Chatbot migration. New channels, integrations, GenAI

The main challenge in changing technology is the migration and transfer of logic from the ‘old’ to the ‘new’ chatbot. Arguably, these are different tools and different technologies, which means that not everything can be transferred automatically applying import/export rules. Partly or even in a large part of the project these will constitute manual work. How do you carry out a project so that it has the least possible impact on the business and staff? How to reduce its cost and time?

The article brings together Comtrust’s experience in AI technology implementations and changes, transferring chatbot processes and logic, launching new channels, voicebots and complementing existing AI technologies with new ones.

The project of chatbot migration to a new technology can be carried out in several steps:

• Step 1: Chatbot analysis: workflow, scripts, intentions, languages, channels, integrations, etc. It is necessary to assess how relevant the conversation history is, which processes and cases are worth transferring to the new chatbot, what to improve and which ones are pointless to transfer. If we want to save time and costs, it is at this initial stage that unnecessary schemes must be abandoned, and not only after the transition. However, it is usually the case that the existing chatbot is well-trained and needs to be transferred 1:1.
• Step 2: If you want to implement changes or new features in the new chatbot then create a roadmap with the improvements. Do you need integration? New channel launches? Perhaps the chatbot should also be a voicebot? In addition to traditional channels, do you need, for example, a chatbot on Whatsapp? Is the chatbot supposed to trigger functions in business systems? What range of information and data will be available to customers and what to employees or partners? etc.
• Step 3: Defining indicators and milestones to track the performance of the new chatbot and KPIs for the new technology and new features and monitoring the migration. Probably, chatbot responses to customer enquiries in the new and old technology will be generated at similar times, the differences perhaps imperceptible to the user’s eye, but only visible in the code, to the administrator or analyst. On the other hand, the correctness and quality of the answers will be important here. This is an area that needs special attention.
• Step 4: Time and budget estimates. Remember – with a new technology comes new costs, and for a while the costs are duplicated (you need both technologies for the testing period). Time, of course, depends on how complex processes the old chatbot handles and how much of the logic has to be transferred manually.
• Step 5: Once the chatbot content has been transferred, duplicated, transfers and integrations need to be transferred. Complications can arise from firewall settings that may block traffic from the internal network to the bot, which will require new settings. Finally, it is worth checking the so-called ‘dead ends’, i.e. places left in the scripts that do not trigger the next step.
• Step 6: Final testing and fine-tuning of the new bot.

The main goal is to get all the AI logic onto the new platform and start optimising the new bot and adding new features and running additional channels.

What does the new technology give us that the old technology did not?

AI bots have been implemented for several years, but even those implemented more recently may already be obsolete. Did the ‘old’ technology provide the ability to integrate securely, handle processes or was it just for searching the knowledge base and providing simple answers. It is difficult to say how ‘old’ a technology you are starting from. However, what you can confidently expect from the new technology is definitely more capabilities, the most important of which is generative artificial intelligence (GenAI).

Why change the technology, if only for the sake of generative AI?

GenAI creates new data, such as text, images, music and video. It creates new content by using algorithms and machine learning models. GenAI in chatbots should be used as a complement to standard NLP algorithms. That is, if there is no answer to a question in the script, then GenAI steps in and searches the knowledge bases. Why only then and not immediately? Because every GenAI query costs money. The cost for GenAI to process and respond is more expensive than using an NLP algorithm. Therefore, it is best to design as many conversational processes as possible ‘rigidly’, with GenAI as a supplement.

Ecology at the end

By reducing the number of GenAI queries you leave a smaller carbon footprint 😊

Chatbot Technology Audit

The chatbot migration project is not heavily complicated, but it can be time-consuming. It’s not like in the past when you started your chatbot adventure. Then you were starting from a completely different point. You probably weren’t sure what to expect, what you would need and what users would expect. You didn’t have statistics or a transcript of all the conversations. You started with traditional chat conversations, with initial assumptions, repetition. Then, after the chatbot had been running for a few months and you had probably reworked the initial script many times, you noticed that the bot had started to adapt to the customers’ requirements. It picked up keywords and matched the response to them. But it could still be wrong and give wrong answers. Once you have trained the old bot well, it turns out that there are new technological opportunities to take advantage of. And it would also be a shame to undo previous work. One does not exclude the other.

Want to find out how new AI technology will improve processes? What will be the benefits and costs? Is it worth it? We invite you to take advantage of the Chatbot Technology Audit service.