Why You Need a Better Data Science Strategy

August 5, 2019 Leah Hoffmann

data science strategy

Ted Bissell is an independent consultant and executive who helps companies bring together data analytics and machine learning to improve digital offerings in the financial services and payments space—and other industries that depend on them. In our latest Expert Q&A, he talks about using data science to deliver business value, overcoming analytics challenges like legacy systems and compliance issues, and why the future of these platforms may be simpler than it seems.

Business functions from strategy to supply chain leverage modern data analytics; many companies have even set up dedicated analytics groups. Yet while most execs see promise in these techniques, only 18% believe their organizations can gather and use data insights effectively. Why do you think that is?

In the best of circumstances, companies have rolled out uncoordinated initiatives in various business units, and they are struggling to pull up and chart a path for driving them forward or expanding or consolidating them.

The worst cases are what I call analytics train wrecks. Someone has managed to cough up the budget to implement a specific analytics platform or environment. They might have hired an expensive team, but the business isn’t seeing results, and now they’re kind of stuck. Highly trained data scientists are doing pretty basic analyses, which demotivates them. Folks on the business side may not have had a lot of exposure to the technology, so they’re unable to foster a dialogue around the organization’s real needs.

How can companies go about trying to fix these issues?

I think the first question is “when.” What’s the best time for organizations to start thinking about and redefining their high-level data strategies? There are several potential triggers. It could be that someone sees the budget spent on a lot of third-party data sources or realizes there’s an imminent decision about whether to upgrade a specific platform or pull the plug on a legacy database. An investment might be tied to what’s happening with data. Or there might be a compliance issue, whether it’s the realization that something’s vulnerable, a specific breach or event, or regulatory obligations such as GDPR in Europe or state-level privacy laws in the US.

At moments like those, it can be a good strategic decision to say, “We know we could be doing something better with our data, but we’re not sure what it is.” And to answer that, you’ve got to get a handle on what’s happening day to day.

And then?

Then it’s time to start framing the potential business impact. What should the company get to be able to justify the investment that’s required to upgrade a platform or tackle the broader question of redefining data strategies? How will deploying advanced analytics help us gain a competitive edge? Even if it’s something we’ve got to do because of privacy, well, what’s the cost of compliance worth in terms of impact on earnings per share or the overall IT or compliance budget?

Part of making big questions actionable is breaking down what you are really going after and defining what getting it right will deliver.

How does that tie in with some of the communication issues you mentioned earlier, where executives and data scientists struggle to align around what the business really needs?

It’s not just about communication. Data scientists are trained on the latest, advanced models, applications, and approaches to data types. But the moment you reach an average big corporation, you run into all kinds of compliance and regulatory issues—and the oldest of legacy systems.

Most US banks, for example, have held customer and transaction data for years. Yet often, that information is trapped in one of a handful of core platforms never envisioned for our cloud-connected world. Getting valuable legacy data out in a safe and legally compliant way is a huge challenge. You might have data scientists who can do a fantastic analysis, but they won’t get off the ground if they don’t have access to a coherent dataset. That’s not a communication issue or even a coding problem.

It does speak to the importance of understanding the investment that’s required to retrieve and format data, not just analyze it.

Getting data right isn’t trivial. It’s expensive. Yet obviously, if you want to do anything interesting, you need to start with good data.

Unfortunately, most data scientists have been trained on techniques that rely on clean, large, and easily accessible datasets. They simply haven’t been tasked with dealing with sloppy data or extracting it from a legacy system. To a degree, I don’t think anyone would ever expect to find such antiquated elements still in use. Yet many healthcare companies and banks rely on databases that were built on technology developed by IBM in the 1960s. Often, these systems rely on batch processes that run overnight. So if a bank has just recorded my payment transaction, it’s not likely that anyone on the analytics side will be able to access that information quickly—at least not quickly enough to create an algorithm that reacts to my behavior in real-time, for example.

No one in academia would be willing to recreate an environment like that simply for the purposes of teaching people how to take it apart and fix it.

And yet the opportunities are so powerful.

They are. Let’s stick with banking for a moment. The bank website is typically laid out and run by the marketing department. But the moment a user logs in, IT is running the backend. So they can analyze the traffic to their website, and they can analyze how customers use their services. But they can’t leverage real-time machine learning that could dynamically alter the online experience for each customer type.

My son applied to college last year, for instance, and I paid for his application fees. I was struck by the fact that my financial institution did not connect those transactions with a timely offer for a parent student loan or related services.

What lessons do you take from companies that have managed to get it right with respect to the resourcing side of data analytics?

One of the BTG clients I worked with—an insurance company—had a very centralized analytics function, and then collaborated with different business units on a project basis. As a result, they were able to deliver a very consistent quality across their work. Perhaps most interestingly, rather than reporting into IT, analytics exists as a distinct business unit at this company. Other insurers have gone on to create sub-brands for their analytics businesses.

That’s fairly unusual.

Insurance has bought into analytics for a long time. More often, what happens is that someone sources a new platform—a Tableau instance or a Hadoop infrastructure, say—with some analysis tools, and not many people know how to run them. IT, responsible for the platform implementation, goes out and hires the first wave of programmers and data scientists, but often acquires that expertise without a realistic sense of what data can and can’t be put to immediate use, or how the benefits to the company will be measured and reported. And very quickly, you’ve lost contact with the boardroom decision that led to the acquisition in the first place. Day to day, people are not saying, how is our analytics delivering value and what goals do we want to reach?

What about the future?

In my view, maybe five years out, all these data science capabilities will be integrated into most business platforms and used by all business units. For example, Salesforce, which just acquired Tableau, will have some sort of ready-to-go, basic analysis capability for any user. You won’t need a specialized programmer or data scientist to run it. Instead, you’ll rely on some easy-to-use weak AI capabilities that are embedded into whatever package that you’re implementing. Meanwhile, stronger AI-based automated decision-making will be hungry for real-time data at faster-than-human speeds. This will force companies to connect new data sources and invest in Internet of Things (IoT) technologies.

Obviously, to be competitive, you’ll need to stay ahead of that. But the mainstream is very likely to skip over some of the current complexities that everyone’s trying to understand.

Need Data Science Support?

Our independent data science experts can help you do more with your data.

START A PROJECT

 

About the Author

Leah Hoffmann

Leah Hoffmann is BTG's Content Strategist. A former journalist, Leah worked for Forbes.com and The Economist before joining BTG. She is passionate about clear thinking, sharp writing, and strong points of view.

More Content by Leah Hoffmann
Previous Article
Overcoming Data Science Challenges: Expert Panel
Overcoming Data Science Challenges: Expert Panel

Top AI and talent management experts discuss how to find, develop, and retain data science talent—and overc...

Next Article
Why You Need a Better Data Science Strategy
Why You Need a Better Data Science Strategy

Most executives see promise in data science. Yet few have been able to develop a data science strategy that...