Building the Data Science Dream Team

Before joining the SSENSE data science team, I used to consult with organizations to help them capitalize on their data. I’ve witnessed analysts, data scientists, managers, and executives vent their frustration.

One of the most common pain points for data teams is not being able to communicate the valuable insights they’re sitting on. Oftentimes, it feels as though executives expect them to perform magic and to answer all their questions, while misunderstanding or oversimplifying their work.

These same executives also often complain about how data science operations don’t provide the results they had hoped for, and lament about the lack of ROI. The results aren’t communicated in a language they understand, so they don’t see tangible outcomes.

Here’s my take on this challenge and how SSENSE is working to get the most out of its data science operation.

The more things change, the more they stay the same

Even before fancy software and multi-core computers, business and technology have always had a hard time remaining in sync. More than a hundred years ago, Willard Brinton explained the last-mile problem in his book Graphic Methods for Presenting Facts: “Time after time it happens that some ignorant or presumptuous member of a committee or a board of directors will upset the carefully thought-out plan of a man who knows the facts, simply because the man with the facts cannot present his facts readily enough to overcome the opposition… As the cathedral is to its foundation so is an effective presentation of facts to the data.”

As data science and data visualization tools become increasingly intertwined, the notion that data experts also need to be experts at communication is quickly gaining popularity. In reality, data experts often don’t want to, or aren’t able to properly communicate with business stakeholders. Communication, visualization, and analysis require very different skill sets.

The default output of data wrangling tools can’t compete with the output of well conceived, fully designed data visualization software; hence we are often left with a gap in the quality of data manipulation and visualization.

I’ve heard many data scientists say that fancy visualizations “dumbs down” their work, getting executives to draw conclusions that don’t take into consideration the nuances and uncertainties in their scientific analysis. In the rush to stay ahead of the curve, most companies getting into data science hire the most technically proficient people they can find, sometimes ignoring their ability or desire — or lack thereof — to communicate with their collaborators.

Common Pain Points

I’ve learned in my work that most leaders recognize the value that data science can create, but only a few are satisfied with how it’s being created. In many companies I’ve worked with, data scientists complain that bosses don’t understand what they do, and therefore are underutilized.

Discontent from both sides usually falls in one of the following three scenarios:

Delayed collaboration between subject matter experts (SMEs) and data scientists (DS)

A manager wants to implement a recommender system in a client-facing app. Say a product discovery feature. There is plenty of data to work with. A data scientist takes over the project, the feedback loop is 2 weeks within the data science team. The model produces good results based on the available data. Just as it’s time to present to other parties: SMEs don’t agree with the results.

Back to the drawing board, but now with SMEs to help evaluate the measures. The SMEs’ deep domain-specific knowledge allows them to understand nuances about different concepts and types of metrics, and how these can be used to solve a specific problem.

A general misunderstanding of what data science is

A business stakeholder wants to start a new initiative but can’t back up their hypothesis with facts or numbers. They ask the data science team to produce the analysis and charts for the presentation. Data scientists know the hypothesis is shaky, and offer ways to tackle the problem. The business stakeholder wants only charts and speaking notes. One of two things will happen: the meeting will go south when someone challenges the data analysis and they can’t provide answers, or the project will go through and then fail to meet expectations because the analysis was more of a sales pitch than an analysis.

Lack of communication skills within the data science team

A data scientist develops a suite of insights and presents them to decision makers in great detail. They believe that their analysis is objective and rock-solid. Their charts are “click and viz” with some text added to the slides — many data scientists think that design isn’t something worth spending time on. The language they use in their presentation is foreign to their listeners, who become confused and frustrated. Their analysis is dead on, but their proposal falls on deaf ears.

There are multiple ways to address these three scenarios. The sections below outline some solutions to tackle them through talent acquisition and allocation, cross-departmental collaboration, and frictionless operations.

Managing for skills

An effective way to manage a data science operation based on teamwork is to cultivate a portfolio of skills that can tackle a wide range of project types, from standard data analysis to the most sophisticated big data efforts that use obscure machine learning algorithms.

