Once a year, I teach a course to aspiring data scientists. At the beginning of the first class, I ask my students what they hope to learn. Often, their answers are “analyze data” or “build good models.” Compare those responses to the workshops I run with practicing data scientists who speak in different terms when discussing what they do. They call it “solving problems” — a step in the right direction — but even that is too narrow.
As reliance on data and analytics continues to expand across industries from agriculture to manufacturing, health care to financial services, it stands to reason that the next generation of data leaders will have far-reaching roles that impact strategy, decision-making, operations, and countless other functions. To help prepare this new talent, I have developed a framework composed of four key areas of skills and capabilities that will help current and future data scientists hone their abilities to add maximum value to a business. This is done by ensuring that data science work is seen as important and indispensable by their business-function counterparts.
Using this framework, and with greater understanding of what each area of business entails, today’s data scientists and those entering this field can see how their knowledge and experiences stack up — and where they need more development.
1. Problem Spotting: Seeing the real issue
As they delve into analytics across the business, data leaders have a front row seat to nearly every operation and function. This provides them with a unique vantage point for both solving problems and identifying new ones. Here’s a real-world example. The head of guest relations for a mid-range hotel chain was getting flak from upper management for low ratings on their check-in process. Surveys revealed that guests thought check-in was poorly managed, took too long, and didn’t provide the results they wanted — specifically, a seamless and pleasurable experience. Management also found that people who rated check-in poorly had a lower rate of returning to the hotel.
The guest relations department reached out to the data analytics team to figure out the root of the check-in problem. Even when they looked at customer demographics, the kinds of rooms they wanted, and whether they checked in at the front desk, at a kiosk, or on their phone — plus time of day, time of year, and whether customers were in the loyalty program — the data team couldn’t put their finger on the underlying cause.
Then an employee suggested they look at customer surveys that had been collected on a rolling basis. Some natural language text analytics teased out some themes — namely, the hotel infrastructure was not optimal. Guests would have problems with the Wi-Fi, room keys occasionally wouldn’t work, furniture was broken, or rooms weren’t clean when they arrived. These problems were not directly related to check-in, but guests attributed them to the check-in process because that’s what they remembered. Bottom line — the problem was with how the hotel was being managed, not the check-in process.
The Takeaway: Solving the problem that is in front of you can mean missing out on opportunities to help the business improve in other ways. Those who work with data often have access to deep, unique insights into numerous aspects of the business. To become adept at problem-spotting, data leaders need to embrace that big-picture view and gain deeper insights, with greater transparency around what matters most to business leaders. In this way, data leaders can add value by identifying problems that otherwise escape notice.
2. Problem Scoping: Gaining clarity and specificity
Once a problem has been spotted, the next step is determining its scope — that is, gaining clarity into the nature of the problem and how analytics can help solve it. This is especially important if a business leader has approached the data team with a vague concern or challenge.
In my classes and my workshops, we practice scoping with an exercise. I assume the role of a product or strategy or marketing leader with a well-defined problem in my head. For instance, perhaps I manage customers, and want to be able to identify which customers are at risk of giving low net promoter score (NPS) ratings so that we can intervene and improve their experience. Any reasonable data scientist would know how to select the right data and data science techniques to solve this problem. But business leaders rarely speak like this. And so I scope the problem using exaggerated jargon and overly general terms. It goes something like this: “We’re struggling to hit our customer sat targets — we need to zero in on our go-to-market strategies. It could be a pipeline issue, but we just don’t have alignment. I think we’re playing in the right sandboxes, now we just need to know the who and the why. Sound good?”
A student, in the role of the data scientist, practices asking clarifying questions — perhaps starting with, “What do you mean by ‘alignment?’” and “How are we measuring customer sat targets?” and “What measures indicate that we’ve been successful (or not)?” What ensues is an iterative process of extracting information to help craft a well-defined problem that can be solved with data analytic tools and concepts.
In my Chief Analytics Officer work with clients, one of the most important (and challenging) parts of my job is to take what’s in a business leader’s head and turn it into a well-scoped business problem. I have a checklist of probing questions that I ask, such as:
- What, precisely, is the problem we’re trying to solve?
