While Artificial Intelligence (AI) is not a new concept by any means, it's still relatively new in the design realm. Engineers have been working at making AI and machine learning as a whole smarter and more independent for years. At Dollar Shave Club, our brand has something unique that no brand in our category and not many other brands have, a defined and distinguishable voice and tone. From day one when our CEO Mike Dubin busted onto the scene with the outlandish and some would say daring viral video, it set the tone of the brand in a way that the Internet had not seen at that time.
Fast-forward 5 years and the way brands can speak to consumers has changed dramatically. We've got new communication channels opening up every week and the main-stream is adopting these new channels at a rapid pace.
When I was managing the Summer UX Intern, I gave him an assignment to explore how bots could positively effect both our brand perception and the end-users experience in finding the right products for them. Naturally, we started prototyping on Facebook Messenger platform, primarily because that's the channel that we spend a majority of our marketing and acquisition dollars. Oh, and it's our most effective / highest converting acquisition channel. We iterated through a few ideas and ended up with a bot that's sole purpose was to help you find the right Boogie's hair-styling product for your needs. As we shopped the idea and proof of concept around to different departments, we needed to work through a large bot strategy if it was to actually live.
After auditing the landscape of emails in the world, we then had to audit our own system. Like I said earlier, we had over 50 unique templates, 250+ variations of emails, A+B tests that had been running for over a year and legacy emails. Looking at these emails, I determined the similarities between them and tried to fit them into buckets. These buckets would then turn into our 10 basic templates. We had to be really selective and creative in how we these over 250 emails with drastically different content could work together. We wanted to keep this system future forward, so we went a step further to brainstorm and design concept emails for a totally new CRM onboarding experience.
Like any project, what we thought we were building ended up being something completely different once it was built. Hell, it's probably different as you read this. We change(d) it everyday. Through our quant research and analyzing quant data, we evaluate the dead ends and what exactly people want out of this bot. Bots are different than any interface you'll ever design. You can think of a strategy, the best strategy you could possibly think of, but at the end of the day people will use it how they want. And bots, are the extreme case of this. The amount of use cases are hard to manage, which makes formal UX documentation kind of pointless. For something that is forever evolving, growing and getting smarter, what's the point?
In order to sell this channel in at the Executive level, we needed the bot strategy and data to prove why we should invest in it. By interviewing and gathering data from our Member Service team (MSA), we mapped the current MSA touch points with our members and the time of day. Doing this helped discover an opportunity, automation through Facebook Messenger. With 800+ messages from members a week (and growing) and a more than 24 hour response time, we could surely add a customer service layer to the bot which would eliminate 78% of MSA contacts through this channel. Those 78% of contacts we would be eliminated through automation are often times 'one-touch' tickets in which the member has a simple one response question.
As a business, DSC's goal is that we want to help members find the best grooming regime for them across our entire product portfolio. What started as a prototype that we were testing remotely with users, turned into something much longer. Essentially, this project fed into my OCD, really bad. But that's a good thing. I continued to refine and refine and refine. As guests and members would use the bot I would tweak flows multiple times daily and add to the Natural Language Processing (NLP) with every conversation. Eventually, the bot got to a really good place. Between the constant communication with the DSC Member Service team, guests and members, the NLP and overall MVP feature set of the bot was ready to launch. This process really helped open a lot eyes to people in the company. It helped them realize that we can invest little resources and little time to produce a platform or new channel to communicate with our members and potential members.
After calculating all of the current Facebook Messenger data we had available to use, we decided to start with FAQs and basic Member Service as we dove into the development of the bot. Daily, we would meet with the DSC Member Service team because they are on the front lines, speaking with and answering member concerns every second of every day. The whole Member Service team helped pin-point exactly what people were asking for and how they were asking. One of the first features of this track that we implemented was live-chat. This, we assumed, would be used quite frequently for frustrated members who just was an actual human to speak to. ChatFuel has a really awesome live-chat hand off to the Zendesk Messenger platform, which is what we leveraged for this functionality and to collect all Facebook Messenger tickets moving forward.
As we the Member Service portion of the bot became more sophisticated and functioned off natural language pretty smoothly, we transitioned into remote usability testing with real members and non-members. We ended up testing the bot in 4 different phases of its development with 50 non-members and 30 members. This helped uncover great features that help members beyond just FAQs and member help. Some of these features include: DSC Razor Match Tool, Subscription to latest DSC Original Content articles, search and read Original Content articles based on various tags, detailed flows of a few member favorite products and easy exposure to all of our grooming brands and categories.
Another insight from our usability testing was that whenever we enjected more emojis, fun language, insterstcial conversation or funny .GIF, the delight factor and love reaction we heard from the test participants was very noticeable. As the remote usability testing progressed we collected 5 words that described their experience with the bot and this interaction changed their perception of the brand, from each test participant.