Chatbot migration

Afraid of chatbot migration – don’t be

This year we have had numerous discussions with customers who feel trapped by their current technology and are considering a change, chatbot migration. Technology that felt fresh and state-of-the-art a few years ago might feel a bit clumsy and outdated today. Does this big decision to change technology mean that you’d be throwing all your investments down the drain and returning to square one?

Understanding the structure is half of the journey already
When people start their bot journey, they often do not know what to expect, what is needed, and what the end-users are asking from them. It is not easy to build comprehensive statistics or analyze your chat logs unless you have great NLP tools and transcripts of all your calls. Therefore, you need to start with some initial guess and develop and iterate.

After some months and probably quite a significant rework of your initial estimate, you have reached a state where the bot starts to be in line with the customer demand. Then, of course, the question is that is your bot high quality enough to understand the semantics of the customer questions – or is it just picking up some trigger words that may lead to a significant amount of so-called false positives. For instance, instead of answering the “I am a bit tight financially at the moment and cannot pay my invoice, what should I do” – question, the bot recognizes the words “pay my invoice” and happily provides instructions on how to perform a payment smoothly.

One-click migration
Anyway, back to the original topic. Once you know what the questions from your end-users are, and you have decided the correct answers (which is not always so trivial, either), you are actually in the halfway milestone already. Changing technology is not that scary anymore after that. Today, almost all bot vendors provide a JSON export and import for the bot training, intents, and answers. If you do not know what JSON is, you do not even need to. It means that you can migrate all the work you have done to the new platform with one click and start optimizing the new bot.

Finetuning the new bot
OK, does the new bot work with the old training data then? No, unfortunately, it does not, but it is a good start. You have now identified the needed training and structure of the customer dialogues. You have also transferred everything to the new platform. This is going to save you quite a lot of typing work. Of course, you probably need to shuffle the structure around a bit. You also need to add or remove or clean some of the training data (it might be a good idea with the old platform as well), and finally, you need to optimize the AI model to finetune it to take the best out of new technology.

Sometimes it is better to start from a clean table
Alternatively, if your current model is not at all compatible with the new technology, you can feed your chatlogs to an analyzer or run an Automator to analyze the structure. You may also scan your website with Automator tools to create the starting point. Even if you decided to start from a clean table, what you have gained already is an understanding of your customers’ needs. You also know what is missing from your current tool. That is important as well when you are choosing your to-be tool.

Some clients did it already. Are you next in a row?
We have made this journey already with several customers, to mention Kela and S-Pankki. And I can tell you that the project is pretty different from the first-timer project. The challenges you meet and the requirements the clients have are entirely on a different level. This is telling me that customer service automation is a journey where you learn with every step in and the beginning of the path you do not even see what is around the next corner. People say about financial investments that the best time to start was yesterday. The second best is today. The same applies to the conversational AI journey. If you have not started yet, start now. If you were fortunate enough to start yesterday, consider if you are still running the best tools – or if you should consider migrating to a better one.

Got interested in? Feel free to contact us info@front.ai or book a meeting.

For more information, visit our Conversational AI services.