We have prepared a complete guide to guide you in the success of your chatbot project, from A to Z. Follow the guide!
My name is Thomas and I am the co-founder of Vizir.co . We have developed a solution for creating chatbots to automate business processes.
I wrote this article as a feedback to help future DSI, HRD, project managers, etc… to find their way in the creation of their chatbots projects.
As in any IT project (solution, website, mobile applications, etc.), the first and most important step is to fully understand what chatbots are used for and in which cases they have an interest, and in which cases it is better get by.
In another blog post, I offered a simple definition of chatbots , which I will remind you here:
A chatbot is a computer program accessible from an instant messaging interface.
This computer program can be used for 3 main categories of use cases:
Their goal is to automate business processes that involve a customer or employee and a business. In B2C (customer), there are use cases for customer service or transaction automation ( Bruno , OuiSNCF ). In B2B, this involves automating business processes (document creation, purchasing process) or also user support (help desk).
Their goal is to assist you in your daily life. They now take the form of voice-controlled robots present on your smartphone (Siri, Google Assistant) or on connected speakers (Google Home , Alexa , Apple HomePod , etc.). They respond to simple requests like: “OK Google, turn on the light”, “Alexa, add a pack of milk to my shopping list”.
Their goal is to allow you to access content from instant messaging. They are present via email ( Flint ) and especially via Messenger ( Jam , Techcrunch , Maroon5 ,…).
We will focus here on the 1st use case: business chatbots .
In this context, the chatbot is useful if the following conditions are met:
The use case requires a conversation between the user and the company,
These conversations are currently managed manually (telephone, email) or not managed at all (new process).
The number of conversations is high
Here are the advantages of the chatbot compared to other formats (software, web, mobile):
Access the program through a transparent and natural interface (instant messaging)
Removing friction during use: exchange in natural language
Collect structured data
Before embarking on a chatbot project, you need to understand how they work. You will then know what resources you will need to mobilize.
Any chatbot is made up of 3 elements that combine:
These are algorithms capable of transforming a request expressed in natural language (voice or text) into computer code.
The objective of these algorithms is to detect in the sentence:
The NLU will understand that the user intends to create a new contract.
And that this contract must be carried out for the Renault customer.
For your NLU to be able to detect these elements you will have to indicate to it the intentions and the entities that it must understand.
So, and this is important: you must configure the topics that your chatbot must be able to understand.
So you need to make a list of topics that the chatbot needs to understand, otherwise it won’t understand them, it’s as simple as that.
Here’s how to build that list:
Often, our customers who automate existing processes have access to a list of requests made by their customers or their employees (eg calls to the call center, etc.), which simplifies their work.
Once you have your list of topics to cover, you need to translate them into combinations of intents and entities.
It is then necessary to configure these intentions and entities in the NLU tool.
Finally, you will need to fill in 10 to 15 user requests for each entity intent couple.
These queries allow the NLU to calculate the intent and entity probabilities in other queries formulated differently by other users, in production. That’s the beauty of machine learning.
Setting the NLU is only the first step in the conversation. Then you have to add logic.
Indeed, as in a discussion between humans, the fact of understanding each other is not enough to have an intelligible conversation. Conversational logic is needed.
This is what we will see here.
Once the request is typed by the user, the chatbot performs an action that corresponds to the logic given to it.
This logic can be handled in two ways:
Today, in the majority of cases, option 1 is implemented. Several tools can be used to create these conditional trees (Excel, mind map, etc…).
Option 2 is at the R&D stage at Vizir, and requires already available and well-structured conversation data.
The chatbot that automates processes has (normally, otherwise badly) predefined goals.
Examples:
These actions are not strictly speaking in the chatbot. They call on external services through connectors called APIs.
These APIs make it possible, for example, to send data into an algorithm to make it work and then retrieve the result.
Well done for making it this far!!
If you’re here, it’s because you really have a chatbot project in mind ( don’t you want us to call 30 minutes by chance? )
So we’re going to get down to business: the steps to create your chatbot.
This question is closely related to the type of solution for which you will opt.
There are three types of players in the chatbot market:
So-called “core” technologies: these are chatbot development frameworks, mainly based on the NLU brick.
Off-the-shelf technologies: these are products, often SaaS, which integrate the entire value chain of the creation of a chatbot, and which do not require specific development
Agencies and consulting firms: they will use technology 1 or 2 and will support you on the project.
Here is a summary of the internal and external resources to deploy in each of the three cases.
I have detailed each of the solutions just below, be sure to read them.
These are development frameworks based on the NLU brick.
Here is a summary of what you will need to do:
According to our experience, here are the human resources you will need to carry out a classic chatbot project:
A chatbot made on such a framework takes an average of 4 months of development before going into production.
After going into production, you will have to keep the whole team full-time to carry out the modifications and optimizations of the chatbot. It will be necessary to see these recruitments as an R&D investment because to be effective, the team will have to keep informed and test the new technologies which are constantly arriving on the market.
You will still have to pay the product costs if you do not opt for an open source solution ( rasa.ai and snips.ai ). These costs remain low. At Luis.ai (Microsoft), the solution costs €1,265 for 1000 requests. A conversation has an average of 15 requests, so it will cost you €0.085 per conversation.
