Build a bot for $285 — and validate it in 4 weeks

No coding advised

I’m going to walk you through how to validate an idea for a bot for less than $300. Example of a bot idea: “Wouldn’t it be cool to build a bot to order movie tickets?” Answer: Yeah, maybe. But let’s validate it, to see if other people think that’s a good enough idea to actually use your bot over and over.

When thinking about building a bot, coding is your last resort. I strongly advise against writing any code until you’ve validated your idea without coding. This post will tell you exactly how to do that.

Though there are several platforms out there that claim you can build a bot without coding. Forget about those too, at least for now. I’ve found most of them to be really buggy, aren’t that intuitive to use, and one of them totally messed up my Facebook page! (But, fair enough, because it’s still really early days in here in the botsphere, even Facebook is still figuring it out.)

I’m going to share what I did to get my bot from idea to protoype and validation phase in a month. I’ve broken down the posts into sections, so it’s easier to follow, but also so you can see what I did on a week by week basis.

At the bottom I’ve left my contact info, so feel free to get in touch if I left anyting out or if you have questions.

First, start a WhatsApp group

Cost: free – Week 1

Create a group, give it a name and a logo, (after all, this is a bot!) and then invite a few friends. It’ll feel a lot like a bot. You’ll have to create different groups for each friend (so each group has you plus a friend), but it’ll be a solid enough experience, and keep the conversations organized in one place for them.

I had about ten friends that I was testing a bot with using WhatsApp groups, and named each group like, “Bot name (Alex),” “Bot name (Alan)” and so on., with my friends being Alex and Alan. Ideally you want all the groups to have just the bot name, but if you do that, it’ll be more confusing for you as the “bot”, even though it may be a more realistic experience for your friends.

I had an idea for a bot that I wanted to test: booking train tickets. Because in Germany, booking train tickets, while not impossible, was cumbersome. And unlike airlines, in Germany at least, there’s really only one option: Deutsche Bahn.

I felt like that contain, giving the user less choices, may actually bode well for a bot, since you are going to be limited anyway by the UX. Turns out, it was a little more complicated than I anticipated. Playing the role of the bot gave me a chance to try on the idea, and then decide if we wanted to spend more time on it, before getting into developing a product.

Build a great landing page

Cost: $29/month – Week 2, part 1

Again here, you need not code. I created stunning, desktop and mobile landing pages with Instapage. With Instapage you can build an awesome landing page, set up A/B tests, and have complete control over the design, without coding. Anyone can do it, no matter how technical you are not.

You can use the forms feature to create a way for users to submit their phone numbers, which is was the first step in my funnel. Users put in their phone number, and then got a text message from me, the bot, to kick things off. Ultimately, you are trying to get a user to sign up, so make sure that’s prioirty numero uno on your site.

I found that the simpler the website, the beter the conversion rate. Remember, bots are new. The people who are going to be trying out these new apps are probably open minded, early adopter types. Don’t try too hard to sell them on your bot, that may back fire on you. But you can figure this out with A/B testing!

Run Facebook ads

Cost: Variable, but be prepared to spend $200 – Week 2, part 2

Now that you’ve got a great landing page set up, throw some traffic at it. While it’s great getting your friends to use your bot, and hopefully provide valuable feedback, eventually you’ll need to test it out in the real world. Facebook is the place to go for the fastest results, and lets you show ads to who you think your target audience is. You can get creepily specific in targeting on Facebook, so take advantage of that.

To get results, run ads over a period of three or four days, a week if you can afford it. Try A/B testing your ads. The ad creative (photo and text associated with the ad) is going to be where you have the biggest chance of swaying your conversion rate. I recommend spending $40–50 a day on ads at this point.

At the end of the week, divide the amount of money you spend on ads by the number of users that signed up on Instapage to see what your cost per user acqusition is.

If it’s cheaper than a cup of coffee, you’re on the right track. If no one clicks on your ad, something’s wrong with your offer. And if people click on your ad, but no one signs up, something’s wrong with your landing page.

Make sure your ad and your landing page say the same thing. Make sure the call to action on your Facebook ad is ‘Sign up”. Iterate, but make sure you track changes, so that you can know what variable caused change. Once it’s working, stop tweaking and turn up the spend.

