Keeping Up With Data #99
5 minutes for 5 hours’ worth of reading
There are many reports demonstrating the value of data and analytics to companies. The chart above is a prime example — the more analytically mature, the more profitable you are.
Leveraging data and analytics to make better decisions or streamline the operations certainly has a value attached to it. But moving on the analytics maturity ladder is incredibly difficult. And expensive. And long.
When we are pitching data as a value-creation opportunity to executives, we often meet with careful, cautious, or even sceptical looks. These are typically caused by past experience with expensive and delayed data infrastructure projects, failed analytical PoCs, or dysfunctional and unproductive data teams.
Every time there is high demand for senior professionals it drives the supply of junior ones, making it hard for executives to identify the data A-players who will get them among the analytical leaders.
So, the scepticism of executives is understandable. It’s up to us in the data community to deal with that. We need to do our best to increase the number of initiatives generating the desired business outcomes. Because every failed project doesn’t reflect badly just on us. It reflects badly on the whole industry.
I’ve read three great articles this week. One reminding that fighting complexity with complexity isn’t always the solution, one advising data teams to treat data as a product, and one about the dark side of training data collection. Enjoy.
- One Data Point Can Beat Big Data: We live in the era of very complex algorithms (think GPT-3 or Dall-e 2). These typically work well under ‘normal situations’ that are well defined and stable — with the scoring data assumed to be similar to the training data. But occasionally, the situations are not stable and the algorithms don’t work as expected. One approach is fighting complexity with complexity. The other goes the opposite direction and relies on very simple yet robust solutions. As always, there is no free lunch in ML, but remember there are other options than adding more data and features to your (often already very complex) models. (Behavioral Scientist)
- Data As A Product, Redefining Our Approach To Producing Value From Data: With the typical adoption of analytics sitting at 26% it doesn’t pass the Sean Ellis test (a.k.a. the 40% test) for product-market fit. What are we doing wrong? We regularly put too much focus on the defensive side of data strategy, which “tests the patience of business executives looking for immediate ROI when your team is stuck in defensive, downstream activities for up to 18 months — unsurprisingly, about the same length of time as the average tenure for a chief data officer.” We also shouldn’t be asking the business what they want or just sending people to data literacy trainings. The solution? Data teams need “to shift from treating data as a project to data as a product.” A product that is solving a big problem to its users, is intuitive, accessible and usable. One with a product-market fit. (Forbes)
- ‘It seemed like fun, I decided to join in’: Inside the biggest human surveillance experiment on the planet: The global population is creating training data for artificial intelligence all day every day. We are writing texts, taking pictures, interact on social media, we regularly share our browser history, and casually disclose many things like location through our phones. And as is the ML community moving the spotlight from algorithms to data, we see data tagging and labelling growing in importance. It is hardly surprising that some people offer their faces for the purpose of training algorithms behind surveillance systems. Disturbing, but hardly surprising. (The Sydney Morning Herald)
It has been a rough week. The Friend’s theme song comes to mind. But I hope that while it hasn’t been my week, it will be my month and my year!
Don’t forget to enjoy your weekend.
