RAG for Text Generation Processes in Businesses


Iñaki Peeters

AI Solutions Analyst

Welcome to the first part of our four part series on Retrieval-Augmented Generation for text generation processes in businesses. In this series, we will discuss one of today’s hottest topics within the field of Generative AI: context-aware Generative AI with Retrieval-Augmented Generation.

Distressed office worker seated at his desk. With one hand, he’s clutching his hair in despair, while the other hand actively writes on a piece of paper, hyperrealistic digital art – Dall-E 3

Common Business Cases for Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) is a technique that augments the capacities of Large Language Models (LLM) by combining elements of both retrieval-based models and generative models. Where LLMs can only access the data they have been trained on, RAG allows for additional information retrieval. This means that information from data sources that have not been used to train the LLM can be accessed and used as additional input for LLM’s to generate text, providing extra information and context. On a first level, this additional information retrieval allows for what is called 'ChatGPT with your own data'. With a RAG system, in essence, businesses can, in a chatbot-like manner, ask for relevant, company-specific information similar to how we use ChatGPT. On a higher level, this additional information can be used to generate a specific response or output, allowing businesses to generate text based on or containing company-specific information. In this series, we’ll primarily focus on the overarching text generation capabilities of RAG systems.

Throughout the series, we’ll adopt a business-centric approach. By defining clear and distinct business cases, we will discuss the associated considerations for deploying a RAG system tailored at the needs of your specific business.

  • In part 1, we will introduce three different clusters of business cases we identified and what business value can be realised by deploying a RAG system. In each of the following parts, we’ll delve deeper into the practicalities of RAG, progressively adding more complexity by elaborating on one of the clusters identified in part 1.

  • In part 2, we’ll take a closer look at the basic architecture and implementation of a RAG system.

  • In part 3, we’ll explore how a RAG system can be deployed in an enterprise context.

  • Finally in part 4, we’ll take a look at some of the possible extensions to build on top of a RAG system.

Let’s not waste any more time and dive into it!

Business Value of RAG: a Cluster-Dependent Question

The business value of RAG for businesses boils down to business cases where some common characteristics can be distinguished. As explained above, we identified three clusters of different business cases each expressing a different (and cumulative) need. As can be seen from the figure below, every cluster shares some similarities with the other two. However, there are also some differentiating factors. The more we move to the bottom right cluster, the more stringent these factors become, hence the RAG system also becomes more complex to fit the needs of these specific use cases.


For all clusters, we see that the business use cases share the following features.

  • Dependence on manual work: typically, processes that can be optimised using RAG require substantial manual effort. Implementing a RAG system can help businesses reduce the dependence on manual work, making the process more efficient and also important: less boring.

  • Time consuming: this one goes hand in hand with the previous one. A lot of these manual processes obviously take quite some time. The same reasoning holds here: implementing a RAG system to streamline these processes can increase efficiency and free up time for other value adding activities.

  • Use of internal and external data sources: one of the powerful features of RAG is the ability to access both internal and external data sources. Internal, or company-specific sources are for example organisational databases, client information, domain specific datasets, etc. External data sources can be anything that can be found online, either public or non-public like websites, external API’s, (research) articles, etc.

The Need for Automation

The first cluster of business cases involves companies using RAG to automate time consuming, manual processes mainly for the purpose of speeding up text generation processes. The generated text just serves as inspiration and guidance for employees tasked with writing specific content. The employee can mix and match, make adjustments, and select the most relevant pieces for the content to be written.

Examples of business cases within this cluster are:

  • RAG-assisted writing for content creation (I promise, this post wasn’t generated using a RAG system).

  • Support documents for employee onboarding like guidelines, manuals, etc.

While the context of the text that has to be generated might differ, these use cases do have something in common besides the characteristics listed above. For the use cases in this first cluster, there is some ‘poetic license’ (or: ‘dichterlijke vrijheid’ as we say in Dutch). Of course, the generated output should preferably be accurate and complete, however, there is some room for variations, as the generated text mainly serves as ‘evidence-based inspiration’. Typically these use cases do not have a very strict predetermined output.