These are off-the-shelf products that allow users (you) to configure their chatbot. You will be autonomous on chatbot management.
What you will need to do:
In our experience, here are the internal resources you will need:
Development of a chatbot with a SaaS solution takes on average 8 weeks before going into production.
After going into production, a chatbot manager must be retained (often the project manager). He will be in charge of analyzing the robot’s performance, and modifying its content if necessary.
You will still have to pay the product costs. The price of this type of solution generally starts at €2,000 excl. tax / month.
Some companies prefer to be accompanied by a communication agency or by their consulting firm.
In this case you will have to:
Here are the internal resources you will need:
The overall duration of a chatbot project developed by an agency or consulting firm is 16 weeks. They often take more time upstream of the project (reflection, etc.). This is where their added value lies.
After going live, a chatbot manager should be retained and your contract with the partner should continue. He will be in charge of making the changes and billing you for them.
You will still have to pay the product costs which, if the solution has been chosen, should be nil or almost nil (see core technology budget).
In my opinion there are two schools, and that is really what we see in the market.
Some companies make the strategic choice to internalize the chatbot creation skill.
The leader has a strong vision of what conversational robots can bring internally and externally.
The company has already carried out chatbot projects in POC mode, knows the project process and the necessary skills.
It is financially able to invest in recruitment dedicated to this area of activity. The recruited team must be able to keep up with the constant evolution of technologies related to chatbots and must master several different technologies that can best serve each of the targeted use cases.
Other companies have understood that chatbots are a tool that can serve very broad use cases, and that the technologies and skills required are different depending on these use cases.
They opt for outsourcing to SaaS solutions, experts in their field. This allows them to reduce the project budget, accelerate the delivery time and maximize the chances of success thanks to the autonomy of the teams on the solution.
My humble advice would be to avoid outsourcing your bot project to an agency.
That’s expensive. Increases the chances of failure. Makes you dependent on the agency/firm you work with and doesn’t help you build enough chatbot skills internally.
On the other hand, it is interesting to work with an agency / consulting firm in addition to one of the previous solutions, for example to work on the content or the design of the chatbot.
One of the questions that often comes to me is the estimated duration of the project. More concretely :
"When can we put our chatbot into production?".
As far as we are concerned, the duration of a chatbot begins as soon as the client has chosen the solution and gathered the necessary internal resources. No need to start the project if some necessary actors are not available.
Then the duration of the project depends on the solution for which you have opted (see above).
Here is the order of magnitude that we observed:
Using a SaaS product specializing in the creation of chatbots will allow you to go much faster than other solutions.
The technology is already ready and the majority of the work consists of setting up the chatbot (content, etc…).
If you leave it to the experts of the solution to do the configuration for you, you will save even more time.
Time is precious for a chatbot project. The sooner you have a final version, the more time you can give yourself to test the chatbot. And the more you test the chatbot, the more you can improve it (NLU, scenarios,…) and the more you increase your chances of success.
This is a central question (of course) for the success of a chatbot project in process automation.
In this type of use case, the value generated by the bot can be correlated to the human time it saves.
On the other hand, the creation of a quality chatbot represents a budget that can be significant because several technologies are involved, require extensive testing, etc…
The equation then becomes very simple:
Does the time the chatbot saves me + the value perceived by my customers> cost of creating and maintaining the chatbot?
This equation is true under one condition:
The chatbot has one or more defined, measurable and comparable objectives with the pre-existing process.
These are often repetitive tasks with low added value, which your teams no longer have the time / desire to manage.
Here is a summary of the overall cost of a chatbot depending on the chosen solution:
Coming back to my previous review:
Remember, one of the main values of the chatbot is to be able to connect with your IS to carry out personalized actions.
Here are some examples :
All of this cannot be done without connecting to your IS and without having an SSO login between your system and the chatbot’s system.
It is therefore necessary to think about the elements that you can connect with the chatbot with more or less complexity.
Your IS must therefore be accessible via APIs, connectors that allow programs to exchange with each other.
The first thing to do is to identify your use case and understand what technologies are needed to carry it out (NLU, logic, content, API, etc…).
Then you will have to choose which development method to choose: set up an in-house team, use a dedicated off-the-shelf solution or use an agency.
In summary, here are the pros and cons for each solution:
Finally, it will be necessary to ensure your ability to mobilize internal resources , particularly at the IT level (architecture / security).
Bravo (and thank you) for finishing this article.
Don’t hesitate to click on the big white button in the yellow insert below so that we can discuss it together!
The importance of the DSI in the company is all the more obvious when we talk about artificial intelligence....
Following the release of the final season of Game of Thrones this Monday, April 15, we present to you...
Where to limit the use of artificial intelligence? Should it be limited? Here's a real-life, sci-fi movie-like example of...
Learn how to optimize your NLU (natural language understanding) to create a successful ITSM chatbot.
Learn how to create a skill that will create tickets in Easyvista from a chatbot conversation!
Create step-by-step resolution scenario with ticket escalation option and problem solve button.
Book a 1 to 1 demo
with a Vizir expert and discover what a
chatbot can do for you.
Receive this type of content and many others (tools, news, testimonials, podcasts…) every week directly to the mailbox of your choice. Unsubscribe at any time.
