8 professionals you need to build your analytics dream team
By Laura Squier, SAS
Avengers, assemble! If we learned anything from the summer 2012 blockbuster Avengers, it’s that having a team with a variety of skills can accomplish almost anything. I’ve been fortunate enough to work with a wide range of talented professionals, and from these experiences, I’ve cultivated my top-eight list of people I’d draft to my analytics dream team. Just remember that outside of the Marvel universe, it’s very unlikely that one person can be a superhero with all these skills. Instead, realize that your team members might have one focused discipline, overlapping areas of expertise or could bring something completely new to the table.
As you read on, you’ll note the first four are all data-related. After all, analytics is dependent on having high-quality and comprehensive data. Without these team members, the project could get off to a rocky start.
Subject matter expert (SME)
I’m sure everyone wishes they had a panel of experts available when faced with difficult situations like buying a car or filing taxes. When it comes to an analytics project, having a subject matter expert isn’t a luxury – it’s a necessity. Without it, your project will likely be poorly framed and out of touch with the organization’s objectives. Understanding the business, relevant processes and relationships between variables are a few of the key skills that this expert should be able to provide. Without one, you run the risk of having a model that isn’t appropriate for the problem your team is trying to solve, or the end-result analysis may be inaccurate. Importantly, the SME should be consulted regularly throughout the process so that any costly errors can be avoided or corrected.
Be wary of starting a project with poor guidance. Years ago, I worked on a commercial project where we were trying to reduce the number of SKUs sold, without impacting the overall revenue. What we failed to learn from the so-called SME was that the SKUs offered changed quarterly. The recommendations that we analytically derived were null and void because there was already a process in place for removing the SKUs – and by the time we finished our analysis, the organization had an entirely different set of SKUs to optimize! Had we involved a true SME throughout the project’s lifespan, he would have picked up the fact that our project didn’t align with current processes.
Even if you never trekked through the woods as a child and learned the Scouts’ motto of “be prepared,” it’s still sage advice, especially for analytics projects! A data preparer is able to pull data together from multiple sources and do the appropriate aggregations, calculations and joins for analysis. Often the data preparer works hand in hand with the analytics professional (another member of the “Elite Eight”), making incremental changes to the data to ready it for analysis.
It’s also the data preparer’s job to not only to prepare the data for analysis, but also ensure that the data preparation steps that are completed are well-documented and deployable. For example, if I am developing a real-time scoring solution for a client, I can only build the model off the data that will be available during scoring. If there is third-party data that I want to include, or other sources of data, I must either alter my deployment design or remove those data elements from the modeling process. Alternatively, they may make a business case to add additional variables in deployment.
Organization is everything. When you’re about to launch a project, don’t overlook selecting a data steward. A data steward is responsible for knowing what data is stored in each repository and how each variable is precisely defined. They should also ensure that the data elements being used are accurate, correct and fully documented. Having a great data steward on the team is essential for having a strong foundation for your project.
I hadn’t really understood the value of data stewardship until I worked with a large US bank a few years back. I was coaching a customer loyalty project, where we were trying to increase a customer’s overall engagement. We spent considerable effort trying to decide how to define a customer – is it a household, an account or individual customer Social Security number? There were problems with each. An account may be shared by multiple people, related or unrelated. A household may contain unrelated people sharing a home. A customer might hold multiple accounts, some shared, others not, and linked or unlinked. Having a strong data steward allows you to make sure that the data elements selected are what you expected and choose the best path forward given the data at your disposal.
Talk is cheap. All too often, organizations develop analytic “projects” that never get beyond a report in a binder. Results from analytics need to be pushed out into operational systems so that they can be USED. This may be 5 percent of the effort or 80 percent of the effort, depending on how the models are used. While operationalizing analytics may seem like a big effort, without it, the value of analytics is minimal. When solutions are developed that guide business people to act, the ROI is very high. These factors make an IT specialist or implementer an essential component of an effective team.
Take for example, the easiest form of model deployment – develop a model to push out scoring code for a direct-mail offer. Give code to a database manager, or third-party marketing service provider. They run it and pass a score back to the marketing manager, who then decides how deep in the list to mail. Here, deployment may be a 5 percent effort where we just need to make sure that the variables that were used for modeling are also available for scoring.
On the other hand, an organization that is doing real-time fraud detection may need to automate the consolidation of multiple data feeds, score the data with multiple rules and models, and push results out into an operational case management system along with relevant graphs and visualizations. This type of solution would likely take 80 percent of the effort, if implemented accurately and designed to be extensible and configurable in the long term.
If you think that implementation is only one small part of the your project’s success, consider this: In an April 2012 study, Nucleus Research found that the ROI for analytics climbs from 188 percent in the initial stage to 1,209 percent in the fully implemented/deployed predictive phase. Having an IT specialist or implementer is essential for seeing a project through to completion and unlocking the project’s true value.