Thu Sep 26 2019

The Right Road to a Data-Driven Future

The development of data capability is a journey. As companies’ capabilities grow they go through phases - first beginning to collect data, then recognising the value of data, and crucially beginning to center data in their business model (we call this the credibility phase) before they take off towards a data driven future. We see that many companies struggle to pass the ‘credibility’ phase, which isn’t strange in this highly complex but promising field.

A recap of Xomnia’s Robin van den Brink’s talk Big Data Expo talk (19/9/19)
Robin is an Analytics Translator for our X-Force team.

Some of the barriers we see most frequently are:

  • Teams with only data scientists or only data engineers.
  • Models run in a Jupyter Notebook that have to be shared manually.
  • A view of Machine Learning and/or Artificial Intelligence as a goal rather than a tool.
  • Proof of concepts that end up spiraling into months-long projects with a lot of code

These issues prevent companies from truly reaching the full potential that data can offer, but they can be overcome by focusing on three core elements: finding the right problem, assembling the right team and applying the right approach. FIND THE RIGHT PROBLEM AI/ML alone aren’t going to revolutionise your business – these tools are only as smart as the problem placed before them. The shift to a data-centric organization shouldn’t begin with the algorithm, the best problems begin with the end user. Good problems can come from different areas; from top down strategic priorities or bottom-up explorations following the why-funnel to get to the heart of operational inefficiencies. It could also be that your direct competitor launches a data driven disruption that you have to match. For any problem you identify, prioritize fully understanding it first, challenge yourself to asses the added business value of working on it, and only then should you begin developing solutions.

Build a multidisciplinary team

Finding a single candidate who can interpret business needs, find best-fit mathematical solutions, and bring a product to scale is about as likely as encountering a unicorn (read: near-impossible). Expecting that the technical experts in these fields will also have the business savvy to find a problem that will provide maximum value (and these people are expensive!) is unreasonable. When putting together a team, we’ve found that it’s best to compartmentalize roles. By having people with specific skills in your team, we see in our own X-Force teams that you can deliver better products faster.  By breaking your teams into multiple roles, they can build strong relationships with other organizations/functions within your company. At Xomnia our division of labor looks like this:

Data scientist: laser-focused on translating a perceived problem into a mathematical model. The more time they have to hone a model to align with reality, the higher the value you can expect. They’re plugged into the organization’s data team.

Data engineer – entirely infrastructure focused – count on them to write proper, efficient, scalable code. They partner with in-house engineers to create the data pipelines the scientist will use, and make sure the solution runs on the organisations infrastructure.

Analytics translator – interprets the businesses needs and translates into a problem to be solved. They keep clients up to date on project timelines, continuously involve the right stakeholders, and ensure that the solution is delivering maximum value. The analytics translator frees data scientists and engineers to excel at what they do best.

Iterate and ship as often as possible

Once you’ve found a problem with a strong business case, don’t start building something complex or sophisticated. Create a baseline model, start developing simple models to test your hypothesis, and be prepared to abandon what you thought was a great idea. Good Data Scientists/Engineers are rare and expensive, and you want to constantly challenge them to get the most out of the capability. Given the uncertainty involved in any data science project (both in terms of technologies/algorithms and requirements), we recommend using the Agile Scrum methodology for work – it provides a way to quickly prototype and rollout solutions.

Walk the walk

If you’re able to pull these pieces together, you’ll have a team with the skills and adaptability to zero in on strong business cases for your data. Remember, the road to a data-centered company is winding, assemble the right pieces and trust the process of continual experimentation and improvement. If you continue to encounter issues accelerating your data ambitions.