Improving Data Science Storytelling
The data scientist’s second occupation is to engage business minds with analytics. For many data scientists, however, this wasn’t exactly the job they signed up for. When they are called upon to speak their findings in the language of business, some find it to be natural. Others find it to be difficult.
We established in my last blog that the data scientist works within the insurer in multiple roles to place analytics in front of insurance stakeholders. In my next two blogs, we’ll explore specific issues related to engaging business minds with analytics, an area where both data scientists and analytics consumers (business users) are far less comfortable. What can a data scientist do to make sure that his or her findings are communicated effectively? What can insurers do to give their data scientists what they need to communicate effectively?
There are numerous ways that a data scientist will be called upon to provide information to business stakeholders. There may be weekly e-mail reports, periodic presentations or impromptu consulting. For our purposes, we’re going to start with the idea that data scientists are often going to be called upon to present data’s story in person. Here are a few ways that they can improve their storytelling impact.
Set aside adequate time to prepare.
This principle is the most obvious and the most often unheeded. No matter what format a presentation takes and no matter how well analysts have done their job in analyzing data and finding conclusions — none of that will matter if the presentation flops. Inadequate preparation time will take excellent work and render it ineffective. Block out the calendar, turn off e-mail and cell phones, hide in a remote conference room…and focus on the story and any visual concepts related to the story.
Start with the audience.
Some data science stories will be small, told to an audience of one or two. They might involve only Claims or only Underwriting or only Marketing. With these, messaging may be simple and straightforward. But when the data science story will be told to multiple stakeholders, the success of the presentation should take all participants into account. Make a list of participants…their departments and their roles. Keep that list handy and use it to think through their individual departmental concerns/issues/opportunities regarding the data message. The last question presenters should ask themselves is, “Have I covered everyone concerned?”
Build the foundation for trust.
Set the stage by building the backstory. Don’t let findings rest upon comment alone. A data scientist needs to quickly establish a foundation of trust by making a short comment regarding the sources of the data used for analysis and the reasons those sources were used. What were the initial assumptions or questions being considered? The inevitable question, “Where did you get the data to support your findings?” can be circumvented at the outset with, “We took data from these sources and here is why.” Don’t linger here, however, because the detail may be boring. Place deep detail in the appendix.
Lead the audience. Don’t lose the audience.
A good data storyteller leads the team down the path of the story, trying to keep everyone on the same page, headed toward the climax of the presentation…the findings. For some, it is best to start with the findings and work backward, but this can often lead to losing audience anticipation and attention. It is best to quickly move from backstory, to plot line to final conclusions. By not overstepping the line from the findings into the potential impact, a seasoned presenter can often lightly sketch conclusions that will give those in the room business epiphanies. When business leaders generate their own ideas with the data scientist’s findings, they become deeply engaged in the process and grateful for the data science partnership. To accomplish this “light bulb” effect often requires a measure of restraint.
Don’t neglect the subplots.
Without sidetracking the presentation, don’t be afraid to highlight dead ends and errors along the way. These will also help toward establishing credibility and potentially even warn others against logical assumptions that may run contrary to the result of the analytics. For example, “We thought that teen drivers with straight A’s were actually better drivers than those with lower grades (a correlation) but discovered through the mileage data that they simply have fewer accidents because they drive fewer miles (a causation).” Sharing these types of occasional sidebars will assist in subtly encouraging a data-driven culture by showing data science’s value in identifying accurate assumptions.
Support every point with clear language and/or clear visuals. When the message is feeling muddy or sluggish, consider using an analogy or cut the point entirely. Sometimes interesting but impertinent information can simply be shuffled into the appendix slides. Likewise, acknowledge findings that may have no real story or may not relate at all to your story. Sometimes a finding is just a finding. It’s a stray point or a fact.
Humanize the analysis where it makes sense.
Data is very often people-related, so it makes perfect sense to use people stories to relate the data. When translating analytics into statistics, use the storytelling license to paint some memorable scenes. Instead of, “36% of our prospects are shopping during the workday, 33% between 10:00 p.m. and 1:00 a.m. and the rest on the weekends,” the presenter could say, “Tim, Sharon and Kyle represent our three most common prospects. Sharon (36%) is time-crunched and finds time to shop around during her workday. Tim (33%) is a night owl and takes care of his business in the evenings. Kyle (31%) can only find time on Saturdays and Sundays.” Listeners are more likely to relate to the mental pictures with a greater understanding of how to grab Tim in the evenings and how to help Sharon make the most of her few moments at work.
Condense. Tighten. Rehearse.
Before finalizing the presentation, look through the story for unnecessary fluff. Tightening the story will give it more punch. Depending upon the importance of any presentation, it may be helpful to rehearse in front of someone else. If visuals need help, solicit advice or help from a presentation-savvy coworker.
In the end, though data scientists have a story to tell, it may be more helpful to view analytics presentations and reports more as a conversation and less as a presentation. A well-honed cycle of scientific reporting and feedback will foster a partnership between business and analytics and allow creative synergy and excitement to flourish around data science within the insurance enterprise.
In my next blog, we’ll look at how an insurer can improve data storytelling by supporting its data scientists.