Digamma.ai CEO Q&A Series: Kieran Snyder of Textio

Leave a comment
Artificial Intelligence / Machine Learning
textio

Digamma.ai CEO Q&A Series: Interview with Kieran Snyder, co-founder of Textio

1. You previously worked at Microsoft and Amazon and have a PhD in Linguistics and Cognitive Science. Today, you and your team run Textio, an augmented writing platform that has been used by companies such as Atlassian, NVIDIA, Square, Starbucks, Twitter and Vodafone. How did you and your co-founder and CTO Jensen Harris come up with the idea for Textio back in 2014?

We started Textio with a very simple vision, which is that if every time you wrote something you knew exactly who was going to respond. Now imagine you had the data to tell you why someone would respond. If you know who will respond, and why, you can change your approach to get a different result.

Jensen and I founded Textio together, though we come from very different backgrounds. The union of those backgrounds makes Textio what it is today.

My background is in measurement in of language. I was a Linguistics and Math major in college and my Ph.D. is in natural language processing. I spent around a decade leading natural language and search at Microsoft and Amazon. Jensen’s background is in user experience. He designed the first ever UI for e-mail in the form of Outlook. He led the design and implementation of Word, PowerPoint, Excel kind of core Office UI. Where my experience has always been in measuring the impact of language in software, his has been building enterprise software that’s usable by a billion people.

When we started talking about the future of digital writing and looking at what had been built and where things were going, we realized that the collection of predictive technologies made it so we are now at the point where you can actually tell which people were going to respond to your writing ahead of time. From there, we set out to build a writing experience—what we call augmented writing—that gave people this kind of powerful experience that could really change the measurable outcomes of their writing but in a way that anybody could use. You don’t need to be an engineer or a linguist to figure it out.

Originally, the very first thing we built was actually a KickStarter predictor because all of the outcomes were known. It was a great place to prototype our technology.

Obviously, now we have deep data around job posts, but it’s still only the beginning. We think that in five years there will be a Textio power augmented writing experience everywhere showing you how to get better results.

2. You mention in a recent blog post that your $20M Series B round led by Scale Venture Partners will help your company build upon Textio’s first project, Textio Talent, an augmented writing experience for job posts and extend it to the other important things that people write every day at work. What are the next set of projects that you are excited to tackle? What new features are you looking to add to your platform?

You’ve touched on two important points here. One is making the existing application, Textio Talent, even better for the people who use it and the other is about extending the platforms to other kinds of writing. We are really fortunate that we have a number of large enterprises using Textio now as a key part of their hiring and communication process. Textio needs to be robust for companies like CVS, Johnson & Johnson and Expedia to use it on a global scale. These companies are deploying Textio to thousands of people involved in hiring a huge range of roles worldwide. The first big focus area for us is making that workflow for those people even better, more seamless, and more connected to other things that they’re doing.

The second piece is really what the investment round is about—starting to take our existing platform for job posts to begin building models that work for other kinds of writing. Customers are currently seeing 25% more qualified candidates apply for roles, and filling roles two weeks faster – how do you take those results and bring them to other types of writing. Right now, we are working with our customers on one to one candidate communication—the emails that you use in hiring. For most recruiters the process of hiring someone is more than just a job post. You also do a lot of directing in case something like, “hey Jensen, I saw your background, it’s really interesting and you should probably come work for my company.” How do you optimize those emails to make sure you get the best chance of return with a really specific and measurable results? Working on improving candidate e-mail is a really important investment for us this year.

And you can imagine how the platform can go beyond that. It opens up a lot of interesting other kinds of communication not just in the recruiting context but also for the enterprise itself.

3. Jensen discusses how Textio isn’t about making words adhere to some grammar rules — it’s about producing a version of a piece of written work that gets the outcome you wanted when you set out to write the email in the first place (e.g. great applicants, the ideal client, etc.). How does your platform achieve this?

