Unbetitelt

[businessanalyticsenthusiast]

Business Analytics: Trends or Necessity?

What is Business Analytics?

Business Analytics is everywhere. You hear people talk about it, but do you understand what they mean by it? Business Analytics, at its core, is the scientific process of turning diverse data into meaningful insights that organizations can use. The field combines statistics, programming, data management and domain knowledge to help businesses better understand what is happening, why it is happening, what might happen next and what they should do next. It is used for operational as well as strategic decision-making in businesses around the world. Today, business analytics plays a huge role in organizations as they rely on it to improve efficiency, reduce costs, predict future trends and enhance customer experiences. This can go from small decisions like how to optimize inventory to huge decisions like how to expand into a new market. Business Analysts provide evidence and clarity to support efficient choices in a quickly changing business world.

Why People Think It is a Trend In recent years, there has been an increasing hype from the media about analytics, big data, AI and machine learning. More and more people started to use large language models like Chat GPT and the need for data scientists and analysts has risen drastically. While companies talk about big-data and a newly developed data-driven culture, behind everything are business analysts that deal and work with the data and turn it into the insights the business need. Due to the influence of social media and the increasing supply of online courses, people tend to only see analytics as a trend recently.

Why It is Actually a Necessity

Companies generate massive data using apps and a huge number of transactions every day. The need for real-time decision-making increased over the years and companies are in the need of having analysts that help them make business-driven and data-driven decisions. Due to large amounts of competition, companies need to have evidence for their decisions and a manual analysis is just impossible for the modern data and its volumes. That is when Business Analysts come into place.

They use their skills of programming, working with the data, machine learning, etc. to

Make data-driven decisions. They generate what-if-scenarios and ask just the right questions. They transform the data and build useful dashboards and give invaluable recommendations for the businesses by telling compelling stories. Additionally, they leverage augmented analytics. Further, with increasing regulations in regards of compliance, governance and business ethics, the data needs to be managed properly.

Examples of Business Analytics in Daily Operations

Business Analytics is everywhere. Think of any field and you will realize soon enough that data and transactions happen in every area of life. Let`s think about retail. Business analysts are needed to make demand forecasts or improve inventory. In Finance, a huge part of machine learning is to predict. Imagine without Business Analysts, banks would have a hard time predicting if a potential customer could be a fraud or not and lose lots of money. With Business analysts, this risk is reduced immensely. In Healthcare, business analysts work with data in order to predict patient pathways and to allocate resources. In Marketing, businesses need to segment their customers or make churn predictions which can be done due to classification by business analysts. Think of any other fields and you will realize that there is most likely someone behind the company that dealt with the data before there has been a decision made.

Skills Needed Today and Why They Matter

A basic foundation for any Business Analyst today is knowledge bout SQL and databases. This is the background of retrieving data. Before even using SQL and doing any query, analysts need to really deal with the data, ask lots of questions and make sure they understand the project requirements. By using SQL then, they can simply query the most important information. Next to that, they need to know about ETL and ELT in order to extract, transform and load or extract, load and transform the data in a way that makes sense for their specific business context. Further, they need to have basic statistics knowledge which is a base for understanding data problems. Then, machine learning skills are needed in order to make classifications and predictions for numerical and categorical data. Further, business analysts should be able to code and use different programming languages like R or Python. Lastly, it matters a lot that they are able to break down the most difficult queries and code into easy to understand graphs and visualizations so that communication and storytelling is facilitated.

Why Companies Cannot Ignore It

Next to all reasons stated above, companies need to invest in their data analytics teams as this will increase the customer experience, it will reduce costs by optimizing processes and operations and will lead to better product decisions and predictions. Moreover, business analytics helps to have a stronger data governance and be compliant with the law and regulations. Additionally, by having prepared and ready to use data and dashboards, companies can get faster insights and make quicker predictions which can give them a competitive advantage.

Future of Business Analytics

There has been a huge increase in business analytics in recent years and it looks like real-time analytics will become the new norm. There will most likely be more increases automation in business processes as well as more low-code analytical tools and more integration of AI in data workflows in the workplace.

Conclusion

So, is Business Analytics just a trend? No, it is a long-term business requirement and therefore a necessity. Organizations that ignore it will most likely fall behind and will have trouble keeping up with customer demands and competition. It is most important to use business analytics efficiently as data literacy is becoming a core skill for everyone.

[businessanalyticsenthusiast]

The Thinking Revolution

Do you still see AI as just a chatbot? Or have you accepted AI as your efficient coworker? In my earlier posts, we looked at how AI reads using tokens and how it finds facts using RAG. But there is a shift happening in 2026. We are moving away from System 1 AI which is fast and can sometimes be wrong to System 2 AI which provides Reasoning models that are slower and more logical which can be seen in the image below.

Figure 1: System 1 vs System 2. Google. (2026). Gemini 3 Flash LLM.

