Building a Team to Build Data Products

The ability to create Data products is mainstreaming from the edgy early adopter phase, to becoming more feasible for companies to consider applying to help improve their customer experience and insights as well as driving efficiencies through internal operations. The needed technology is there, it is accessible and there is precedent to help companies find ideas for applying the technology. The data you need still can be challenging to find, organize, label, and clean your data sources, but there are many sources of supervised data to help get started. These are important discussions. We’ll talk about them at some point in the future. However, today I want to spend a little time discussing the skill requirements you will need in a team, and how to determine what you should prioritize to get started.

It is hard to find and afford talented and experienced data experts. Many companies will want to start with a key resource to anchor a team. Others will be curious if their existing resources can be retrained to become productive data experts. The basics of team building is slightly less complicated than “stochastic gradient descent”, but this is new to most companies, and a little bit of a blueprint would be helpful.

I will caveat, what you are doing matters. If you were to call me and talk about building out a new deep network, the specifics will matter. You may be starting from scratch, or you may be able to leverage an existing trained model through a technique called “transfer learning”. What the deep network needs to learn to identify matters. You may need to do something basic like detecting puppies in photos, or something more difficult like interpreting stress in someone’s voice. These specifics will matter to tune your actual plan and the associated team, but let me share with you some generic rules for building a data team.

THE ROSTER

A data team is generally made up of three type of resources:

• Data Engineer - The data engineer is concerned with the processes and architectures for acquiring, housing, and cleaning data. This work is most closely related to traditional IT roles that have worked with data in the past. It is possible to potentially start with someone on your staff.
• Data Scientist - The data scientists are educated on the algorithms, math and procedures of applying the math to the data. Their job is to find the best algorithm to fit the data. They should, and usually are, intelligent, curious, and good at problem solving at a scientific level. They may also be challenged to fully conceptualize the rules of business that they are applying the problem to. Unless you are hiring new college graduates from a rigorous technical college, it is not likely that you have existing resources who will naturally convert to becoming a data scientist. Albeit, I graduated well before data scientists existed, so who knows.
• Data Translator - The data translator may easily come from your business. The role is best served by business analyst types who have a deep technical skill set, and are a good balance of curious and cautious. Someone who is a trained problem-solver, worked in product development, or may someone from the agile team could all be good candidates. This role needs to understand everything at a working conceptual level so that they can bring the business need or opportunity, through the technical process of developing a deep network, ensuring it solves the problem, considers accuracy, inherent bias, etc. 

THE ART OF DATA VALUE

Now that we know the players, who is most important to you and your potential data products? Data value is the generation of product or insight from the effort of data analysis. There are three key attributes of data value that usually manifest in the relationship between the different roles. Note sometimes, your first identified team member may be someone who has skills that span two roles. In that case the shared relationship attribute would be a desired personal attribute. In general, the three relational attributes between the data specialists are:

• Creativity - Is the product of art and science; of finding a complementary fit to the business problem, or potential evolving new IP from an uncovered market opportunity. Creativity can be found throughout, but should be demonstrated in the relationship between your translator and your scientist. If creativity is critical to your venture, then these two roles will be very important to your work and need to have priority. 
• Assurance - Is the need to carry forth the principles and risk profile of the business. It is the governance policies that make sure the data product output is properly aligned with the corporate mission. Assurance is most represented in the relationship between how your translator and your engineer represent needs, specs, architectures, etc. to complement the business intent and culture. If you are an established multinational, and you want to start a maturation process to incorporate data products into a product portfolio, assurance will likely be a priority for your initial steps. 
• Rigor - Is the adherence to the best practices of science and data techniques to ensure your data operation is ship-shape. Are your models tuned, maintained? Is your data valid? When do you retrain? Are we asking legitimate questions of the networks? Not all data products are deep learning based. This also applies to machine learning techniques being used to embed intelligence into products. As you continue to build out your team and start to manage a farm of deep learning networks, rigor will grow in importance to your operation.

AN EXAMPLE

I thought it may be helpful to discuss a personal example. I was the first person on a team for a startup to build data products. I am a senior resource, and thus have a little of each of the skill sets at my command.

Let me breakdown my rough assessment of my skills:

We could spend more time assessing me, but let us assume I am right. The question is, was I a good fit for the group building data products? The group needed technical help, and were considering how to develop more insights from their growing dataset. More specifically the group wants to use long form stories to generate attitudinal scores that could be used to assess how a community of people holistically think and feel. This was not something we could buy off the shelf, we were going to have to look across the state of art for the industry, consider technique and benchmarks that could be used to define our foundational grounding. In many ways this was a research journey and needed some level of rigor to assure our efforts and results were justified, and how we could validate that justification in our products. In the process we created a whole new product line that has been delighting customers, and continues to grow and evolve.

It was my job to build the team with “me” as the starting point, so maybe I am too close to it. With that said, I would say I was a pretty good choice for these reasons. My strengths are that I have spent my career consulting, facilitating/leading change and running various operations. I am a pretty good translator in general. I have a scientific mindset. I spent a great deal of time reading academic journal publications on data science, anthropology, psychology and from other fields. I was a STEM kid if you will, and I felt compelled to find a scientific grounding for our work, and though we were a startup, not a college lab, we respected these principles. As a startup creating data products leveraging NLP and other machine and deep learning techniques, we needed a good balance of creativity and rigor which I supported naturally as the initial person doing the work. Assurance was less important, and for us becomes a growing concern as that business scales.

I hope this article helps you with your thinking. If you think there is an opportunity to work data into process, there is. I suggest getting started. I also suggest you develop a team with a nice balance of creativity, rigor and assurance to product offerings that complement your brand.