Bot yourself with Sonar

Cost: free – Week 3

I’m stealing this phrase from my friend Alex. But before you build a bot, you should become the bot. For my group of friends, WhatsApp was totally fine. But once you start expanding your beta, and getting real users from the internet, you’ll need a help desk solution, or something web-based for sending and receiving text messages to your users. I hacked together something with Sonar, the best platform that I could find for sending text messages to useres.

Instapage lets you set up a form, which you can then point to an email. When a user lead came in via Instapage, I literally copied and pasted the user’s phone number, one at time, into Sonar to start the conversation with them.

If you are technical, you can upgrade your Instapage account, and use their API to send the number to a spreadsheet or to Sonar, to at least automate the welcome message.

Sonar offers canned responses, so you can work fast and consistently. You can also you have mutliple agents logged in at the same time, so you can outsource your messaging work later on. Which leads me to the next step…

Hire a freelancer

Cost: $56 for 1 week ($4 per hour) – Week 4

The next thing I did was hired a freelancer in the Philippines to be the bot for me. The reason I did this was two-fold. (1) I just couldn’t keep up with all the messaging alone, and (2) outsourcing would help validate the idea further. If you can’t outsource it, then you can’t scale it. So, by bringing in another person, I was proving I could eventually train a computer to take over.

For my “bot”, I needed about 2 hours of human labor messaging per day, which means I spent max. $56 per a week. What I found was surprising: as I outsourced the work, user retention increased! The reason for this is that hiring someone to focus on the bot, let them focus on the users. That mean more consistent interactions, and thus more replies from users.

Total costs: $285

From scratch to finish, my bot experiment cost less than $300, and took about a month to run. Of course, there were countless conversations with friends, late night research and all day thinking that went into bringing the bot to life. But, the gist of it was this:

Week 1: Validate with friends using WhatsApp
Week 2: Set up a landing page, Facebook ads 
Week 3: Start messaging with real people that you don’t know via Sonar
Week 4: Outsource your bot using a freelancer

After a month, you should have enough data to decide if you want to pursue the bot further. I ran this process on a few ideas. Your first idea probably won’t work out. Mine didn’t. Once you have the confidence, conviction, and, importantly, data, you should be ready to start building your bot for real.

Metrics that mater

Some of the numbers I was looking at to validate this for myself in order to move on to the next stage were:

  1. Click thru rate of the ads, showing, Are people interseted in this? I had a click thru rate of about 2.5%. That means that 2.5% of the people who saw the ad clicked on it. If no one is clicking on your ad, it could mean that the market isn’t that big.
  2. Conversion rate on website, showing if people are actually willing to give their phone number and sign up. This takes some iterating to get right. My first stab at this resulted in a 7% conversion rate, but after iterating I got it up around 25%. The higher this number is, the lower your cost of user aquisition is. You’ll want to throw 1–2 thousand people on your site to make sure you get enough data to sample, but you should alredy have a good indication of where it’s going after 200 visits.
  3. Daily active users: How many users exchange messages with the bot every day as a percentage of the total that have signed up. I wasn’t concerned with outbound messages, just inbound messages. Make sure you don’t lie to yourself here. I’d be as strict as possible in definig a DAU. For me, it’s all about messages received from users. Of course, your bot may not require users to reply to get value out of your service, and if that’s the case, how you define a DAU will differ.
  4. Cohort analysis: Which is to see, what percentage of users are active after 1 day, 7 days, and so on still use your bot. You’ll want to run your experiment long enough to have some data around cohorts, so you can see how sticky your bot is. I’m curious to find out if we will have “new norms” for retention, since messaging, by its nature is just stickier. Your retention should be going up as you improve your bot. If it’s not, it’s you (your bot), not your users. For my bot, I had some cohorts with 60% retention after 7 days, which I felt really good about.

If you feel good about your your metrics, you probably have enough validation, and conviction to start coding. Building a bot is a lot about asking the right questions, having the right tone, and buliding a personality. It’s a lot of fun, but there are also a lot of unknowns, which is probably a bit part of what makes it so much fun!

If you have any specific questions, feel free to shoot me an email: Thanks for reading!

My name is Schuyler, I’m working on something bot-related, backed by Techstars. Follow me on Twitter, @schuylerdeerman, or Medium to stay up to date about what I’m building —I’ll be sharing more soon.