PAYMENT MANAGEMENT

A system supporting customer request processing and quick responding

Case study image
756
hours per month saved for other tasks
50%
– the time for document processing was shortened by this much
86-95%
automatic document classification accuracy (including categorization, prioritization and intent detection)

ABOUT THE CLIENT

Universal international bank


Credit Agricole is a universal international bank that has been operating in Poland for nearly 20 years.

Credit Agricole Bank Polska is one of the most frequently recommended banks in our country. It operates in the area of retail, corporate and agricultural banking, small and medium-sized enterprises, and in the area of Consumer Finance.

Credit Agricole Bank Polska is part of the Credit Agricole Group that belongs to the 10 largest banks in the world in terms of asset value. This group operates in 48 countries globally and serves over 52 million customers.

CHALLENGE

Improving reply speed and quality

Credit Agricole Bank Polska faced challenges related to enhancing efficiency in assigning customer cases, prioritizing them, and preparing replies.

Credit Agricole Bank Polska has been our long-term business partner whom we effectively support in the field of technological solutions. This time, the bank turned to us with its challenges regarding dealing with customer requests. The first challenge was the classification of incoming documents. The manual handling of a large number of requests led to them being assigned to the wrong categories within the company’s system. This had consequences in the form of request processing delays, potential financial penalties from regulatory authorities (e.g. due to failure to meet the reply deadline) and decreased customer satisfaction retention rate.

The second challenge was to prioritize responding to customer requests. The incoming documents covered not only a wide range of issues to be resolved, but also a varying degree of the customers’ emotional intensity (e.g. a customer asking for help vs. a customer dissatisfied with a service). Assessing these two matters by a human made the task highly time-consuming. This, in turn, affected the request processing time.

The third challenge involved preparing replies to customer requests. It was a long and arduous task; it should be also noted that preparing a reply to a non-standard request took 2.5x more time in comparison to standard requests.

Case study icon

Key areas to improve with the right solution

Case study icon

reducing the risk of incorrect request assignment in the bank’s internal system

Case study icon

eliminating the risk of a request getting stuck at one of the processing stages

Case study icon

reducing the long response time in terms of customer requests

Case study icon

increasing the correctness and accuracy of replies to standard and non-standard requests

Case study icon

speeding up prioritization based on the case type and emotional intensity

Case study icon

automation of tasks previously carried out manually

SOLUTION

Using an AI-based system


ROC3 – the world’s most human AI assistant

ROC3 is an innovative solution that supports the handling of complaints and requests as well as other after-sales processes. To achieve this, it uses artificial intelligence (AI), especially natural language processing (NLP), natural language generation (NLG), and machine learning (ML) algorithms.

The reason for using ROC3 is to shorten the time needed to prepare replies to customer requests (up to 50%), and full automation of standard operations regardless of the style and format applied to the customer requests. In addition, the solution turns unstructured data from the incoming requests into knowledge about the customer (e.g. satisfaction with the services or customer care). An additional advantage is the system’s ability to independently recognize some cases and execute appropriate orders, e.g. withdrawing marketing consents at the customer’s request. In such cases, no employee time or attention is required. Achieving this level of automation is guaranteed by three basic ROC3 modules: Extractor, Classifier, and Generator.

Main areas of supporting Credit Agricole Bank Polska S.A. employees by ROC3

Case study icon

Significantly shortening the time of handling complaints and requests while increasing customer satisfaction regarding after-sales processes

 

Case study icon

Automation of certain tasks of Back Office employees to eliminate case categorization errors and reduce repetitive activities so that the employees can deal with tasks requiring human creativity and decision-making

Case study icon

Shortening response times as per SLAs (up to 50%) thanks to predefined answers, even for unusual cases

Case study icon

Standardizing replies and ensuring compliance with simple wording rules to guarantee easy-to-understand communication with customers

Case study icon

Identifying customer intentions regarding the execution of obvious orders by robots (RPA) instead of engaging employees

HOW DID WE DO IT?

Enhancing the capabilities of a banking platform


The project was cross-implemented. There were two teams participating

Case study icon

a team permanently cooperating with CABP, responsible for developing the IT architecture, CI/CD (DevOps)

Case study icon

a team of data science and AI architects, responsible for the entire implementation, AI algorithms and MLOps

Experts involved in the project on Credit Agricole Bank Polska’s side

Case study icon

Director of Customer Service Quality Development Department

Case study icon

Director of Innovation

Case study icon

Leader in the Customer Advocate Office

Case study icon

Configuration Management Senior IT Specialist

Case study icon

Senior Business Process Robotization Specialist

Experts involved in the project on Deviniti’s side

Case study icon

Data Scientist

Case study icon

Machine Learning Engineer

Case study icon

DevOps Specialist

Case study icon

NLP Specialist

Methodology


Due to the project’s innovative nature, we jointly decided that the best solution would be implementing the project through the Agile methodology. This methodology was intended to translate into quick software implementation and close cooperation between the teams. The iterative mode of work allowed for flexibility in product deployment. 

