Photo by Carlos Muza on Unsplash

Dialogue data is pure gold when building chatbots

Chatbots is a hot topic these days, but as we’ve developed the chatbot platform Kindly.ai ourselves, there are a few things that frequently remind us that it is still early days for the whole chatbot space.

We are currently involved in many projects with clients, but whenever we ask them during the initial phase if they have any dialogue data, this seems to come as a surprise. Data such as emails or chat logs can be pure gold when starting a chatbot project. It reveals the true nature of the interactions your customers have with your business today, and it can provide an accurate overview of what you should focus on in the beginning. We always advise our clients to start out by letting us do this analysis for them. In this regard, it would be very helpful if customers have already begun an internal process of surveying what data is available. Following this advice is something we feel will give your chatbot project the right direction, effectively narrowing down the scope of the first building phase of the project.

Our data scientist, Petter Hohle, performs data analyses for our customers.

Another advice is to never just hand out data like this without signing a legal document with your chosen chatbot provider, a so-called Data Processing Agreement. An NDA should also be signed before moving forward.

Don’t try to do too much in the initial phase. Start with a good Q&A bot, and then move on to valuable integrations with your own databases and various systems, which in our case would mean using the Kindly API. And then just release it after the Q&A phase! Get it out there to measure how it can improve over time. You can never anticipate all the numerous ways your users will interact with your bot, so getting it out in the wild is an important part of the learning process.

Companies need to start collecting their own chat DNA, as it will constitute the backbone of future chatbot solutions. It may take large companies some time to realise that they should build chatbots, but having a live chat is already running on overtime, as your competitors are already doing this to prepare for the future.

Also, don’t try to build a new Siri. That’s not the point of the project. The point is to make sure the bot answers the most relevant questions for your business, day and night, all year round. You are perfectly right to tell the user: ‘Sorry, we can’t help you with that’, when someone tries to order a gin and tonic delivered to their rooftop. Unless that is your business, of course.

If chatbots interest you, please check out our own chatbot (only in Norwegian) on our website or on our Facebook page. We might not be the perfect use case, being a fairly small organisation, but it’s nice to have automated customer support 24/7 on both our website and on Facebook Messenger. It handles the most relevant questions for us, such as ‘I would like to order a chatbot’. The most interesting part of this showcase is probably that one of our copywriters built it in just two weeks, using our platform. Another thing of interest is that we had no previous data when we built it, but we’ve improved it over time, after seeing what users have been asking it. A chatbot only gets better! The bot was released in July 2017.


Convertelligence is developing the Kindly.ai platform for building chatbots, with currently 17 employees working at our office in Oslo. Among our customers are DNB, Visma Group, OBOS, Olav Thon Group, SiO, Posten, MyBank, and many more.