members for a partial time period. The recommendation here is to choose customers that became enrolled before the date of the analysis and ensure that specific pairs of customers have similar tenure, as tenure can influence the level of engagement in certain industries.

In most cases customers do not only engage with one type of reward. In order to ensure that the analysis is specific to the reward being evaluated, its important to try to choose customers that exclusively participated in the given reward.

Linking ROI to Various Customer Loyalty Rewards

The result of your analysis will give you the metrics you are looking for relating to the effectiveness of each reward and the amount of lift it is generating. You can conduct a bit more analysis to compute the incremental revenue as well.  For example, if you know the dollars spent per month by a group of customers you can calculate the average revenue in the post-reward period for the control group customers and compare that value to the target group customers to calculate the percentage age increase due to the redemption. If the number of customers is known, you can now compute the total incremental revenue.

There are many online resources that explain how to compute lift or incremental revenue, but it’s important to understand your unique situation and business problem, discuss the approach with the stakeholders to whom the results will be presented, and take the approach that best fits your situation.

Conducting an analysis will reveal the types of rewards that will likely be most profitable and what makes sense for your company and for your customers. Once you have made that determination, you may want to begin to do more broad market research into the different reward types to see if you can identify any trends. You may also choose to create a customer profile to determine the most profitable types of rewards and to understand the types of customers that take advantage of these rewards. Your findings will help to inform your customer acquisition strategy.

You Know Who Will Respond…Now What?

As with any analytics project, understanding what the numbers are telling you is not always easy. It is entirely feasible to come up with varying results based on underlying assumptions. Once you have put all of the work into understanding who is participating in your organization’s loyalty program and how that is affecting their decisions as a customer, be sure to find a partner to help you fully realize the potential of the program you have in place and to keep your customers coming back for more.