Zenoti Success Story
As your business grows, customer support is often one of the first areas to feel the strain. The more customers you have, the more support tickets you get, the more support agents you need to hire to handle them. Skimp out and you risk alienating your customers with long first response times and ultimately – poor customer experiences.
There is a smarter way. Modern businesses are focusing on making their services more efficient by investing in CX analytics and self-service help solutions such as chatbots, knowledge-bases, product tours and other initiatives.
One of our customers, Zenoti, is serious about self-service help and recently managed to achieve a 25% ticket deflection rate with the combination of conversation analytics and chatbots.
I caught up with Sagar Garuda, Zenoti’s Director of Customer Education to learn more about their process. Below is our interview – enjoy!
Hi Sagar, can you tell us a little about Zenoti?
Sagar: Zenoti is the best-in-class cloud-based solution for spas, salons and medspas. We power 80% of the world’s enterprise spa, salon and medspa market with our all-in-one software, designed specifically for these industries.
Our founders knew the stresses of running spas, salons and fitness centers without comprehensive software to help, so they leveraged their professional software development backgrounds to create a solution in 2010. Zenoti covers online appointment bookings, POS, CRM, smart marketing, employee and inventory management, and much more.
Today we’re truly global, serving over 1,000 spas, salons and medspas in North America, Europe, the Middle East, India, Southeast Asia, Australia and Africa, enabling them to focus on doing what they love.
What is your focus right now?
Sagar: As the Director of Customer Education, I lead various teams and initiatives that help with efficient onboarding, contain support tickets by providing just-in-time help using bots and overlay help, and increase product awareness and feature adoption via product tours.
Zenoti is growing at a rapid pace, and like any SaaS business, we want to scale efficiently and sustainably. Human support agents are an essential part of our customer support strategy, but we know we can improve our efficiency by investing in self-help.
What is your strategy for containing support volumes?
Sagar: We use Intercom as our primary support channel, so we wanted to deploy their new Resolution Bots to automatically respond to common user questions and free up support agents to focus on tickets that need in-depth analysis or troubleshooting.
To train bots effectively, we needed to build a deep understanding of commonly asked questions and the associated terminology.
I’m a strong believer in, and advocate for analytics-driven content, so I wanted our initiatives to be informed and backed by data. We already had all the data in Intercom. We just needed an efficient way of identifying all the “how-to” type questions.
What was the first step you took to uncover commonly asked questions?
Sagar: We tried manually tagging conversations in Intercom to uncover common user questions, but we couldn’t get insights with the right level of detail to build effective bots. That’s when we started looking for text analytics solutions and discovered Prodsight.
We hooked up Intercom and Prodsight to analyze our historical and incoming Intercom conversations. Then we created a topic in Prodsight to collate “how-to” questions in one place. We specified keywords such as “how do I”, “way to”, “can I” and other similar terminology to tag and gather any historical and future mentions automatically. Based on that, Prodsight produced a real-time report with all “how-to” questions and volume and trends charts to help us gauge the big picture.
By looking at the trend chart, we realized that at any point in time, approximately 25% of our support conversations contain “how-to” questions. This helped us set our expectations around what proportion of conversations can potentially be resolved by bots.
Once we laid this foundation, we then had to break down “how-to” conversations into the specific product areas they were relating to.
How did you go about segmenting “how-to” questions by product area?
Sagar: Once we isolated all the “how-to” questions, we created subtopics for each major product area such as “inventory” and “membership”. The subtopics instantly uncovered all convergences of “how-to” intent and terminology associated with different features.
Once we completed this exercise, we ended up with a list of questions broken down by product area and prioritized by volume. This was crucial in helping us decide which areas to start building bots for.
How did you translate questions into bots?
Sagar: Intercom requires us to provide at least 10 example questions to help train each Resolution Bot to correctly match questions with answers.
This was particularly challenging because of the differences in terminology our customers use across geographic markets. For example, when enquiring about our checkout feature, US-based customers would say “ring up the customer” while in Australia, it’s “checkout”. It’s almost impossible for someone to think of all of these variations.
We used this iterative process to build each bot response:
- On Prodsight, zoom into each subtopic to isolate FAQs in each product area
- Read the user questions to deeply understand the terminology used by users
- Capture illustrative question examples and use them as Intercom Resolution Bot training examples
- Once satisfied with the configuration, launch the Resolution Bot
- Refine the triggering criteria as needed to reach desired performance
We repeat this process for each commonly asked question.
Sounds like a neat process! Do you measure how your bots are doing?
Sagar: Yes, we do track the performance of each bot and their combined performance.
One of the most important metrics is the resolution rate. It represents the proportion of tickets that are solved by the bot without requiring the support agent to intervene.
For example, the most popular questions come from users forgetting their username or password. While we have self-service options for recovering those, many users still contact support. Resolution Bots now automatically solve nearly 60% of such requests each week.
Another key metric is the coverage rate which is the proportion of tickets that the bot triggers on. We are always looking to increase coverage by reviewing incoming questions and bot triggering rules.
What is the impact of this project so far?
Sagar: We have now deployed around 450 bots to automatically handle various common customer queries. Some bots are performing better than others but we managed to reach an average resolution rate of 25%. This means that one out of four tickets the bot triggers is automatically resolved.
This is huge because it not only gives us the productivity of one support agent, but most importantly it offers our customers instant assistance.
That’s an amazing result! What’s next for this project?
Sagar: We are working to increase our coverage to automate even more support conversations.
Our goal now is to triple the productivity of our chatbots.
To achieve that, my team is continuously monitoring frequently asked questions on Prodsight and we’re using these insights to inform various self-help initiatives – bots being one of the key pillars.
We are also looking beyond bots and investing in other customer experience initiatives such as product improvements, product tours and contextual pop-ups, building our knowledge base. We are particularly proud of Zenoti University, which provides Zenoti customers with online learning courses tailor-made for specific roles, from front desk staff to management.
Amazing! Thanks so much for sharing your process and insights with us, Sagar!
It’s a great blueprint for all CX leaders and practitioners on how to drive efficiencies and improve the customer experience at the same time by leveraging existing data and modern tools like Intercom and Prodsight.
I wish you and your team all the best in continuing to hit new records with your customer education initiatives.