How to Keep a "Go Big or Go Home" Attitude: an AI Application Story

Article by:
Anton Oparienko
8 min
AI-based applications are among the hottest trends of the 2020s. In a new episode of "Startup Stories", Upsilon talks to Roland Polzin, co-founder of Wing, an AI-based personal assistant app that works better than Siri and Alexa. We discussed what can give an AI assistant that extra value for the users, and which personality traits are important to cultivate in order to be a successful entrepreneur.

Anton Oparienko (COO, Upsilon): Today, I'm speaking to Roland Polzin, co-founder of Wing. AI startups enter a market which is already hot and is growing really fast. What does Wing do and how is it different from its competitors?

Roland Polzin (Co-founder of Wing): Wing is a personal assistant app, kind of like Siri. With the exception that it actually works. What makes us different is that we never say that we don't understand user requests. We do that using a hybrid approach, meaning that we use not only artificial intelligence, but also human intelligence. So, whenever our AI algorithm fails and does not understand something then the human will take over seamlessly, and take over whatever the request the user had. And then follow through and have it done for the user.

The second big difference to all those chatbots out there is that we get something done for the users in the real world. It's not only that we give back information. Of course, we can do that. What we can also do is - let's say, order pizza, or we can book a plane ticket or something like that. We work like a real personal human assistant, only that we're an app.

Roland Polzin, co-founder of Wing
Roland Polzin, co-founder of Wing

Anton: You say you use both an AI algorithm and human intelligence to process your users' requests. In order to provide your current user base with the best service, how much of it is done by an AI and how much is done by people?

Roland: I can't give you the exact number, but of course it's a lower percentage right now because we're still developing the algorithm. And there are really two main things, let's say, how we improve the algorithm. One is with hardcoding. And we are seeing it right now already after about two weeks after our launch. Let's say about 18% of people's requests are related to food delivery. And now what we can do is we can go and hardcode skills to automate food delivery with APIs from third-party services like DoorDash, Uber Eats, or Postmates. So that's what we will do for all the big buckets out there. That way we can achieve an automation rate of well over 50% within half a year or so.

And then the other thing is we leverage deep learning. That means that our AI algorithm is constantly listening to everything that's happening, user requests and human operators that service those requests. And our AI algorithm learns from those interactions that the human people take. And then actually becomes better at predicting how to answer and service similar requests in the future. And so at some point, we'll not only have the hardcoded skills but also an algorithm that learns by itself and gets better and better every day.

Wing AI
Image Credit: Wing Website

Anton: What was the weirdest user request to Wing that your team had to solve?

Roland: We had someone who requested a private jet which we could do. We didn't have it automated because it's pretty rare. We also have interesting requests from people who need help from a real assistant in terms of creativity. One person asked "Can you give me some ideas for quotes that I can put on a bumper sticker?"

This is very interesting because you get to know your users better and of course collect information in a safe we call the Vault which enables us to be even better in being a personal assistant than any other product out there.

Anton: You are a competitor to the giants like Siri and Alexa. The key thing that you are doing is actually introducing humans where AI cannot solve it. And I as a user would not receive "I don't know what you mean". How do you scale this approach? And what happens when your user base grows very big very fast?

Roland: Scaling is a very big topic for us. The biggest asset for us is, of course, the algorithm. And the better the algorithm gets the lower the need is for human interaction. By progressing forward, the more we can improve the algorithm the more users we can let on exponentially.

How we do onboarding right now is we leverage a waitlist whenever we have a bottleneck in human operations. Let's say we have a hundred more people that want to join today but that would mean we need to hire another human operator. Then we put those people on a waitlist, hire a new operator, train them and a couple of days down the road when it's done we let the people on.

We control growth, so that's one important aspect of scaling. Over time, the need for human labor will diminish, but for the foreseeable future, we'll always have some human element to our platform.

Anton: Where do you take courage to enter this complex market with multi billion companies? Where is it coming from?

Roland: Many investors ask "So, what do you do when Google becomes better? And what do you do when Siri and Alexa become better?" It is a valid concern for sure. However, we do really regard those chatbots (let's call them chatbots) as indirect competition because they are purely AI-driven. So what's always going to happen with them - they do have a limited set of skills and at some point they hit a roadblock where they can't understand and they can't do something.

Inevitably, there will always be an instance where the user says I want to do X and the chatbot says I do not understand that. That gap will probably exist for a couple more years, probably 10, 20 years down the road until those algorithms are much more sophisticated and they can do much more. We bridge that gap now!

Anton: What led you to joining a startup in California (Newport Beach) and become a co-founder of a company? What did you do before that and what made you think - ok, I'm ready for this?