The Need for Scalability

The second cluster of business cases goes one step further by adding a scalability requirement. When deploying a RAG system, or any AI system in fact, businesses have multiple options. On the one hand, an AI system can be built locally. This implies that the application is running on the business' own servers. In essence, this means that all necessary infrastructure is located on the premises of the organization. On the other hand, businesses might opt to outsource infrastructure needs to cloud providers like Amazon Web Services, Google Cloud, or Microsoft Azure. This outsourcing, also known as cloud computing, implies that, through an Internet connection, businesses can make use of infrastructure, software, or platforms that are not a located on their own premises. Basically, this boils down to ‘renting infrastructure’ to run an AI application. This infrastructure renting can have several benefits over locally deploying:

  • Cost efficiency as cloud computing requires less upfront investment compared to setting up and maintaining on-premises hardware

  • Scalability as businesses can easily increase computing capacity when needed

  • Reliability as data backup and recovery are often more robust compared to on-premises hardware

  • Security as cloud providers typically invest heavily in security measures

Examples of business cases within this cluster are:

  • E-commerce businesses leveraging RAG to generate (personalised) product descriptions

  • Tech companies using RAG-assisted writing for handling FAQ and customer complaints

As the examples indicate, some businesses might need a scalable cloud solution to deploy their RAG system. Businesses might benefit from having the ability to access extensive computational power and scalability options to handle a large and volatile number of requests. Besides that, integration opportunities with other cloud services and maintenance considerations might as well be incentives to opt for a cloud-based alternative.

The Need for Accuracy

The third cluster of business cases again goes a step further by adding two additional characteristics. This cluster contains business cases where the text to be generated requires a substantial amount of domain expertise. Besides that, also the accuracy and correctness of the generated text is taken to a new level. Where in cluster 1, there was some leeway in terms of structure and content of the generated text, in cluster 3, there is a high need for 100% correct output as the business value is too high to account for errors. Examples of business cases within this cluster are:

  • Pharma and biotech companies compiling regulatory documents using RAG.

  • RAG assisted writing for Vehicle Design Specifications documents in the automotive industry.

  • Software companies drafting Software Requirements Specification documents using a RAG system.

Although the purpose and content of the documents to be generated in the business cases above might differ, it is clear that, given the specific structural and content-related requirements of these documents, accuracy has to be taken to the next level. In order to safeguard these accuracy requirements, additional extensions can be built on top of the conventional RAG system to ensure optimal accuracy and efficiency.

What’s in It for You?

From these three clusters of use cases, it is clear that automation of the text generation processes using a RAG system can reap multiple benefits. Automation will not only accelerate these processes, it will also help to reduce the dependence on manual labor and induce cost savings, which has a direct and positive effect on the bottom line. Moreover, automation increases process efficiency, productivity, and scalability throughout (growing) organisations. Companies that adopt a RAG system can gain a competitive advantage by being able adapt more quickly to dynamic environments. Overall, there is a clear need for the automation of text generation processes using RAG systems, as it has incredible potential to create additional value for business in a wide variety of areas.

And What About the Data?

For many years, the conventional paradigm within the field of Artificial Intelligence was a model-centric approach. People worked for hours trying to improve the algorithms in order to achieve better results. For years, this was a justified way of thinking. Over the last years however, impressive progress was made when it comes to models and algorithms. First of all, access to open source code and models, together with the rise of cloud-based AI-platforms has made AI more accessible than ever. More importantly, a clear need for data interpretability, reliability, and robustness have initiated a turnaround from model-centric to data-centric AI. Nowadays, frontrunners like Andrew Ng are convincing people to actually implement this new paradigm. Ng defines data-centric AI as: “the discipline of systematically engineering the data needed to build a successful AI system”. People should shift their focus from trying to improve models to trying to improve their data. At Faktion, we share this opinion. We believe that proper data quality is what will unlock the full power of an AI system and what makes you stand out from the rest. After all, garbage in, garbage out. For this reason, throughout this series, special attention will be paid to data quality and its implications when it comes to RAG.


Qualitative data that is structured in an efficient way is an important prerequisite to get the most out of AI applications. However, setting up your database is also a very boring job, especially when it’s done manually. Luckily, AI eats boring tasks for breakfast. At Faktion, we developed the Intelligent Data Quality Optimisation (IDQO) Toolbox. The IDQO toolbox is an AutoML toolbox that enables companies to streamline ‘data quality tasks’ (e.g. classification, matching, enrichment, validation, clustering, similarity search, etc.). We envision IDQO as reusable building blocks that we can easily tailor and apply depending on the specific requirements of a use case. At the end of each of the following posts, we address a specific data quality task or requirement for setting up a RAG system, providing a solution by explaining how one of the features of our IDQO toolbox effectively tackles the highlighted issue.

In the next part of this series, we’ll elaborate on the first cluster of use cases and explain the solution suited for this type of use cases: a basic RAG system. We’ll explain what the basic architecture looks like and what considerations should be taken into account before setting up your own RAG environment.

Get in touch!

Inquiry for your POC

Scroll to Top