- What outcomes, if improved, would indicate that the problem has actually been solved?
- What data would ideally be available to solve the problem, and what data are actually available?
- How will the analysis lead to a solution?
Answering the last question is arguably the most important part, as it will determine the appropriate analytic technique — e.g., some simple insights or a more formal predictive or causal inference model. Here, I run through many “what-if” scenarios with the business team; for example, “What if the results show this, or that? How will that help you make a better decision?” Often, business leaders try to push this question off, suggesting we can consider actions once analytic results are available. That’s a mistake — knowing how the analysis will translate into a solution is a key part of formulating the analytic plan.
The Takeaway: To excel at problem-scoping, data leaders need good communication skills to talk through the problem with the business leader to arrive at the requisite specificity that will enable data analytics tools and concepts to meaningfully contribute to the business. Only then can the problem be turned over to the data team for analysis.
3. Problem Shepherding: Getting updates, gathering feedback
Once the problem is identified and scoped out, many data analysts go into isolation and only emerge when they have found a solution. This approach is highly problematic. To be most effective, the process requires a great deal of information sharing and setting of expectations — or what I call problem shepherding.
For data leaders, this means empowering their team to get more comfortable with providing preliminary results to the business team. Each exchange then becomes an opportunity to gather feedback. For example, “Are these initial results of interest to the business team?” and “Are we defining terms correctly?” From one update to the next, the results come together with sequential updates until the project is concluded.
This approach runs counter to how some data scientists prefer to work. Sometimes they get enamored with their models and their creative problem-solving techniques, and they can’t wait for the big reveal. But “big reveals” are a bad practice — in danger of backfiring. Too much surprise in a final presentation can put the audience on the defensive. The reason? Surprising results often prompt people to start questioning the underlying data and methods.
Every data model requires assumptions (e.g., what to do with missing data, how to treat outliers, etc.). If data teams actively working on analyses don’t disclose and discuss their assumptions ahead of time — and, instead, wait until the end — the business team is going to pile on the questions and nitpick the weaknesses. However, by bringing the business team into decision-making along the way, they will buy into the results and commit their trust.
Many business leaders have shared with me that the best final data deliverables are those in which there are no surprises. They’ve been working closely with the data team all along, and the final deliverable or presentation is simply a culmination of their work to date. This is how problem-shepherding gets buy-in through collaboration, exposing the difficult choices that data scientists need to make.
The Takeaway: Problem-shepherding sets up a process of providing regular updates and gathering feedback from the business team. Data scientists and team leaders who are strong in this area are able to encourage and facilitate candid discussions that ensure the final deliverable hits the mark with the business team — with no surprises.
4. Solution Translating: Speaking in the language of the audience
At this point, we transition from problem to solution, the success of which depends on how well data leaders and their teams have executed on the first three steps. More than determining a final answer, the data team must also deliver a solution that’s understandable and, therefore, actionable.
This isn’t just about putting the data in a chart or another visual display. Rather, the solution — whether data insights or a new course of action recommended by the model — must be conveyed in language the business team can understand. One tool I’ve recommended is the two-page data analytics memo, which highlights the most important elements of the problem to be solved. While two pages may seem highly condensed, especially compared to the hefty reports that data teams often generate, brevity is the power behind this secret weapon.
The two-page limit can avoid the temptation to go on and on about details of the data analysis and encourage focus on the recommendations being made and the evidence for them. I’m certainly not alone in advocating for shorter memos. Amazon founder Jeff Bezos required executives to present their ideas in six-page memos (versus a PowerPoint presentation) that could be easily digested and discussed.
The Takeaway: Solution translation requires data leaders to step back and consider how to make the most impact with their analyses and recommendations. By using simple language, while not compromising the complexity, data leaders who excel in this area can deliver the equivalent of an elevator speech to engage business leaders with compelling and understandable solutions.
As data and analytics become increasingly embedded in business decision-making and solutions, data teams must move well beyond merely solving the problems they’ve been assigned. Data leaders and their teams must focus instead on the terms “collaboration and communication.” This means becoming more adept at broader roles that help them spot the real problem, scope out its nature and importance, shepherd the process with periodic updates, and deliver and translate solutions that will truly make an impact.News