Finding Carmen

Preston Richey
Moonshot Lab
Published in
5 min readNov 10, 2017


After a year and a half working from the 4-story TWA building Barkley calls home, you’d think I’d know my way around. Here, meeting rooms follow a theoretical naming scheme: celestial bodies (Big Dipper, Galaxy) are on the 1st floor, people (Neil Armstrong, Ellen Ochoa) are on the 2nd, planets (Mercury, Neptune) on the 3rd, and everything else (Flight Deck, Magenta Lounge) on the 4th.

The historic TWA building that Barkley calls home is deceptively large

But even if you’re on the right floor, finding the right room can be a challenge, especially if you’re scrambling to get to a meeting you’re already three minutes late for. Not to mention all the rooms that break the naming protocol (Michael Bay, though a person, is on the 3rd floor next to the video team) or are just too obscure to know off hand, like Er’Kit, the fictional planet in the Outer World Territories of Star Wars lore.

Where in the World is Uranus?

That’s all to say, finding your way around here at Barkley can be a problem. I know that I’m not the only one who thinks so. I see dozens of people a week wandering around, looking puzzled. Some ask, “Where’s Uranus?” (Mark Logan particularly enjoys this one, although it’s a joke that works better spoken aloud than in print). And though my official title happens to be Creative Technologist, it might as well be Creative Problem Solver. So, I and a small team of writers and designers put our heads together and created Carmen.

Carmen serves up directions, maps, images and occasional snark.

This quarter, Moonshot is doing a deep dive into Conversational Interfaces. What better way to put these learnings to use than to create a chatbot to help Barkley partners find their way? We first decided on a platform — while Facebook is nearly ubiquitous and has the benefit of delivering messages over data, we chose SMS because we didn’t want to exclude anyone. We also wanted to most nearly match the model of simply asking someone for directions.

Carmen, named after Carmen Sandiego, will help you find any room in the building at any time, night or day, and won’t even get annoyed if you forget where Ellen Ochoa is (for the 3rd time this week). You can ask for instructions to a room, as well as a picture or map of the room in context. Just under a month after being launched, Carmen has helped 123 Barkley partners, received 1,147 messages (of which 90% were understood), and sent 1,121 messages in return.

Carmen’s streamlined graphic system serves up visual cues to aid navigation.

But Carmen Why?!

Throughout the process of building and updating Carmen, we’ve learned a ton about how people interact with a chatbot, what they expect and how they prefer to receive information. By logging each message Carmen receives, we also have insight into which rooms in the building are most often searched for, as well as questions that Carmen doesn’t currently handle but maybe should.

Users expect Carmen to be able to point them not only to rooms, but also to individual people (as if anyone actually sits at their desks!). Some users want Carmen not only to know the location of rooms, but to book rooms or to tell them if rooms are available right now. Cool features, to be sure, but well beyond the scope of Carmen 1.0. And some just ask the really deep, existential questions that Carmen can’t and probably shouldn’t answer. All-in-all, this information is incredibly useful and can inform long-term decisions about wayfinding and onboarding new partners.

Carmen’s intelligence is basic, but we learn a lot about our users’ expectations from her error logs

We’ve also learned that people are more comfortable interacting at the speed of human conversation, even if a bot is technically able to respond more quickly. We introduce an artificial pause of a second or so when sending a series of messages, since it can be jarring to receive 3 or 4 messages in less than a second. This practice can carry across to bots on platforms like Facebook, where ‘typing’ bubbles can be used to indicate that the bot is ‘thinking’. Even if it’s not technically necessary, it gives the bot a bit more humanity, and fosters a more personal conversational feel.

Basic Bot

We’re still early in this new era of conversational interfaces, and we still have much to learn. Carmen is one of the most basic kinds of chatbots you’ll find, there are a lot of features we could have included but didn’t and a lot of technology, like NLP (Natural Language Processing), that she doesn’t possess. But as basic as she is, she proves that chatbots can be simple and still highly useful.

For Moonshot, building is an intrinsic part of our method of technology exploration. Creating tools like Carmen offers us a way to learn about the technology, processes and best practices, while also improving the lives of our Barkley partners. At the same time, it also exposes them daily to the tangible benefits of this technology, and we hope that experience inspires ideas for new solutions to client problems and opportunities in the future.

Next: The BotShop

And we’re not resting on Carmen’s laurels. We’ve begun working on more bots, both for clients and for Barkley. We’ve convened a group of Barkley partners called The BotShop to refine our processes and work on a more expansive Barkley bot. We’ll introduce you to him/her/it in the near future. Stay tuned.