Don't overlook the value of having a data science team play an operational role at your company. Here are three steps to follow when assigning analysts to these roles.
When most people imagine a data science team, they envision a think tank centrally located in corporate headquarters that is assiduously working to propel the company's next big strategy. Some even think of data scientists in more of a research and development function, doing deep data analysis in hopes of a breakthrough innovation.
While these placements in the organizational structure are viable and appropriate, there's another, less-obvious place where your data scientists can really make a difference -- at the heart of your company's operations in what I call operational data science.
Houston, we have a problem
It may seem a bit odd to put your data scientists on the front line; however, in some circumstances it makes a lot of sense.
I recently did some work for a large financial institution to help increase their cybersecurity defenses. When we considered options for organizational structure, it made sense to put a small team of data scientists in an operational role when a possible breach was detected. As the technical team was deployed to defend the incoming attack and protect sensitive systems, the data science team was deployed for real-time informational support.
Reminiscent of the brain trust that's assembled in mission control whenever NASA launches a rocket into space, it's worth considering a real-time, operational data support function that your data scientists would man when necessary.
3 steps to using data scientists in ops roles
1: Define all possible use cases
If the opportunity does present itself for operational data science, the first thing you must do is thoroughly define all the use cases under which your operational data support is required. This starts with an inventory of all the operational business functions and continues with a thoughtful analysis of whether advanced analytical support would be valuable.
Cybersecurity is just one example. Consider a delivery company like FedEx that is responsible for delivering packages on time. In unusual circumstances like a tornado, it may be helpful to have a team of data scientists analyzing routes for optimal delivery times. Once opportunities for real-time data support are identified, clearly articulate use cases so the team can simulate and prepare accordingly.
2: Think about data system support
The next thing to consider is your data support system. There's an architectural construct in data warehousing called an Operational Data Store (ODS) that seems to fit the bill in most circumstances.
If you don't already have an ODS, it's time for your architects to start designing one. An ODS typically sits between the transactional system and the data warehouse. It acts as an intermediary to provide operational analysis from the transactional systems, so they don't carry the burden of simultaneously processing transactions and crunching through analytics.
The class of the ODS determines how frequently it's updated from the transactional systems. Class 1 systems are near real-time (within seconds); Class 3 systems are daily; and Class 2 systems are somewhere in between. Class 4 ODSs are refreshed on a less-than-daily basis -- typically weekly.
For most intents and purposes, your data science team will need a Class 1 or Class 2 system. Even with the recent advances in technology, Class 1 systems are very expensive to build and maintain; they typically use messaging technology instead of classic Extract, Transform, and Load (ETL) technology. So, keep this in mind when surveying the overall value of your real-time data support capability.
3: Be selective in your assignments
Finally, you must be very selective about the actual data scientists that you assign to these roles. The reason why you don't automatically relate data scientists to high-pressure, operational roles is because you assume analytic-minded people don't function well in these roles. And you are correct. The stereotypical analytic does not like pressure. They need time to think and analyze, and they don't want someone hovering over them for answers.
I was recently working with an analytic to move through some analysis, and we were running out of time. I gently nudged her by saying, "We only have 10 minutes left," and she quickly retorted, "Don't do that! Countdowns freak me out!" This is very typical.
So you have to search for the atypical data scientist. You need data scientists that are comfortable under pressure and think and act fast. Usually, they've had some other passion or experience outside of their data science world where handling pressure is normal; soldiers, first responders, and athletes are a good place to start. You can also look through their job history to see if they've previously been in an operational and/or high-pressure role.
Just don't assume you can place any data scientist in an operational role, or you'll likely regret that decision.
Although data science certainly has its place in strategic analysis and innovation, there's also a case to be made for using your data scientists in real-time, mission critical scenarios. Work with your General Manager or Vice President of Operations to uncover opportunities for analytic support, and talk to your data scientists about how they might be able to help. If you see a match, define your use cases, build an ODS, and select top-notch data scientists that can handle the pressure.
You may not be launching rockets into space, but there's nothing wrong with a little on-the-fly data science when it makes sense.
- 5 ingredients for adding a sous data scientist to your team (TechRepublic)
- Two programmers are better than one for your data science team (TechRepublic)
- 10 tips for chief analytics officers (ZDNet)