If you’re reading this article, then you might be looking for a better and faster way to train your chatbot.
Look no further - say hello to Bot Improvements!
Traditionally, training a chatbot has involved the manual process of sorting, creating and adding samples of “training data” for the bot to learn from. This process is similar across most AI chatbot platforms and requires time, effort and scrutiny.
So if you’re responsible for building or maintaining a chatbot, it probably means you’ve read through large volumes of customer chat logs, identified keywords and phrases, and created many different replies and dialogues, based on your customers input and FAQs.
All fun and games, right?
To simplify and speed up this ongoing training process we partnered with Innovation Norway, Norway’s state-owned innovation hub and development bank, along with a select number of our dedicated customers.
Our mission was to develop a set of automated tools using Machine Learning to analyze all chatbot conversations for you. The result is a highly advanced system that automatically suggests improvements for your chatbot conversations!
We’re calling this system Bot Improvements, and it’s being launched as part of the Kindly chatbot platform. This feature will be available for all existing and new customers.
Why is this good for you?
What Bot Improvements does for you is to significantly reduce the manual labor and time spent on the training process. It also contributes towards increasing the precision and relevance of replies to customer enquiries.
This is functional AI at it’s best.
Even sophisticated chatbots can still improve
As mentioned, the task of going through archived customer conversations one by one is tedious and time consuming, but it is essential for improving your chatbot’s overall quality.
Bot Improvements will help you maintain your chatbot and keep it up to speed. We’ve built an automated solution to save your chatbot teams even more time, so that they have more time to focus on the bigger picture!
In short, it taps into the advanced machine learning capabilities of the Kindly platform and uses the technology to support maintenance tasks for the chatbot’s upkeep.
There are three key areas where Bot Improvements can further enhance chatbot maintenance. Here’s a brief rundown of these added benefits.
Benefit 1: Sample Candidates
Chatbot maintenance consists mainly of a manual review of all archived conversations. The findings from those logged chats are then used to make manual improvements to the bot itself.
However, a better way to do this work is by using the Sample Candidates within Bot Improvements. With Sample Candidates, the machine learning technology goes through the conversational archives for you. This way, you can use the data from that report to quickly make changes or iterations to how the chatbot is programmed, saving an invaluable amount of time and effort for your team.
“Our custom made machine learning model runs through all conversations in your inbox, and comes up with suggestions for new training data to add to a given dialogue. No more need to spend hours going through endless conversations.” -Gjermund Norderhaug, Product Owner at Kindly
- Sample Candidates automatically suggest “samples” (bot training data) to add to dialogues, improving dialogue precision
Benefit 2: Dialogue Confusion
(Coming, will be available H1, 2022)
As your chatbot grows, it goes through “growing pains” as it becomes bigger and has more conversations with shoppers. This can result in two or more dialogues that conflict with each other, which can confuse the chatbot in terms of which dialogue it should present to customers asking it questions.
To avoid this, you can use our Dialogue Confusion feature. Kindly’s machine learning will be able to automatically tell you, "Hey, I have found two conflicting dialogues, and I don't quite understand which one you want me to use."
You can now fix this easily and choose which dialogue is the most appropriate for your needs. This is an easier way to take the necessary course of action that improves the chatbot for your shoppers.
- Automatic identification of similar dialogues makes it easy to edit them directly, improving dialogue quality
Benefit 3: Incomplete Dialogues
In order to improve interactions with shoppers to the best of its ability and evolve, your chatbot needs exposure to enough customer experiences to learn how to have those conversations.
If your sample size is too small, it’s hard for the chatbot to gain that level of understanding to become the sophisticated product it can be for your brand.
Our Incomplete Dialogues feature helps address this challenge.
- We recommend minimum 20 samples per dialogue. The platform automatically shows which ones need more data
Incomplete Dialogues is a task-list for the responses that require more samples and data, in order to meet our recommended number of interactions for the machine learning technology to sufficiently learn from. Simply put, these are dialogues that need a little more love and maintenance.
“Incomplete Dialogues is a list of dialogues that need more love. These dialogues need more samples in order to be good and strong bot dialogues. You can find it under your bot -> Build -> Bot Improvements” -Gjermund Norderhaug, Product Owner at Kindly
Why Kindly AI chatbots are better than traditional bots
Based on our own research, few companies have the same level of automated suggestions for training data as Kindly does. This makes us the right option for future-oriented companies and bot builders looking to improve their everyday work-life.
✅ For managers: your team will increase efficiency and reduce costs
✅ For bot-builders: you will save time, effort and reduce repetitive tasks
✅ For your customers: they will interact with a better and more precise chatbot
We believe chatbot training is a core part of how our products and services are going to evolve, so that the training process is done in an efficient and effective manner.
Better conversations with your customers is good for your company - learn more about Kindly’s platform, chatbots and conversion tools.