How to create a data flywheel for your business
How to create a data flywheel for your business
How to create a data flywheel for your business; AI is full of fear; you should act on it; Facebook shares three ways of dealing with this fear.
This is the Three Data Point Thursday making your business smarter with data & AI.
Want to share anything with me? Hit me up on Twitter @sbalnojan or Linkedin.
How to create a data flywheel for your business
AI is full of fear; you should act on it.
Facebook shares three ways of dealing with this fear
Let’s dive in!
How to create your own data flywheel
“Picture a huge, heavy flywheel—a massive metal disk mounted horizontally on an axle, about 30 feet in diameter, 2 feet thick, and weighing about 5,000 pounds. [...] You keep pushing and, after two or three hours of persistent effort, you get the flywheel to complete one entire turn. [...] You keep pushing in a consistent direction. Three turns ... four ... five ... six ... [...] The momentum of the thing kicks in in your favor, hurling the flywheel forward, turn after turn ... whoosh! [...] The huge heavy disk flies forward, with almost unstoppable momentum.”
The FlyWheel Concept explained in Good to Great - Jim Collins.
That’s a flywheel. The key success tool of many companies. The key question always is: How does YOUR flywheel turn?
So what about data? What some people call data network effects is what is behind the success of Google, Amazon, Airbnb, Uber, Netflix, Spotify, databricks,...
And it turns out, this is just a specific kind of flywheel. So the true key question is:
How do you build your own data (!) flywheel, and how does it turn?
Let’s back up a sec; what's this network thingy again?
A data network effect is when the value of a data-powered product grows as a result of its adoption. (Yes, this is a slightly more general and better-fitting definition than what Nfx uses).
Example: If you click on a series of Google results, Google collects this data and thus makes the algorithm better for ALL users, not just you. The whole Google product becomes more valuable because you use it; that’s insane.
Why is that special? Because other products don’t work that way! If I use a shaver, my neighbor's shaver stays the same… (unless I borrow it, which would be gross).
So how do you build a data flywheel?
Luckily, there is one template flywheel all these companies use, and it looks like this:
As you might guess, “data magic” is pretty abstract. And depending on the company, it works in very different ways. It could be…
More data leads to better algorithms.
More data leads to different products based on different algorithms
More users lead to more feedback lead to better algorithms on existing data
All of it leads to scale effects on computing, making the algorithms faster and better simply by better computing
Scale effects on the knowledge of the people creating these algorithms and collecting data
Money
A combination of any of the above.
So what should you do? We got you covered! Use that template, think through the ideas above, and then watch out for these five traps!
Your data flywheel does not accelerate.
If you cannot explain why your flywheel will fly, it’s not a flywheel. If you only get more data once and then never again, or your algorithms do not improve above a certain point, it’s not a flywheel.
You get lost in the hows of the data magic.
There’s an infinite amount of data magic boxes; if you have one that makes your flywheel turn, that’s fine unless…
Your flywheel doesn’t accelerate fast enough.
There is a straightforward question you need to ask to have a successful data flywheel: Does it turn faster than any competitor? You must step back and make it turn faster if it doesn't.
You don’t need the most extensive computing facilities, the most data scientists, or the most data. You need none of that, but you do need a flywheel that turns faster than anything else.
You’re basing your flywheel on money.
Yes, if you sell your product to more people, that will help you make your product better. That’s not a flywheel because it will never turn faster than your competitors; there’s always a giant whale with more money to spend.
You stop pushing
It’s a flywheel; it needs to fly! It is hard work, and you will constantly need to keep pushing; if you stop, you will stop accelerating. Other companies will surpass you, and their flywheels will reach a point of momentum at which they turn by themselves…
Want to share your data flywheel with us? Feel free to contact us!
Fear the fear of AI
AI will be smarter than us. And we think we need to be able to control that.
OpenAI launched a group to solve this problem within 4 years based on technical and scientific breakthroughs (yet to be delivered).
Essentially their idea is to train a monkey and tell him to do alignment research on humans - to be then able to create well-aligned humans smarter than monkeys.
Oh, but the monkeys stay in total control, or so the story goes.
So what? We argue it’s more about appearance than actual progress. But when every one of us is deploying more and more AI, we need to watch the fear of people!
The AI might be unpredictable, but the fear is not: With disruptive innovations like the one we’re undergoing at this moment in time, fear rises significantly and produces polarization.
What do we do? We can discard the openAI efforts for us, we think. But we cannot discard the fear; we must do something about it because every company uses more data and algorithms.
How Facebook tackles the transparency challenge
Closed source for algorithms is scary. Explainability as a must-have is on the rise. Algorithms and data need transparency for customers and users to use tools comfortably.
Facebook just started playing around with transparency for most of their used algorithms by featuring a “catalog of explainability.”
But that’s not all; fb also shares new tools for users to control what they see and transparency tools for AI in general.
What's in it for you? A small brainstorming list on how to increase transparency in your systems:
An explain like I’m 5 kinds of catalog to explain your algorithms
Tools to let people change the behavior of the systems (even small inputs let people feel in control)
OS tools (or data, frameworks, methodology) that help with transparency in general