When I joined our Growth product team in 2021, I could not have imagined how our work would drastically change the user experience for those who interact with HubSpot Help. You may be wondering what “Help” is. Help, by definition, remains unnoticed, as long as it’s successful. It is not even considered as part of the product, yet it empowers customers to self-serve, increasing their likelihood to adopt the product and it reduces our costs to serve them.
The True Nature of Self-Service
Help, or Self-Service, covers all the UX flows a user can go through in order to be unblocked and be able to use the product again. It is a massive (and most of the time) unnoticed play for a business. Help spans from reactive Help - when the user asks for it, to proactive help - when we can reach out before the user does.
When self-service is done well, users won’t notice it and will go back to the core product. When it fails, it’s a disaster. Customers are frustrated and sometimes even disappointed by our core product. This ambivalence and the thin frontier between failure and success is exactly why I love my job. I love being part of the team that operates in the background and makes the experience seamless and natural. We’re building a product called Help. Yet, sometimes it feels like we’re creating illusions. How could we make Help more intuitive and frictionless than any company did before?
Our Growth Approach: The 3 Things We Learned From Our Users
When I started in this role, we kicked off our work by looking at top user needs. By analyzing historical support data, we found questions that could easily be solved: “how do I add a user?” or “how do I connect my inbox?” are setup and deflectable questions. We then ran further user testing to understand our users’ expectations. We asked them questions like “think about a time where you are blocked, where would you go?”. The qualitative interviews allowed us to unveil a few themes that needed to be confirmed or rejected by quantitative data. So we ran multiple, small iterations of Help in 2021. These small experiments helped us confirm or reject hypotheses we made.
At the end of this growth and experimentation led approach, we had a solid list of learnings:
- Customers struggle to articulate their questions in a clear and straightforward way
- Customers do not read articles or long paragraphs due to limited time, bandwidth or expertise. Sometimes, it’s actually a lack of self-confidence
- Customers like to have clear instructions, listed in a few steps as it’s easier to read, it’s more digestible and it’s actionable
Two Challenges: Parsing Users Questions and Matching Them to Answers
By running all these small experiments, we also uncovered two challenges:
- Parsing our users queries in order to understand what they were asking
- Matching their queries with the best self-service resources we had available
We found that when users open Help, they are most likely already frustrated. We had to ensure that our Help feature could not add any further frustration to their product experience.
To solve the parsing queries challenge, we decided to use Natural Language Processing. This branch of machine learning (ML) focuses on understanding texts and speeches, in the same way humans can. We used Dialogflow, a ML tool owned by Google which parses conversational queries. We successfully matched more than 15% of our queries after a few months of training the model.
When it comes to matching their queries, we had a huge amount of HubSpot content at our disposal. Yet, neither quantity nor quality means matching relevancy. We needed to match the exact right resource to our users’ questions. As we parsed queries through this machine learning model, we were able to know exactly what our users asked and we answered them with step-by-step instructions.
Our Design Approach: A Minimalist Conversational Search Experience
Help was redesigned with a focus on simplicity. We were very inclined to let the user lead the way. We wanted customers to ask us the questions they had, yet we wanted them to ask it in the way we needed to receive it.
We started to design the simplest entry point for Help, with a traditional search bar - intuitive, users know how to search. Yet, guidance was injected into the search experience, to make sure users would learn how to articulate their questions. When typing in search, the Help app would try to auto-complete their search by suggesting pre-populated questions. And even without typing anything in search, the search bar would suggest the most popular questions asked by other users. Pretty cool, right?
So what did we learn? Experimentation is our motto in Growth. We drive product changes by testing our hypotheses first. With this work, we learned that experimentation needs to be balanced: small iterative experiments should build up over time, to give space for a big swing. With Help, we accumulated learnings with these small iterations and then used all of them to go big. By redesigning Help entirely with this experimentation mindset, we were able to increase self-service rate by 22%.
We will continue to build the best and most simplistic Help application that users can use in order to answer any questions they have. And to be honest, we’ve only just scratched the surface.