The learning loop revolution

Kieran Snyder
Textio Blog
Published in
5 min readJun 9, 2016


The concept of using social data collection to power a feedback loop is hardly new. Still, the speed at which users benefit has accelerated rapidly over the last decade. When Microsoft began asking users to submit crash data, it was to help engineers find bugs to fix for the next version. The update arrived three years after you sent the crash report. Network effects, yeah, but not exactly overnight.

Help Protect and Improve Microsoft Office, just three years later!

The first time I used Waze was a revelation, not because of the network effects it uses but because of how transparently the UI exposes them. By participating in the system, I improve everyone else’s driving experience, and everyone else participating improves mine. In being part of the learning loop, all of us reach our destinations faster than we could by ourselves.

Two important things have changed in the last few years to accelerate both network effects and the feedback loops that they support: first, the commoditization of machine learning technology, and second, the transparency in the relationship between using a piece of software and contributing your behavior to it. These two together power the next generation of technology: learning loops.

What are learning loops?

Learning loops combine machine learning with large-scale data sets that are socially provided. But what you provide isn’t your status, your photos, or the books on your reading list. You provide a direct feed into your behavior.

In learning loops, everyone in the network programmatically benefits from the experience of everyone else in the network. Just as Waze gets me to my destination faster in exchange for providing my whereabouts to the app, learning loops give me better results than what I can get on my own.

Driving in NYC without Waze? Good luck!

And they’re faster — a whole lot faster. In the new wave of learning loop apps, the lag time is not just zero, it can actually be less than zero. Learning loops not only react to behavior, but predict it.

Learning loops have taken over consumer software. Whether you’re figuring out what books to read, what music to listen to, or who to date, someone has built a learning loop to support you. Your experience is built upon millions of other behaviors that have been instrumented and captured from people just like you. Your side of the bargain? Use the product and contribute to the network.

Ask any VC how many consumer products they’ve funded in the last year that don’t include learning loops, and you’ll see how pervasive this is. When it comes to enterprise software, though, things are just getting going. But as learning loops take root, which is steadily happening at forward-thinking enterprises, learning loop software is transforming companies’ ability to compete, sometimes overnight.

Learning loops in the enterprise: sharing data as the way to win

One business segment where learning loops have been highly operationalized is in security and fraud detection. Consider Area 1 Security: Participants in their network send sensor data back to the core platform, which in turn makes the detection of security risks better for everyone in the network. If I’m part of the network, then I benefit from the experience of everyone else in the network.

It seems like a no-brainer, but remember that just a few years ago, for a company to share this kind of data outside their immediate firewall was a seriously big deal. But enterprises with foresight recognize that the security protection they can get on their own lags far behind what they can get by participating in a benevolent network.

In a totally different domain, we see analogous effects at Textio. Companies contribute their hiring data: the job posts they published, along with applicant stats and how long the role took to fill. In turn, when they write new posts, Textio’s predictive engine compares their language to the language of successful posts from the broader network and gives them concrete guidance that is proven to fill their roles faster with more diverse candidates.

Textio’s brain knows that this job will fill faster than only 35% of similar roles. Ouch.

A year in, companies using the technology fill roles 17% faster, with candidates who are both more diverse and more qualified, than their competitors who do not. As in the security software case, that gap is only widening month over month as the data set grows.

In both cases, companies in the loop get concrete advantages that change their competitive standing within days of adoption. Gone are the days of waiting three years for a version update; learning loops are fast. And of course the larger the network grows, the faster and more effective the learning loop becomes, and the more painful it is to be outside of it.

Ask yourself which other parts of your business are ripe for upending by learning loops: Finance software that considers your spending patterns in the context of everyone else’s to tell you how to budget? A CRM that tells you when to pitch certain companies based on when in the fiscal year they generally purchase services like yours?

Your data vs. hive data: no competition

This goes way beyond discovering a cool band or figuring out what movie to see. The social implications of the learning loop revolution as it takes hold in enterprise are profound. If you’re in the learning loop, you will hire, sell, market, and build better than anyone outside the loop can. Everyone inside the learning loop wins because they benefit from the experience of the entire community; everyone outside the loop loses because they’re operating with data that is just too limited to compete.

If you’re building a learning loop, developing machine learning technology is no longer a barrier to success. It’s also no longer your differentiator. Machine learning technology has become increasingly commoditized as larger tech companies release their libraries for all to use.

Instead, your success relies on applying this technology to novel combinations of social data. To get social data, you need to give people in your loop an experience that far surpasses anything they can get on their own.

Across so many domains, our collective experience and a predictive engine consistently solves problems better than one brain in isolation. Whether I’m trying to get home during rush hour or I want to find out how I sound to other people, learning loops take me vastly further than I can get on my own.

Consumer products have already embraced this. The writing is on the wall for enterprise. If you’re not inside the learning loop and programmatically benefitting from your network’s experiences, your business won’t stay viable.

Learn more about how language impacts your hiring at

Thank you to Gordon Ritter, Jensen Harris, Oren Falkowitz, Roy Bahat, Shivon Zilis, Steven Sinofsky, and Tim Halloran for their great discussion of the ideas in this piece.