What I’ve learned two weeks into building a bot version of myself
It’s been about two weeks since I put up the chikai.co signup page and one week since I launched Chikai 🤖. I went into it with only a vague notion of what it was supposed to be and yet a good number of people signed up. It was enough that I thought I should actually build it to see what would happen. So in a week, I built the basic system to allow me to test what people would do with it.
Here is what I learned.
To start, here are some basic stats. 21 people have signed up so far. 2/3 of them have engaged with Chikai 🤖 and of those that engaged over half initiated the engagement (i.e. they did not wait until they were prompted) either at the very start or later on. It is made up of about 25% women and 75% men, which is exactly the same demographic split as my Twitter audience (Medium needs to do a full audience demographic section on their stats page!). During the first week of signups, I personally knew every person who signed up except for one. The second week of signups, I knew less than half.
To be upfront, I responded to the vast majority of the conversations that came in, very few were scripted responses. The conversations I had were incredibly diverse and nuanced. I don’t think many of the questions could have been easily scripted by a simple bot via a regex and a template-based response. It made me appreciate how hard the NLP problem is and that it may be farther out in the future than people realize for this type of application. But the flip side is that the more people I interacted with, the more clarity I got in how a bot script or heuristic response system could be incorporated to either create a better experience for the user or to allow me to scale up the number users I could handle at once. There is nothing like real interaction with users to viscerally understand the nuances and details around this and it’s probably the most valuable insight that I’ve gotten from Chikai 🤖 so far.
Now what do people do with Chikai 🤖? It ranged from people asking what the weather was going to be like tomorrow to people trying to figure out whether it was a bot or human to people assuming it was just me and using it like a messaging app, but the most interesting one was some were treating it like a friend or wondering if it was a friend. I did not expect that last one and as I talked to some people about it, it’s not as unusual as you might think.
Another surprising thing was that the bot concept provided a lot of opportunity for humor. Because I was pretending to be a bot, I could make jokes about how cold the datacenter was or refer to myself in the third person as “human” Chikai and talk about what he was doing. Several of the users got into it and played along with it, so it became as much about fun and entertainment as it was about providing some kind of service. It also loosened things up a bit, so that it became easier to get to know the users better.
So all of these insights I’ve mentioned so far are from the user side of things, but probably just as interesting is the “administrator” side of things and the tools I created to manage and respond to all of the users. It’s a very simple system that is a wacky blend of Slack, Twitter, and traditional Unix command line tools like sed, awk, and grep. I won’t go into too much detail here, but I think there is as much opportunity for this being a tool for others to use to engage/manage an audience or a community of people as there is for it to be a bot. I have some off-shoot ideas I’m going to try out to see if there is anything of value on that side of the fence, but just wanted to note that it’s another potential direction that may bear some fruit.
All in all, it’s been a worthwhile experiment and I will definitely continue with it and see where it will go. I may write again in a couple weeks about how it was just a novelty and is now a ghost town or maybe it will continue to do well and I’ll have more insights to share that will tell me where to go next.
Post Script: Advice that I’ve applied in this experiment
I wanted to highlight some of the advice that I either personally received or read in a blog post that I took to heart and applied in this experiment.
Fake Doors
This is a technique that Jess Lee mentioned in a blog post, where you create these “doors” to products that don’t exist yet to get data on what features/products users actually want. When I put up the sign-up page for Chikai 🤖, I had not written a single line of code. I just wanted to see if anybody would sign up at all and surprisingly people actually signed up!
Progressive Reveal
This is something that Kevin Rose said to me in passing when I was meeting up with him while I was an EIR at GV. Basically he said that you should give people sneak peeks on what you are working on and slowly reveal it versus doing it at one big moment. I’m not sure I got the wording exactly right, but I tucked the advice away in the back of my brain and was somewhat skeptical at the time on whether it would work for me. Well, I totally used it with Chikai 🤖 and tweeted photos and shared thoughts as I was writing the code and about to release it. And it totally paid off. I went back and looked at a timeline of when people signed up and how it matched to when I tweeted or blogged and every time I tweeted or blogged, I got a few more signups. I’m sure this is obvious to most people, but I hadn’t seen it work up close and personal until now.
Do things that don’t scale
This is from a well-known Paul Graham post, but the most recent example of this that comes to mind is Ryan Hoover and how he started Product Hunt. A lot of what he did in the beginning was very manual and not something that would scale, but it helped build community and got the ball rolling. I totally used that approach for this experiment and did many things manually. One of the things I did was use a Google Spreadsheet to manage all of my users, nothing fancier than that. I’ve been following Lenore Estrada of Three Babes Bakeshop on SnapChat and saw her use Google Spreadsheets to manage her pie orders! She has *many* more customers than me, so it was cool to see how she manages things and keeps things simple and scrappy. I highly recommend following her and love watching the baking process. She also comes across as incredibly genuine and her tidbits and advice are as good as her pies.
I have a few other pieces of advice that I used, but those are the main three I’ll mention for now. Lots of people to admire and learn from. I love that I’m able to apply all the things I’ve picked up over the years in this experiment. So this post script is my small way of saying thanks to all of those people that have given me advice as well as those I’ve admired from a far.
