Starting a company around a data problem is one thing. Starting it around a business problem which data can solve in new ways is another, entirely different, kettle of fish.
Having traversed the gulf between academia and the Real World™ before, I’m seeing a lot of the same tension. I like to think of myself as a pragmatic person — pragmatic and curious—which means that on a daily basis I’m trying to juggle both the ideal way to do something and the today way to do something.
I’m a product person. Well, I like to think I am. I’m also a data person, and a former scientist who prefers to think in terms of experiments and playing in a data sandbox. I realised recently this is exactly the same mindset as lean UX, which centres around the mantra ‘build-measure-learn’ — in fact, the same mindset as the whole lean movement in general:
Nasty hacks and duct tape
The problem is, data isn’t perfect and users aren’t perfect. We don’t get a startup kit in the mail with a set of clean vectorised signals ready for classification. In fact, I’ve spent the majority of my time over the last couple of months on the data input part of the puzzle: getting and cleaning data, something they neglected to teach at university, is almost all the battle.
The fun comes when you have to remove the duct tape. You cobble together enough of a system to make the demo work, with a few good hacks thrown in for great justice. Then, newly armed with a fresh understanding of the problem space and the hurdles facing a good solution, you dutifully architect out a system that will work and scale. You start building it, but data quality issues rear their head and the investor demo doesn’t work quite as well as the hacky one. After playing whack-a-mole and whack-an-API, you finally sit back, ready to support the product.
And then, of course, the product side needs something totally different. Fundamentally, planning for now and planning for tomorrow are not the same thing: as soon as you start building up the next layer of product features you realise a whole lot of things need to change. And so it repeats.
Lean data science
There’s no such thing as lean data science. Data science is by definition lean, and it only gets leaner if you think in terms of weekly deliverables rather than yearly papers.
Lean data product, though, that I’m learning.
Lean data product is about understanding your users and assumptions well enough to plan out data-facing features which directly impact product learning. It’s about prioritising your data enough to let the hacks slide, for now, so we can test whether the product will ever need the ideal solution that’s bursting to get written. It’s about knowing when to throw data and code away, and when to change features to take advantage of new data insights. It’s about balancing promises and reality.
At the end of the day, though, to me: it’s all about making it work.
I’m planning on following this up with some thoughts on the data/product team at a super early stage. But I have to go apply some duct tape first.
Email me when Jennie Lees publishes or recommends stories