The first professional analytics project we ever did was a blazing success. With it came recognition, opportunity … and somewhat misplaced confidence.

It led us to believe that with a robust approach, the right choice of tools and methodical analysis of data, we could understand how things worked and deliver a result that someone would pay us for.

How wrong we were.

When the first cracks started to appear, we were naturally inclined to start with engineering flaws.

Wrong machine learning algorithm … poor database choice … weak interface … badly designed data processing … confusing dashboards … elements that would have hindered the seamless customer experience we were angling for.

But data science can be overwhelming

So much technical content out on the web constantly makes you doubt whether you’ve chosen the right approach … and then, the right software.

And when new products promise giant steps forward with no upfront costs, it teases exploration. And a real danger of being sidetracked.

It might take a monk’s discipline to resist the temptation of diving into Tensor Flow with its Google heritage and the delicious tagline “An entire ecosystem to help you solve challenging, real-world problems with machine learning”.

Now if only time wasn’t in short supply … and unrecoverable.

Moving past engineering

Eventually, we were forced to question our limitations in other areas.

Maybe our marketing was poor. Perhaps we were solving an irrelevant problem. Or we weren’t communicating clearly enough.

After all, in a few short years, we’d been scalded by many of the pain points highlighted in those ubiquitous “Top 10 data science failures” lists.

Even knowing this proved frustrating — there were simply too many reasons, too many solutions, too many fixes, too much detail … while we desperately looked for a common thread.

To simplify a world that was threatening to overwhelm us.

Is there a common link?

After enough failures, the odd success and many tries at applying advice, dots began to appear.

And when we tried to join them, we found that for the most part, they all skirt around the human dynamic, the notion that data science and analytics …. the process of discovery, interpretation, and communication of meaningful patterns in data (Wikipedia) …. might just be about people and relationships.

John Thuma touched on this in his thought piece Why do data science projects fail. He reasoned that the complexity of math and data science will likely come in the way of communication between those who understand and those who don’t.

And without common understanding, what chance for the business or project to come out ahead in its analytical endeavours?

A tangled web of needs

We’d just spent hundreds of man-hours building a prediction model for traffic congestion in a 4G mobile network.

This, from a brief that stated “We need you to predict the likelihood of traffic congestion, given a week’s worth of historical data, so we can improve traffic management.”

Over the next weeks, we demonstrated our solution at multiple meetings and struggled to deal with some of the questions that were thrown at us.

Business case owner — What is the dollar value of these predictions over 6 months? (“Fair question but shouldn’t you be asking your team first?” we thought silently)

Middle manager — Marketing wants to know how many users are impacted during congestion. (“Wouldn’t the network supplier know better?”)

Project manager — How long to put this into production? (“Finally something we can answer!”)

Technical architect — We need to forecast traffic patterns over the next 6 months. (“Hang on, isn’t this a different problem?”)

Technical lead — Network capacity will be increased in 3 months, the prediction should work the same way. (“Argghhh!”)

Our answer of “We tested a model that can predict congestion to 70% accuracy” obviously wasn’t complete enough.

Incredibly, no one cared about prediction accuracy, potential biases, data integrity and sample sizes … forget the K-means assisted non-linear regression model that we were just dying to explain!

Instead, all roads appeared to lead to “your black box should be able to tell me what I want to know”.

And it wasn’t the first time …

The role of technology

We reluctantly concluded that in our line of work, having a working data science solution was merely table stakes … nothing more.

It simply gave us permission to hear the real questions on the table … questions that seemed to come up only when enough trust had been built.

Questions that provided just a bit more context about the state of the business and its people, and the relationships within.

At this point, one might mutter “Where the hell did that come from?” to the towering mountain, previously unseen, blocking the path ahead.

And suddenly, that 80% data science project failure rate that the folks at Gartner and others keep throwing around doesn’t sound so made up after all.

What we wish we had known

When there are multiple parties involved, how does one know what’s really going on in all their heads?

What business or even personal context is influencing what is being said?

And how does that relate to their perception of you and what you represent?

At a group level, how does one read between the lines to understand the dynamics between people on the same team?

And get a sense of how much “Will this make me look good” matters?

Given the usually limited amount of interaction, is it even possible to get much of an answer?

A perspective on data science

It seemed that over the years, we had been so caught up solving technical problems that we’d lost a little bit of our humanity.

Perhaps … could it be … that data science, that much-hyped and feted field of sophisticated algorithms, cutting edge software and insight-rich data, really is a PEOPLE BUSINESS? Loaded with technology distractions?

Which scared us somewhat because suddenly, it would no longer just be about mastering technology, optimising pipelines or refining precision … things we had some control over!

And the secret to achieving more in data science?

It might simply be the effort we make to connect with people, build relationships, speak a common language and understand where they’re coming from.

(Jan 2020) Lifeguard patrol on a winter’s day, Oceanside, Southern California