"The show doesn't go on because it's ready; it goes on because it's 11:30." - Lorne Michaels, creator of Saturday Night Live. Data scientists can learn a lot from Saturday Night Live. SNL has a rule: The show goes on at 11:30. Not when it’s perfect. Not when everyone’s happy with it. At 11:30. Many years ago, I was responsible for performing the annual workcover premium rate calculation for the whole of Victoria. It was a calculation of the utmost importance - $2b in revenue depended on it...
6 days ago • 1 min read
For the 4 1/2 years of my PhD, I worked with a de-identified dataset that felt like nothing more than numbers on a page. Cold. Abstract. Disconnected from any real human experience. Each “person” was just a line in an Excel spreadsheet, with an ID in place of a name. When I started my first role in insurance pricing, my mindset initially remained the same. That was until my boss took me along to speak to a policyholder - putting me face-to-face with one of the people my data actually...
10 days ago • 1 min read
12 years in government taught me something surprising about data science. Making money and making an impact aren’t always the same thing. The easiest way to create value as a data scientist is to help your organisation to make more money. After all, everyone wants more money, don’t they? As Elon Musk’s recent $1 trillion pay deal suggests, even the richest person on Earth. Yet, while money is valuable, money and value aren’t necessarily the same thing. And if you work for a not-for-profit or...
13 days ago • 1 min read
Data science requirements gathering is about as popular with stakeholders as vegetables are with kids. The solution is also the same... Most data scientists dread these sessions. For stakeholders, the experience is probably far worse: Conflicting voices talk past each other. Senior executives dominating discussions. Junior team members too scared to speak. Political dynamics derailing productive conversation. The result is invariably requirements that miss the mark and doom projects from the...
17 days ago • 1 min read
Root canals and machine learning models have more in common than you think. A few weeks back, I went to the dentist because of a mysterious pain in my jaw. I wasn’t sure what the cause was, but after mentioning it to family and friends, the term “abscess” came up a few times. By the time of my appointment, I was convinced I was going to need a root canal. It took my dentist all of 5 minutes to conclude that the pain was purely muscular and everything would be fine in a week or so. That 5...
20 days ago • 1 min read
The traditional data science process is backwards. We start with data and end with storytelling. We should be doing the opposite. Most data scientists follow the same predictable process when delivering projects. Gather requirements. Collect data. Build models. Then create visualisations to communicate results at the very end. We all do it because it seems logical and few of us have ever been shown another way. But what if this traditional approach is actually working against us? What if by...
24 days ago • 1 min read
There’s a Calvin and Hobbes comic that perfectly captures why your data science career isn’t advancing. In the strip, Hobbes suggests Calvin prioritise his chores over having fun, so that once he’s done, he can enjoy the rest of the day without worry. The punchline is that by the time Calvin finishes his chores, the day is over and his mother appears to send him off to bed. Many data scientists are Calvin. We tell ourselves we’ll work on that high-impact analysis after we finish these status...
27 days ago • 1 min read
Why do stakeholders keep asking data scientists for the wrong analysis? They don't. They're telling you their symptoms, not their problems. "We have too much inventory in the warehouse." "Our sales are down this quarter." "Customer complaints are increasing." These aren't problems - they're symptoms of deeper issues. But most data scientists take these requests at face value and build solutions that address the symptom, not the root cause. Then they wonder why their technically perfect models...
about 1 month ago • 1 min read
I once had to choose between two candidates for a data science role. One had better credentials. The other got the job. I was hiring for a data science role within my team and we were down to the final two - let’s call them “Amy” and “Ben”. On paper, Amy was more qualified - she had better credentials than Ben. But in her final interview, all she kept doing was telling us how great this job would be for her career. She talked about the skills she’d develop, the experience she’d gain and how...
about 1 month ago • 1 min read