# Flip.kz Voice of Customer (VoC) and Root Cause Analysis (RCA) Study

## Introduction

```{image} Similarity_Match_Pipeline.drawio.png
:alt: Similarity Match Pipeline
:class: bg-primary mb-1
:width: 800px
:align: center
```

In recent years, AI-generated summaries have become a notable feature in the LLM era. By using AI summaries, we can simplify content and structure it by compressing large amounts of data, including videos, captions, papers and books.

AI has taken a major step in e-commerce, becoming an important tool for optimization and recommendation in online retail since the advent of language models. In recent years, top marketplaces around the world have added this feature to their platforms, and it has become increasingly popular. It improves the shopping experience by saving users time when reading reviews.

However, some platforms have started copying AI summaries from others, making their product listings more attractive than those without reviews. We know that the presence or absence of reviews affects purchasing decisions: users are more likely to choose products with reviews because they rely on others experiences. At the same time, tests have shown that products with AI summaries can outperform those with reviews only. In some marketplaces, however this feature is still difficult to implement effectively.

A relevant case is Kaspi.kz. Kaspi offers a wide range of products, from everyday goods to cars. One challenge is that a single product can have multiple sellers, except for original items where the brand or trademark is registered in Kazakhstan. This makes implementing AI summaries more costly and complex.

The underlying issue is quality. Not all sellers offer the same quality, even for similar products, and differences are often reflected only in the price. One possible solution is to generate a separate summary for each seller's product reviews. However, this creates difficulties for new sellers, who may not have enough reviews to build trust and generate sales. As a result, they may lower prices aggressively to remain competitive. A concerning risk is that the marketplace might manipulate pricing strategies and justify them using AI-generated summaries.

Another example is Flip.kz, one of the first marketplaces in Kazakhstan to implement AI summaries on its e-commerce platform. Flip has a well defined market position, a loyal customer base and is a leader in its segment. Its main categories include books, beauty products, home goods and pet supplies.

We collected data from Flip.kz with more than 28 000 products included in the dataset. The main fields are product ID, name, detailed information, AI summary, tags, and reviews [(Parser Code)](https://github.com/salamqt/flip-kz-parser). In total, 431 000 reviews were collected, including the review date, author, rating, pros, cons and review text.

![VoC Analysis pipeline](pipeline.png)

As an important analysis in e-commerce, in this study we decided to conduct VoC Analysis (Voice of Customer Analysis). VoC Analysis helps to define product problems faster based on user feedback. It can be an optimized solution for e-commerce.

RCA (Root Cause Analysis) is one of the main types of analysis in customer service. We can identify product problems and causes of refunds in e-commerce by analyzing customer feedback. It helps sellers make decisions in a more optimized way.