Focus on talents, not roles

If we want to stay away from the “one-man army” data science way of thinking, it would seem natural to assign roles based on skills: data wrangler, analyst, modeller, designer, communicator, etc.

Well, not quite… rather than assign roles, it is more sustainable to define the talents you need to succeed.

A talent is not a person and vice versa; more than one person can possess the same skill. One person may have several talents; three people may possess five talents.

It’s a subtle distinction but an important one for keeping teams versatile enough to configure and reconfigure during various stages of a project.

Every company’s talent requirements vary, here are the five primary skill axes for data science at SSENSE:

Business and management

A good manager will have strong interpersonal and planning skills, helping to bridge cultural gaps by bringing heterogeneous talents and personalities together at meetings, and by getting everyone to speak the same language. You need someone that understands the business objectives and can make sure the project is steered in the right direction.

Tech and data wrangling

Technically-inclined people will look for opportunities to streamline operations — for example, by building repeatable processes for multiple projects and templates for solid, predictable visual output that will jump-start the information-design process. Skills that compose this talent include building systems; finding, cleaning, and structuring data; and creating and maintaining algorithms.

Statistics and data analysis

The ability to set hypotheses and test them, find meaning in data, and applying that to a specific business context is crucial — and, surprisingly, not as well represented in many data science operations as one might think. Some organizations are heavy on wranglers and rely on them to do the analysis as well. But good data analysis is separate from coding and math. Often this talent emerges not from computer science but from the natural sciences and liberal arts. Critical thinking, context setting, and other aspects of learning in the humanities also happen to be core skills for analysis, data, or otherwise.

Domain expertise

It’s time to retire the trope that data science teams are isolated in basements, doing arcane work and who surface only when the business needs something from them. Data science shouldn’t be thought of as a service unit; it should have management talent on the team. People with business-acumen and strategic knowledge can inform project design and data analysis, and keep the team focused on business outcomes, not just on building the best statistical models.

Design and storytelling

“Storytelling with data” is a fatigued statement now. Narrative is an extremely powerful human contrivance, and one that is severely underutilized in data science. The ability to present data insights as a story will, more than anything else, help close the communication gap between algorithms and executives.

This needs to be coupled with good design, not just choosing colours and fonts or coming up with an aesthetic for charts, but understanding how to create and edit visuals to focus an audience and distill ideas. Information-design talent — which emphasizes understanding and manipulating data visualization — is ideal for a data science team.

Expose team members to talents they don’t have

Learning about your neighbours’ skills is the best way to avoid culture clashes. Data scientists value statistical rigour, objectivity, and scientific integrity; building fancy charts and communication pieces impedes their ability to reach their objectives. Design talent, in contrast, often has no exposure to statistics or algorithms. Instead, its focus is on aesthetic refinement, simplicity, clarity, and narrative impact. Neither need to become experts in their counterparts’ field — they just need to learn enough to appreciate the other.

At SSENSE, we tackle this challenge at a large scale by holding workshops on specific subjects where employees can learn about each other’s trade. Recent subjects have included data engineering, Chinese Marketing, Buying & Planning, amongst others. This initiative is driven by our Human Resources department and is called SSENSE University.

Talent Mapping

Laying down a map of the talent you have access to will allow you to better plan for projects and configure teams.

Map those talents to your team members.

And finally assess, the pool you have for each talent.

Putting it to use

With a talent map in hand, managers can now allocate talent to projects and plan their next steps efficiently.

Meetings should include a mix of talents. Stand-up meetings to update the tech team on the progress of a recommender system can still include a marketing team member, who will most likely make the Demo Day deck. SMEs could bring their statistical modelling talent to strategy meetings to give them a higher view of the challenges the company is facing.

If you’re lucky enough to have people with data, communication, and design talents in your organization, prompt them to mentor others. Encourage people who show interest in developing new skills that are required in your team, even if those skills are far afield from their core talents.

