)
Generative AI supporting business. The Autoreply case as an automatic customer care respondent.
.
In the corporate world, front-office operations are often repetitive, laborious and require specific skills from staff. In addition, the human element inevitably introduces a margin of error that can negatively impact overall efficiency and customer satisfaction. The Large Language Model (LLM) represents an innovative solution to improve the efficiency and productivity of companies. These advanced natural language processing models are designed to understand and generate text in a human-like way, offering a wide range of applications that automate repetitive and time-consuming tasks
.
The customer interaction department of a leading company in the online banking services sector has highlighted the need to optimize the flow of managing requests from customers via e-mail. The main focus is on generating quick and effective responses for generic and common requests, proposing a draft email through Autoreply, thus freeing up valuable resources to focus on the most relevant issues that directly influence business performance.
)
)
Automate the response of generic and repetitive questions that have a low impact on the business, ensuring customer privacy while keeping information within the business environment.
Given the absence of a cloud infrastructure, the customer operated with an on-premise architecture characterized by limitations in terms of computational power, affecting access to high-performance hardware and services. An additional restriction concerned the management of customer personal data and the lack of authorization for their sharing with third-party services. This made it impractical to use external platforms that offer generative artificial intelligence solutions.
Target Reply has therefore developed an end-to-end solution based on LLM, fully functional offline and on limited resources.
The project deals with receiving emails from mailserver and uses a knowledge base based on Frequently Asked Questions (FAQ), internal documentation and conversation history as reference data.
The system is able to understand the request, extract the most relevant information through Retrieval Augmented Generation (RAG) techniques and rework it to generate personalized responses in email format.
The system then proposes a draft response, which will always be subject to human validation before sending it to the recipient, allowing it to be improved or to correct any errors.
In addition, the system constantly optimizes company FAQs, integrating new requests and updating existing answers to ensure greater relevance and completeness.