When I started to design Pillar, a digital support platform designed especially for mums, I knew I wanted to use a chatbot initially as a tool to test onboarding with the mums. I am a service designer that primarily focuses on a user-centric approach to digital products and I wanted to use technology to build the onboarding process. A chatbot felt like the best method. Being aware of the different types of software available my main criteria was something that was quick and easy to build (and therefore test). Beyond standard a/b testing, the goal of this prototype was to test the core feature of Pillar, which is where we match women to other mums 1 step ahead in the parenting journey to ask for their advice, and 1 step behind to offer their expertise too.
There were 3 key learnings that came out of the test.
1 – Know your audience, naturally
With any conversational chatbot, you have to know who you are talking to. Much of the work went into writing a script that sounded; friendly, natural and engaging. With any message style onboarding, you might be asking some personal questions in order to make sure you can match them up in the best way, so working on the conversational style is imperative.
Bringing in the mums from the early development stage allowed me to build the script with them. Making sure I was selecting the correct questions and also being able to recognize any nuances in their answers. Some of the questions were simple button responses for the first round of the prototype, and after some initial testing I learned through the process I needed to add some additional questions for the mums to really trust pillar, and believe that we are here for them as a friend. So we added the following additional questions into our final working app, which is now key to our match-up and are more nuanced and rely heavily on the mums responses, such as; If you could offer your past self a piece of advice about motherhood, what would it be? If you could ask your future self a question what would it be?
After my initial research I was aware these mums were using Facebook Messenger and WhatsApp to currently chat with their other mum friends. I wanted to utilize these existing channels and connect BotSociety to facebook messenger directly. Tapping into the right channels allowed for lower barriers to entry and ease of use for the mums.
2 – Fake it until you make it: Be the service
In order for me to gain the most out of Botsociety, I used the tool to onboard the mums by asking them questions such as; would you like to use your name or assign a username (is it important for you to remain anonymous); how many children do you have, what are their ages; where is your home (does cultural references affect our match up/support offered); What are the parenting styles that are important to you (co-sleeping/feeding habits/natural foods). All of these questions were iterated upon because I was the service.
I spent time developing my conversational style, my vocabulary in this space, I talked to mums, I questioned their thoughts, I immersed myself into their lives to fully understand the space I was working in and how I could gain their trust.
I then took all of these learnings and was able to test often and as early as possible. The key to using any kind of software such as Botsociety is to keep iterating on it until you have validated all of your assumptions. It is a cheap, easy to use software that allows for you to prototype (often on the spot!) how a chatbot will look and sound once it is developed!
3 – A/B testing will only get you so far.
The dreaded quantitative Vs. qualitative debate is becoming smaller and smaller, as both designers are learning the basic skills of data scientists and data scientists are learning design thinking methods. Much like how designers used to find it near impossible to talk to developers (and it is still tricky!), the knowledge gap is closing in and people are starting to see the real value in data-driven design.
A/B testing the chatbot allowed to see real data on why drop-offs were happening. Why at certain questions or after a certain amount of time were mums not completing the full onboarding process? Through Botsociety’s user test feature, I was able to obtain drop off percentages along with comments from real users which provided some further context. This allowed me to make tweaks to the design, but I also used my skills as a designer to get out into the field and test the chatbot by placing it in the hands of my real end users. I was able to observe certain things the mums were doing also during the onboarding process, such as, holding and feeding the baby, so one-handed ease was essential, which is why I redesigned the menu bar on the final application to be placed along the bottom as opposed to the top. Some mums would prefer to login with Facebook and therefore have their profile filled in and ultimately fewer questions to answer, and editable if necessary, again something I would not have found out by just the quantitative data.
All in all, testing, iterating and digging deeper into your end users behavior and trying to have fun along the way is the best way to tackle any design opportunity!