The year for chatbot industrialisation
The word Chatbot has been trending since 2016, when Facebook launched their messenger bots, and since then we’ve seen an increased interest in developing this new way of interaction with the technology.
The year 2017 has been the year were most companies wanted to use chatbots for automating processes. During the second Paris City AI event, we saw some examples such as a chatbot by France’s National Lottery games helping people to bet or the chatbot that assists people using an internal company FAQ. Throughout the year numerous proof of concept for chatbots projects were tested in companies to test if chatbots can make a big step in the way we interact with AI. Let’s see what are the outcomes.
2017 learnings
During Chatbot Paris meetup on May 2018, Javier Gonzalez, Co Founder of Botfuel, presented the main highlights of chatbot projets during 2017, highlighting the following outcomes:
- Chatbots are adapted for a single scenario, which is the one that they have been designed for. Outside of that, they would hardly generalise to different scenarios.
- The universal bot is not ready yet. Instead of wanting a chatbot that can do several tasks, you should think about several chatbots where each one is specialised in a single task.
Some deceiving experiences come from the fact that people don’t realise that bots are not humans, therefore, you can’t talk to them yet the same way you would talk to a friend or colleague. The chatbot logic is different, they expect a specific word that can trigger the sequence of the conversation and do what they are meant to.
Another main concern is security. The first tests show that people haven’t been too worried about the kind of data that chatbots can collect and how that would integrate with actual standards like the GDPR and how it can be safely stored.
The rise of chatbots in 2018
Pushing the development and adoption of chatbots further in 2018, the following aspects will play a major role:
Involve more concerned actors in the project: Usually the proof of concept involved only the marketing and innovation teams. This is important to bring the new ideas to life, but now more stakeholders like technical and field people should be involved in order to shape the skills of the chabot.
Include IT: The IT department should be an important part of the project since they define the architecture where the bot will run and also how it is going to be connected with the end users. This is a fundamental piece, if the project needs to be used on a larger scale in a company.
More consistent budgets: During the first proof of concepts, the allowed budget for a chatbot project was between 10K to 50K Euro. In 2018, budgets are more consequent (50-X00K €) which is going to boost the resource allocation and also expand the scope of the projects.
More realistic call for proposal: Nowadays, calls for proposal are defined precisely since chatbots specs have been generalised so people are more mature when they demand a chatbot. They won’t ask for all-in-one bot anymore.
New adapted profiles: Chatbot development needs people familiar with chatbots subjects such as natural language processing and natural language understanding. In addition, software developers are also needed for the implementation of the technical aspects.
Technical requirements have evolved: Chatbots collect information about the user in order to give a personalised answer and to have some context about the conversation. This implies a concern about authentication and the delay that a text content remains alive in a conversation.
Chatbot development 2.0
Typical chatbot projects include the following components:
- Different connectors that can be adapted to the message system of the company.
- A supervision interface for a chatbot, where the logged interaction can be seen in order to improve the bot.
- The ability to connect the chatbot to company data such as a data lake or the data issued from a CRM software.
- Analytics for business intelligence strategies.
- Redirection of the conversation if the chabot doesn’t know the answer so that a human can continue the conversation or the chatbot would give hints about a possible answer.
The development cycle is composed by the following phases:
- Define a use case of the chatbot with end-users and technical staff.
- Format input data that is going to be connected to the bot such as an FAQ or data from a data lake.
- Prepare the technical integration of the chatbot with the IT staff so that the production environment details are validated before the chabot development starts.
- Development of the chatbot, with regular points of progression showing examples of the bot.
- Launch strategy, which involves communication strategy around launch and test of the chatbot.
- Put the bot in a internal test environment to configure the indicators of the bot that are going to be supervised during testing.
- First measures and adjustments looking for optimization. This phase can lead to a maintenance phase of the chatbot.
Bottom line
Nowadays we have realised the capabilities of a chatbot, the fact that they are good only for the things that they have been programmed for and the fact that they don’t always generalise well. This leads to creation of single bots which can do one thing good and leading future development to a chatbot management or chatbot market place such as the one created by Amazon or by Google.
Get involved with the community
If you are interested in chatbots, want to know how to build one or if you are generally interested in applied AI, reach out 🙂. The City AI Paris community is open to people that are interested in subjects around AI. We also organise events
