BNP Paribas Cardif automates 66% of inbound calls using callbot
Channels
Call centre
Industry
Insurance / Banking
BNP Paribas Cardif is the banking insurance leader in 35 countries with 100 million customers worldwide
Background
The company's hotline constantly buzzes, so one would assume that only a human agent could work with clients from all over the world. Yet, the company chose to automate calls with the help of conversational AI due to the prevalence of standard inquiries.
Cardif's call centre used to operate 14 hours a day, while the work of the hotline agents was deemed inefficient.
Not all customers were able to wait in the long queue for advice on critical/vital issues. Typically, agents spend half of their shift answering common questions such as “where can I find your office?” or “how do I get a payout?”.

Goals
The main goal was to help the clients of Cardif to get information faster and to reduce the burden on call centre agents
Respond to customers in loss events swiftly
Increase percentage of requests solved at the first call
Automate the process of handling common questions
Solution
- 1The Cardif team analysed dialogues with its customers to identify the 50 best conversation topics
- 2They then came up with 15-20 key phrases for each topic, having added the assistant's answers and a few friendly phrases
- 3For a month and a half, Cardif was discussing whether its voice assistant should be a cartoon character or a real employee with a profile pic
Cardi bot sends key takeaways via email or text messages, while in complicated cases it transfers the call to the agent
We taught the callbot to answer common questions about the company, insurance payouts, working hours, and website navigation.

How can Cardi help?
The Cardi bot had to be in line with the brand philosophy, be polite and friendly like a real insurance specialist
Respond to customers in loss events swiftly
Increase percentage of requests solved at the first call
Automate the process of handling common questions
Results
52%
Call containment
70%
Reduction in call waiting time
$9m+
Saving in operational costs over two years
Today, the FCR is estimated at
83%
As a result of the automation, the call center agents can now focus on higher-value tasks.
Cardif uses the FCR (First Call Resolution as a percentage of requests solved at the first call) metric to estimate the robot’s efficiency. If the customer does not call back within 24 hours after talking to Cardi, then the conversation may be considered successful.
We are getting ready to fully legalize the interaction of a person with a bot in the telephone channel, use voice biometrics and identification using security questions. The new goal is to automate 80-85% of incoming calls.
Performance evaluation
83%
of NPS survey participants gave the highest rating to the company and its automated services
Cardif regularly conducts NPS surveys to estimate an index of customer brand loyalty
Cardif regularly conducts NPS surveys to estimate an index of customer brand loyalty. A recent survey showed that 52% of customers are willing to recommend the company to their friends and acquaintances. Moreover, 83% gave the highest rating to the company and its automated services.
Before Cardi, the NPS survey was conducted by two employees. Following the success of Cardi's robot, Cardif decided to do a similar project for outgoing calls. Now, another robot assistant, Cardi's colleague, conducts the surveys.
The assistant communicates with 400-500 customers each month and approximately 40% of them agree to take part in the survey
The effectiveness of the survey was lower when the calls were handled by company staff. Cardif plans to teach Cardi to answer questions on contract management and amendment.
The effectiveness
of the survey was
lower
when the calls were handled by company staff
The head office of BNP Paribas Cardif in Paris was also pleased with the results of Cardi's work.
Soon, Cardif plans to scale the adoption of conversational AI technology to all of its offices with Tovie AI
Do you want to streamline your business?