It’s important to differentiate Textio from conventional grammar checker-type products that have been around for 30 years now. For example, a product like Grammarly builds on a very old metaphor, which is: “find 100, 200 rules and implement them.” Textio doesn’t really apply rules. The core of what the platform has been built on, in the scenario of a job post, is a massive data set. Today, the data set is over a quarter of a billion job posts with their outcomes attached. The outcomes are things like: how long did this role take to fill, how many people applied for the job, what demographic groups did you reach, how qualified were they, and were they good enough to get interviewed. All of these are real sets of outcomes that companies are tracking today about the performance of their job posts. Textio models all of that data and can find the patterns that work to get those different outcomes.

The questions then become: how do we write a job post to get a high proportion of qualified people through the door? How do you write that engineering job so that you optimize your chances of getting more women interested in applying? How do you make your finance job in New York perform better than the thousands of other finance jobs in New York? What Textio is doing is continuously taking in data from our customers and modelling what’s working in the market today, because the patterns change over time. One example is big data, four or five years ago when you used the phrase big data you got more people through the day. Then the phrase became popular, so popular in fact it became a cliche. So today if you use the phrase big data, it actually degrades the performance of the post. And by the way, AI is starting to ride the same curve. Textio is constantly taking in this data, the words you use, the format of your post, and many other permutations, and providing you guidance on what will work best in today’s job market.

4. Why are humans, not machines, still the best writers? Do you think we will reach a point where machines will develop their own unique voice?

Let’s take job posts as an example: If there’s one best way to write a software engineering job post and everybody listing a software engineer job in the entire world use that exact set of text, you can imagine what would happen? It would no longer perform well or poorly. The job post would no longer even be relevant since everybody would have the exact same thing. When a text is purely machine generated, it becomes the lowest common denominator. It loses that ability to stand out. The thing that makes a platform like Textio work and evolve over time is that people innovate language. Somebody figured out what came after big data. Somebody is figuring out that deep learning is coming after AI. The point is that people’s voices innovate.

And the patterns that work at one company may be different then the patterns work at another company, even if they are hiring for the same type of role. A great example here, is Expedia, one of our customers. They hire here in the Pacific Northwest and they hire lots of engineers, so there is significant talent competition from companies like Microsoft or Amazon or Boeing. Many of the patterns that work in Seattle for hiring engineers work for Expedia too. However, at Expedia in particular, when you talk Airbnb compete, it fills a job faster, but when you talk about travel agencies, it degrades the performance of the job. It fills more slowly and those patterns don’t apply at Microsoft or Amazon. It kind of makes sense because the type of person who might want to work at Expedia is passionate about travel. So, they’re sensitive to how you talk about travel. It’s not the case that one machine can produce kind of a one size fits all guidance that will work equally for every person.

5. How do you see the field of NLP supporting the development and improvement of augmented writing over the next few years?

I think the thing that makes a platform like Textio’s interesting is the way that it combines NLP with a classical supervised machine learning techniques. Machine learning in particular is really good at making a prediction of something that’s going to happen. Will this job fill quickly or slowly? I can build you a classifier that will tell you. Machine learning is really bad at telling you why and, in the case of a product like Textio where we’re not trying to make a prediction, we’re trying to give guidance to a real person so they know what changes to make to get that job post or document to a different outcome. You really need to figure out why, what changes? What changes can they make it? Of course you need to present that guidance in a way they can follow. NLP is how the platform starts scanning, labeling, and identifying all the various language factors that could make a difference in how a document is going to perform. NLP allows us to really get to that explanation—not as what’s going to happen, but why. Then, how would you translate that into guidance for a real person? How do you know you need more bullets? how do you know this isn’t verb-heavy enough? NLP allows Textio to go way beyond just the individual sort of word or n-gram that you see in machine learning.

With over a decade of AI experience, Digamma.ai’s team are your trusted machine learning consultants, partners, and engineers.

Leave a Reply

Your email address will not be published. Required fields are marked *