Roland: I would say I always had that entrepreneurial mindset and the curiosity and motivation to start my own business. In high school, I did have my own computer service business where I installed computer hardware and did troubleshooting for people's computers at their homes. I did that on the side next to high school and then I really wanted to do something that's bigger than myself.

So, I actually joined the military. I signed the 12-year contract with the German army and became an officer, served in Afghanistan, and in Mali as Chief Communications Officer for the United Nations. And when that 12-year contract ended I thought to myself "Ok, do I want to stay with the military and maybe become a general, or I want to do something else". And I actually thought "Let's do something else". I went to the US because I was intrigued by its great economic success and the freedom exhibited in society.

I did an MBA which I'm finishing right now. And I still had this idea of pursuing entrepreneurial opportunities.

It was just a lucky coincidence that the co-founders of Wing found me on LinkedIn and hit me up because they were pivoting then. Back in 2018, Wing was just an on-campus student assistant, but for this new business idea of a ubiquitous professional assistant, they were looking for new talent with leadership experience. They found me and I was lucky enough that it happened. I fell in love with the product and joined the team.

Anton: You mentioned inner energy, motivation, and curiosity. So is it the most important thing that you have to have your own startup and business? Maybe there's something else that you need personally?

Roland: For sure. I would say it is not that much different from the traits you need as an officer in the military, because you do need determination, you do need to be very self-motivated to do things, and you do need to have a hands-on attitude. You got to be willing to take action, to do things yourself, to push things forward and stick with it, and if they fail try again and pivot, do something else, keep pushing and keep doing it. That's I would say the core of it.

Anton: What were the influencers whose example helped you be an entrepreneur?

Roland: Of course, there have been many people that shaped my thinking and influenced what I value. Specifically for my entrepreneurial endeavors, I would say it's pretty much Elon Musk. To me, he embodies that pure entrepreneurial mindset: coming from a background that didn't provide many opportunities, but employing that self-made mindset and hands-on work, and then build something and make millions of it - and don't lay back, don't say "Okay, I've achieved everything", but start over with something new.

That's what he did after PayPal - start over, and start actually two companies that were blowing people's minds. A space company. Who would do that? And an electric car company. Who would do that? It's crazy, right? That's what I admire and it's what keeps me pushing my company as well.

Wing task management
Image Credit: Wing Website

Anton: How are you developing your startup? You obviously have an operations team that's handling users, but how is the AI part developed?

Roland: We do have a core team of engineers that are working only for Wing. They are very well versed in web development and AI programming. That's how we are able to build all that in-house. All the software we have, the app, the AI algorithm, all the backend software that our human operators use - that's all built in-house and it's all proprietary. That's our strong suit. And we have a couple more interns that help with fleshing out the AI algorithm. We have four more full-time tech interns at the moment which is pretty awesome. And a good part of being in the Irvine area is that we have access to the UC Irvine campus which has a really good Computer Science program. That's where we draw talent and that's how we are actually able to build all this technology in-house. And that's why we are more of a technology company than anything else.

Scaling the Product and Growing the Team: Upsilon's Comment

For companies like Wing, inevitably comes the time when the user base overgrows the capacity of the product to support operations without failing. It is crucial not to miss the moment to grow the team and scale your application. It is not uncommon and actually makes a lot of sense to find a trusted software development partner. There will be no competition in hiring between your company and tech giants like Google or Amazon. In addition, you will be able to operate on a lesser budget in comparison to what you might have spent on the in-house team growth.

In our experience of providing services for early stage startups, the tricky part lies in matching the remote team to the startup culture and promoting the sense of unity between in-house and remote team members.

Extra Tips on Working with a Web Development Team

Here are our tips for a successful cooperation with a remote web development team:

  • Extensive communication: a global team chat, daily and weekly meetings, company updates - all of that is vital for successful development of your product;
  • Do not disregard video calls;
  • Dedicate time for providing feedback one-on-one;
  • Transparency of data reporting and access to startup tools for task-management and other purposes is a must;
  • Encouraging the knowledge exchange and discussions between your core and remote team;
  • Provide remote team your vision for the product, don't just give them tasks out of context;
  • Have the team visit and do workshops.
We provide: Data Science Consulting Services

Is your startup in search of a tech partner?

Helping startups develop products is Upsilon's specialty.

Book a Call

scroll
to top

Read Next

15 Deadly MVP Mistakes to Avoid at All Costs
MVP

15 Deadly MVP Mistakes to Avoid at All Costs

13 min
What Is a Concierge MVP and When to Use It
MVP

What Is a Concierge MVP and When to Use It

6 min
Mastering Startup Pricing Strategy: Unveiling the Steps, Types, and Real-World Examples
Building a startup

Mastering Startup Pricing Strategy: Unveiling the Steps, Types, and Real-World Examples

14 min