"Dumbest thing in data science"
A data scientist claims that writing a script to copy-paste screenshots from dashboard to email is the "dumbest thing in data science".
Sometimes you go weeks wondering what “content” to create, and how to write it, and then something literally falls into your lap, such as this thread on Reddit, which got posted on to Twitter.
For starters, I don’t understand how this thing is “dumb”, unless of course the original poster has an exalted opinion of labour theory of value (on that note, see this).
Back to the topic, this original post pretty much sums up a lot of the conversations I’ve had with respect to dashboards, and how people use them. Basically, senior management doesn’t like to look at dashboards. Laziness here is a feature, and not a bug - it allows you to prioritise what best to spend your time on.
The rest of the Reddit thread isn’t that interesting - most of the stuff is around data leakage from training to test set, building a regression problem as classification, and all such classic modelling blunders. The tweets related to the above screenshot are quite interesting, though, and some of them are worth mentioning / analysing here.
“Push vs pull” - our research has repeatably demonstrated that decision-makers seldom “want to use another tool”, and instead want everything delivered to where they consume the stuff already - such as email or Slack or even WhatsApp.
At Goldman, there was a report that one client liked. Other clients wanted something similar. They created a product but no one wanted it. So they resorted to sending email reports based on the product instead.
“Execs don’t read dashboards. They get their flunky or favourite employee to read it, which leads to a game of telephone”. And so the message gets garbled. If you want the exec to know what exactly is happening, deliver the message in a way that she can parse it very very easily. The moment someone else gets involved, there is noise.
There are also comments regarding Tableau’s “per seat” pricing model. Hinting that screenshotting Tableau might be a way to “cheat the system”. With static dashboards, this is one issue with per-seat pricing. People can forward (and based on everything so far, forwarding a feature, not a bug)
Someone from Zomato has explicitly commented that the CEO looked at a fancy dashboard and said “instead of this, send this report to every city lead every day”.
"If you make me log into your product, I will fire you as a vendor."
Of course, there are also tonnes of comments that say that execs are lazy for not wanting to look at the dashboards. Most of these come from accounts without real-sounding names.
I’d forgotten to mention this in my previous post on dashboards, but one sticking point is the need to be logged in through a VPN to access them.
In any case, with full benefit of hindsight, I guess the time that the OP wasted was in terms of building that dashboard. Had he figured that what the exec required was an emailed report with graphs, the whole thing could’ve been done in R / Python at a small fraction of the time.
Then again, if you have a hammer, you will see a nail everywhere.