While consulting, I’ve heard from multiple data scientists and data analysts that they would love to develop their design and storytelling talents. Frequently, however, they usually don’t have time to commit to it, or lack support from their organization. Others would love for their team to have access to those talents, but their management structure focuses on technical outcomes, not business ones.

SSENSE encourages out of scope work where a member of a team can allocate a few hours a week on a project that is completely unrelated to their work. This is a good way to initiate both resource and knowledge sharing.

This exposure is meant to foster empathy among coworkers with different skill sets. Empathy in turn creates trust, which is necessary for effective collaboration.

Build a roster

With the talents you need identified, unlearn the idea that these are roles you should hire people to fill. Align with your recruiting counterparts to make sure these talents are available in your team, or in other teams. Naturally, some skill sets will appear in pairs: Design and storytelling, data wrangling and data analysis, etc…

Thinking of talents as separate from people will help companies address the last-mile problem, because it will free them from trying to find the person who can do both data science and communicate it. Nabbing a data engineer with stellar data wrangling skills will free data scientists to focus on their strengths. A manager who also has a good eye for design, for example, might be very useful.

Here, we try not to focus solely on how well someone can do the technical part of data science and appreciate the communication part of it. In the hiring process for SSENSE, we’ll run a full set of cross functional interviews to see how a candidate can collaborate with different departments.

Don’t hesitate to throw in some communication questions in your technical test. You can also avoid the echo chamber of only data scientists interviewing data scientists, and bring in a nontechnical person. Ask the candidate to explain their models to this person. It’ll both challenge the candidate’s theoretical depth and their ability to articulate complex concepts.

Structure projects around talents

It is time to put your portfolio of talent to use. Combine talents and interests as needed to optimize project progression. Experience in agile methodologies will help in planning the configuration and reconfiguration of talents.

Seamless operations

Define goals before stakeholders

9 out of 10 times, only a fraction of the people whose talents you will need for a data science project reports to the data science team manager. Design talent most likely reports to marketing, and subject-matter experts may be on the finance team. The deciding parties will most often be responsible for business goals and have a higher view on the business. These people should create shared goals and incentives for the team. This will allow you to avoid the responsibility-without-authority trap, in which the team is left to deal with several stakeholders who may not all be aligned.

Mainstream communication

You should try to have all team members work on the same virtual space for communication and collaboration. It is already complicated enough that we don’t speak the same language. The least you could do is avoid a situation where design and storytelling talent use a Slack channel and DropBox, while the tech team uses GitHub, and the business experts are on Google Suite.

I would also suggest trying “pair analysis” techniques, derived from “pair programming”, whereby team members literally sit next to each other and work on one screen instead of sending feedback back and forth.

Support and carry

Let the nature of the project and the phase it is in dictate who should be leading and carrying.

A deep dive analysis aimed at recommending strategy adjustment will likely involve multiple data sets and heterogeneous infrastructure. In such a situation, data wrangling and statistical skills should take the lead. Management, design, and communication skills should be there to support and super-charge the effort. On the other hand, to prepare a report for this recommendation, storytelling and design, and management should lead with support from data talent.

Build frameworks and templates

You should foster a culture of templating. Have your team leverage each other’s skills by developing reusable functions and classes for analysis, visualization, and modelling. Such frameworks are invaluable for getting a team operating effectively.

Conclusion

Until companies can break from the notion of “one-man armies”, they’ll simultaneously face difficulties in hiring data scientists, and incite frustration within data science departments and executive leadership.

We need to rethink how data science teams are managed and assembled, and find ways to properly leverage all key players, all the way from the first brief to the final deployment of models. By focusing on talents instead of roles, we allow ourselves to move faster both in hiring stellar talent and allocating resources to key challenges.

In a fashion similar to machinery in the industrial age, data is redefining how we do business in the 21st century. In today’s world, every industry is on track to becoming data and technology driven, and building healthy data foundations will give you a head start in this corporate marathon.

Editorial reviews by Deanna Chow, Liela Touré & Prateek Sanyal

Want to work with us? Click here to see all open positions at SSENSE!