Raising a voicebot

Anyone in the need of a voicebot would prefer to go to a shop and purchase a ready-made, wise bot, which has the ability to both take calls from customers and make calls. In reality, however, it is a bit more complicated than that, since each bot is different and is dedicated to its different tasks. Before our bot can take a place “on the handset” of the contact center, it will first have to be “brought up”.

Contrary to what professional investors often like to hear, in most cases it is not the case that a made voice bot can be sold ten, twenty or even a hundred times. Although it would certainly be a great business model. However, we find such bots much more often in optimistic investment plans than in reality.

Even if we consider two bots to make appointments or carry out a satisfaction surveys, they will never be identical. For example, one of the bots arranges visits to a selected facility, and the other, through a completely different set of survey questions, arranges visits to a network of facilities (it must find out in what facility the visit will take place). The bot can also cover a completely different way of data exchange with either a calendar or CRM systems.

I recently wrote about what a voicebot is, and today I want to describe the living process of such a bot. How it is born, how it grows and learns, and finally, how it becomes an intelligent adult bot, which can professionally work in a contact center. Of course, every bot manufacturer can have an individual approach to this process. I describe a process that works quite well in practice.

Planning the birth of a bot

It all starts from the idea to create a bot that will operate on specific processes in our company. The next step will be to define the purpose of the bot’s life. Usually it is clear – we aim to close cases in either more or less time. Perhaps we want to receive completed satisfaction questionnaires in the form of a cyclical report, scheduled visits to the calendar, or information in the CRM that a given customer will pay the outstanding invoice this week.

In short, we want to find a simple solution for a repeatable process in our contact center. A ready-made script that we already use may often appear here, but it may also be a completely new process for which we do not yet have a script. Whether the script is present or just a general idea with clear objectives, we can begin the creation of a bot.

An extremely important factor at this stage is the choice of the technology provider. We have to decide whether we choose an on-premise, cloud-based recognition and speech synthesis systems, or whether we prefer to work with a reader. We need to define the process of integration with our internal IT systems and of processing and importing data. This is a stage that requires many important technological decisions.

Once we have made our choice, the bot producer will sit down and prepare the first version of the dialogue. This version is subject to extensive internal testing so that the bot is ready for first contact with the outside world.

Growing up and education

Using the metaphor of adolescence, one can say that the bot’s school age stage is his first independent attempt to hold a conversation. The bot is prepared by its creators, tested and potentially ready to live independently, although it still needs support and education.
We can start talking to him in tests. It is worth seeing how it works, and therefore how we will ultimately be represented by the bot to our clients.

During a conversation with a bot, it is important to keep a close eye on him, pay attention to how he works, whether he is mistaken, how he is doing, what his shortcomings are. He needs to be constantly trained, both in the beginning stage as well as the further ones to come! Once we, the contracting authority and the creators of the bot, are satisfied with his work, it can be said that the stage of his official education is over, meaning he is reaching maturity. But he is not yet a professional. He still has to pass the areal test of maturity which will be in the form of verifying skills on the labour market.

The final exam

The finished bot should be tested in action, i.e. allowed to talk to real customers. As much as possible, it is best to limit the number of interactions. You can either pass a line with the bot every 10 callers, for example, for outbound processes load a limited number of numbers to call into the bot.

At this stage, urgent monitoring of calls and correction of any problems that arise is essential. This is a key moment in the bot’s education. The first encounter of his gained “school” knowledge with the professional world. Nothing can replace interactions with real customers. Unfortunately, joint internal tests will never reflect the way people talk. Therefore, the tests should only be regarded as a preliminary check to see whether the bot is surely talking the way we want it to and is not wrong in any obvious places. As a customer service team, we know the most frequently occuring problems in this process, but knowing the potential course of the dialogue, we can manage it in a correct way. Only the first meeting with a real client will be crucial for the bot’s education.

It is also a great time to initially measure our assumed bot performance indicators, such as call-back time, duration of the conversation, percentage of correctly handled cases and so on. Here we can see whether the performance of our bot meets our expectations “on a living organism”.

Adulthood and professionalism of a voicebot

Once we see how the bot talks to real-life customers and we train it on the basis of preliminary talks, we can think about the actual implementation of the so-called “production”, i.e. 100% launch of the bot. In case of bigger changes or if we are facing a really huge volume of interlocutors, you can still repeat limited tests or start the bot on gradually increasing traffic.

Either way, it is necessary to maintain close monitoring of problematic talks, especially during the first days and weeks of operation of the bot. The learning process should never end. Over time, people will start talking to our bot a little differently, because they will get used to this kind of interaction with our brand. That’s why the bot needs to be periodically checked and further developed in areas where it can’t perform perfectly, increasing its communication capabilities.

At this point we can say that our bot is already fully professional in customer service, within the chosen process. And like a good professional, he should learn at all times. We can safely assume that the implementation of a well an employee on the handset. However, very important differences should be kept in mind. While a bot will never reach the level of human creativity, the process of its creation and training is, contrary to appearances, much faster, itst scalability of the whole process will increase with time, and the volume of threads handled will be incomparably higher.

Therefore, I recommend that we approach this type of investment with a large dose of optimistic realism. The objectives we set for the bot must not exceed its capacity, and the first implementation must be carried out gradually. We should also remember to verify the work and its effects at every stage. It is important to carefully select the technology supplier, and the investment can bring a return in a very short time.

Piotr Kempa (Head of AI) – the creator of Primebot – an intelligent conversation bot, which conducts a non-linear dialogue and serves customer in repetitive processes.