Kata.ai Case Study: Delivering and extracting chatbot data to reap behavioral insights for AI
How Kata.ai processed over 9 million messages per day and developed better chatbot AI with Holistics, improving delivery of customer analytics and internal product analysis.
Holistics is a business intelligence tool that we use to track everything in a very agile environment.
Gelar Pradipta Utama
Head of Product, Kata.ai
About Kata.ai
Kata.ai is a conversational artificial intelligence (AI) solution for corporations in Indonesia. It uses Natural Language Processing technology to power virtual assistants across various industries, including fast-moving consumer goods, telecommunications, banking, insurance, and retail. It counts among its customers Unilever and Telkomsel, and partners with Accenture, LINE, and Microsoft, among others.
Businesses can customise the Kata Bot Platform according to their needs. This enables them to build their own chatbot on any messaging channel, using the Kata Bot Platform as the underlying technology, and reducing costs associated with developing AI assistants.
YesBoss Group launched Kata.ai in 2016 as a pivot from the former’s human-assisted virtual concierge service.
The Problem
Managing messy conversational and behavioral data
As of 2017, around 143.3 million people or half of the population use the Internet in Indonesia. Many use instant messaging platforms, such as WhatsApp (40 percent of Internet users), LINE (33 percent), BlackBerry Messenger (28 percent), and more. Indonesians use these channels to communicate not only with their friends, but also with businesses.
To cope with massive amounts of queries, complaints, and requests from customers via messaging platforms, businesses need to scale their customer service operations. AI-powered chatbots help marketing and support teams to sort through customers’ messages, categorise them, and respond. In fact, the largest of Kata.ai’s 25 enterprise clients has around 300,000 daily active users who send roughly 800,000 messages per day. All in all, Kata.ai processes around 9 million messages per day.
“Conversational data is messy,” says Afandi, data analyst at Kata.ai. The company aims to build its own analytics pipeline to make sense of the language-based data and further train their chatbot to analyze and understand conversations. For example, they need to identify a customer’s problem or request based on the words they use. They also need to detect the words and phrases the chatbot does not understand so they can train it to analyze such data.
They need to do this for every company and industry, so that their clients would understand their own customers’ behavioral data and find ways to serve them better. Kata.ai also needed a way to regularly transform these data insights into visual reports that even non-technical people would be able to understand.
Holistics gives a very powerful dashboard… we use schedules to push [dashboards] to Slack every morning.
– “Yodee” Ismail Usuluddin, Product Growth Lead, Kata.ai
The Solution
Granular access from micro raw data to macro-level analysis
Initially, Kata.ai’s team tried using open source platforms like Zeppelin, Superset and Metabase, but realised they wanted a platform backed by a strong technical support team who could help them out when necessary. “Some friends who already used Holistics told us it is a good tool for analytics,” shares Afandi. “We decided to start using it because it’s easier to connect with data.”
Kata.ai’s data engineer connects Holistics to their messaging platforms’ database to extract conversational data. From there, they build metrics and reports for Kata.ai’s clients. These reports provide information such as the gender, age, and location of customers who are conversing with the company via messaging platforms. But they also provide more complex analysis and insights into customer behavior, such as their most common pain points.
Internally, Kata.ai uses Holistics to analyze and extract conversational data to train their chatbot AI assistant. This means teaching the bot to recognise common phrases like greetings, and to extrapolate the meaning of the messages based on sentence patterns and word usage. The more data it processes, the more the chatbot learns new words and phrases. For example, it will be able to distinguish complaints from e-commerce inquiries and transactions.
For example, what should they do if they notice 1,000 people saying ‘milkshake’ in their chats? Such data can point out trends to clients, alerting them of high demand for a product. On the other hand, if a large number of people send messages containing the same complaint, that may alert the client of a bug in their app.
The team also uses Holistics to gain a macro view of data, tracking things like the number of active users in a day, number of sessions, number of incoming messages, and the days of the week when usage of the chatbot peaks. Gelar Pradipta Utama, Head of Product at Kata.ai, notes that they use Holistics to understand product usage behavior in more detail, and to track product-based key performance indicators (KPIs) to know how they can further improve their chatbot product. For example, by analyzing conversations and user sentiment, they can decide that their bot needs to learn how to make small talk.
Aside from analyzing conversations and training their chatbot, Kata.ai also uses Holistics to track and measure marketing campaigns. “Yodee” Ismail Usuluddin, Product Growth Lead at Kata.ai, uses Holistics to track reports on active users by channel, number of clicks, new friends on social media pages, and more. This data helps the marketing team decide which types of promotional messages to send to certain customer segments, and assess the return on investment of marketing campaigns.
Holistics kind of solves every problem we have at the moment!
– Gelar Pradipta Utama, Head of Product, Kata.ai
Killer Features
How Kata.ai uses Holistics
Holistics helps Kata.ai parse through conversational data to extract relevant subsets for training their chatbot. They can also categorise words based on intent—for example, greetings and requests.
Kata.ai’s enterprise clients have access to a Holistics dashboard, which means they can access their conversational data and carry out further analysis if they wish. Kata.ai toggles the permissions feature in Holistics, as well as the filters, to make sure each client sees only their own data, and not that of other companies.
Kata.ai connects Holistics to Slack to automatically send out charts and reports to alert the team of important numbers to constantly monitor, such as active users. The team can also push data to Google Sheets. The delivery capabilities of Holistics, to automatically send data and reports to customers, are crucial features for the Kata.ai team.
Future Of Data At Kata.ai
The beginning of AI product development
With a complex product and the need to analyze massive volumes of data per hour, Kata.ai sees the need to grow its data team and build its own analytics pipeline. As they construct their pipeline, they’ve found Holistics to be a suitable solution that helps them make sense of conversational and behavioral data.
Yodee notes that the chatbot is only the company’s entry to AI product development. They aim to develop voice recognition and text summarization products in the future.
The company also plans to have a data science division, with data scientists and analysts available to help out teams that use Holistics.
Using Holistics
Tips & tricks from the Kata.ai team
- First of all, you need to know what kind of metrics you need. If not, conduct experiments to find out which metrics matter to your company and to your clients.
- Give your enterprise clients more visibility into their data by giving them a Holistics account. Use the Access Permissions feature and filters to ensure your clients only see the data their company should.
- Connect Holistics to Slack to automatically push reports and dashboards to your team channels.
- If it’s necessary to build your own internal data pipeline in the long term, use a business intelligence tool like Holistics to flexibly operate across your existing pipeline without locking you into a specific technology, which lets you swap out and connect new parts of your evolving data infrastructure.