Platform


As part of this project, together with the CABP team, we expanded the Remedy system used by the bank to manage the entire after-sales customer service process. In this system, we have embedded a Widget, i.e., an AI-based Writing Assistant (an assistant supporting the writing process based on artificial intelligence).

In addition, Remedy uses all available information classifiers and extractors. It places the extracted materials in its own database. The widget finally generates a print-ready DOCX file containing the employee’s signature and the customer’s personal data.

Technological stack

Development


The work was carried out – per the Agile methodology – in two-week iterations, resulting in subsequent versions of the application that provided new functionalities. The teams met regularly for workshop sessions to review and discuss the effects of their work. Together, we clarified the requirements for the following project stages.

HOW DOES THE SYSTEM WORK?

Automating customer request processing


Using ROC3, employees can save time and energy on duties such as receiving and classifying customer requests, prioritizing them, and preparing the reply text. The system will perform most of these tasks for them, while employees only need to verify the created documents. ROC3 contains three main modules; each is responsible for a specific work stage regarding customer requests: Extractor, Classifier, and Generator.

  • Case study image

    Extractor

    Receipt and preliminary analysis of documents

    • the system scans the document (or performs OCRs) and extracts essential information from them, i.e., what the purpose of the request is and what results are expected by its author
    • the Extractor module determines whether the request was sent from a private person or an institution (e.g. UOKiK [the Polish Office for Competition and Consumer Protection], or KNF [the Polish Financial Supervision Authority]), recognizes the customer by their identifier and determines which product or service the request concerns
    • ROC3 also recognizes the emotional intensity and style of the documents to prioritize them adequately later on (e.g. whether it is a polite request, an aggressive complaint or an official document)

  • Case study image

    Classifier

    Classification and prioritization

    • the system efficiently registers customer requests and immediately forwards them to the appropriate team
    • the Classifier module also finds requests that have been sent to the wrong department and automatically moves them to the place where they belong
    • ROC3 can even recognize simple customer orders (e.g. a request to withdraw the marketing consent) and implements them independently

  • Case study image

    Generator

    Support in composing the reply content

    • ROC3 helps employees with preparing a reply to the customers as it has full and immediate access to all company messages (standard and non-standard)
    • the Generator module offers ready-made paragraphs that match the customer request content and can be input by the employee with one click; it also allows the employee to add customized text
    • the system checks on an ongoing basis whether the prepared answer is linguistically correct, clearly written and contains all the information the customer needs
    • the employee can also quickly find the necessary fragments based on phrases, or easily add/remove individual parts of the prepared reply
    • a request that is ready to be sent in PDF format, is created on a company template, and contains the employee’s signature can be generated with one click

RESULTS

Increasing work efficiency and customer satisfaction

756

hours per month saved for other tasks

50%

– the time for document processing was shortened by this much

86-95%

automatic document classification accuracy (including categorization, prioritization and intent detection)

40%

of cases processed much faster thanks to ROC3

62%

of cases related to certificates processed with RPA (by robots only)

32%

of cases related to account closure processed with RPA (by robots only)

Since the introduction of ROC3 to the system, Credit Agricole Bank Polska has been enjoying many benefits. First of all, the response time regarding customer requests has decreased. In the case of simple requests, customers receive an answer right away because the system executes their orders independently. More complex cases, including non-standard ones, are handled faster than before (up to 50% faster). This positively affects customer satisfaction levels.

Another significant factor for Credit Agricole Bank Polska is compliance with regulations imposed by authorities such as the Polish Financial Supervision Authority. This applies, i.a., to the rules for processing requests which are set out in a specific legal act. Thanks to speeding up customer request processing, the bank can always meet the deadlines for SLA-related tasks and avoid financial penalties imposed by KNF.

The last important issue was to increase the reply correctness and enhance customer satisfaction. The combined forces of the bank’s employees and ROC3 ensure that the replies are correctly composed. In addition, all issues raised by the customer will be considered and duly handled (no case can escape the AI). Correctly written replies relating to the customer’s case positively affect the bank’s image and customers’ willingness to continue cooperation.

references

What does our client say?

Logo image
Extended customer service time and lower customer satisfaction were the main problems we faced before implementing the ROC3 solutions. It was a great challenge to find mechanisms that would shorten the customer case handling time while maintaining high quality of the process. The implementation of ROC3 enabled the proper assignment of customer cases to the units responsible for their processing (including shortening their execution time), the easy composition of replies in complex and multi-threaded topics, and the full robotization of standard processes. Thanks to data systematization, we are implementing robotization for subsequent processes. The prepared improvement solutions were implemented primarily with external customers in mind, and they worked great here. However, we must remember about the new generation of employees who are more willing to use the latest technologies. We notice the need to move away from the usual solutions and invest in new-generation tools to guarantee employee satisfaction and work comfort.
Ewa Traczykowska
Head of Quality of Customer Service Development Department in Crédit Agricole