The image explains the difference quite well. While System 1, so standard LLMs, are fast and confident, they often fail at simple math because they can just predict the next word without thinking ahead. System 2, however, are reasoning models like OpenAI o1 or DeepSeek R1. They use internal thinking tokens. This means that they pause, reflect and self-correct by verifying their logic before they give you an answer. This makes it more accurate or less likely to be hallucinated.

One secret phase that augments all your prompts? “Let´s think step by step”. Researchers found that by simply adding this to your question, the prompt can increase accuracy on logic tasks. This is called chain-of-thought prompting and it helps the model to use reasoning before it answers.

In the business world, we are now using agentic RAG. This means that we not only just use one model, but use multiple and accept them as basically our helpful coworkers. For one, we use the orchestrator which is a small and fast AI that perceives your questions and decides which files to look at. Then, it calls tools like a python script or a vector database to either do math or to find a specific page in the company´s SharePoint. If the answer is wrong, the agent reflects and tries again. This is the crucial step that differentiates it from just being a chatbot.

So, right now we are at System 2 of AI but where are we going? Following steps are System 3 which uses agents so that systems can take actions completely independently to System 5 where AI can actually do the work of an entire department at a company. Will we get there eventually, or do you think we are already at our maximum limit that AI can reach? Let me know by commenting below and let´s get the discussion rolling!

[businessanalyticsenthusiast]

From RAGS to RICHES

Have you ever asked an AI what happened yesterday and it either said “I don´t know” or even worse, it lied to you confidently by stating something that never happened? That is because Large Language Models (LLMs) have a temporal knowledge gap. They are trained until a specific day and are frozen at their training cutoff. This means that they have a domain expertise deficit as they have never seen for example company specific data.

In order to not get hallucinated answers, what could you do? The solution is Retrieval-augmented generation (RAG). It is an architecture that lets an AI reference an external knowledge base before it answers. The interesting part is that the AI does not have to memorize everything, it just needs to know how to look up the external information.

But what are the steps to success? In order to get your private data into an AI´s brain, you need to follow a specific five step workflow that is illustrated below.

Figure 1: RAGS to RICHES Pipeline. Google. (2026). Gemini 3 Flash LLM.

The first step is the preprocessing which creates the chunks. As you cannot feed a large PDF all at once into an AI, the chunking breaks it into smaller segments so it can fit the model´s memory.

Then, using the embedding strategy which I explained in my last post, you turn those chunks into numerical vectors.

The vector databases are used to store and index vectors using for example inverted file index so they can be searched quickly.

The similarity search / the retrieval is when you ask a question and the system turns your query into a vector to find the chunks in the database that are mathematically closest in meaning.

After this, you can augment your prompt as the system pins the retrieved facts to your original question and the LLM then answers only using the context you provided.

So, why should you brag about RAG? RAG is faster and cheaper than fine-tuning a model. It reduces hallucinations as the AI is forced to answer based on the provided text and cannot image stuff. Also, it can actually help you in writing your next paper as it accurately cites which page it could the information from, to build user trust. Further, you can connect a model to sensitive data without that data becoming a part of your model´s permanent weights which builds data privacy. Lastly, it offers transparency as it makes the logic behind an LLM explainable.

Think of it like this: When in the past, an expert left a company, their knowledge left with them and the whole team had to suffer from it by trying to build back the knowledge. Now, however, companies use RAG and upload their “collective brain” into a permanent digital assistant that they can ask any company information at any time.

[businessanalyticsenthusiast]

Teaching Computers the “Vibe”

In my last post, we saw how AI breaks down language into tokens, so basically turning words into a set of numbers. But here´s the problem: To a computer, the number for “dog” and the number for “cat” are just two different integers. It has no idea that they are both pets or why they should be connected to each other. To solve this issue, we use vector embeddings which allow the AI to understand semantic similarity.

Are you confused? That´s okay, I was to. Let´s start with the basics. What exactly is an embedding? At its simplest, an embedding is a dense list of numbers, so a vector, that can represent unstructured data like text or images. Compared to just tokens, it carries the meaning of the data. So, when two words have a similar meaning, their vectors in this high-dimensional space are mathematically close to each other. As an example, the vector of “dog” will be closer to the vector of “cat” than it will be to the vector of “airplane” because dog and cat are both pets and therefore closer related.

Very cool, right? However, it wasn´t always like this. Before we had modern embeddings, we used a method called TF-IDF which stands for “term frequency- inverse document frequency”. It counted only how often a word appeared which made it great for basic searches, but it disregarded grammar and order completely. Then, with Word2Vec, there was a major breakthrough. The philosophy was: “You shall know a word by the company it keeps.” So it learned word relationships looking at their neighbors in a sentence which allowed the model to actually learn meaning for the first time.

Now that we have embeddings, we can actually perform math on them! This is called semantic arithmetic. Because vectors capture relationships, researchers found that if you for example take the vector for “King” and subtract “Man” and then add “Woman”, the resulting vector will be…you guessed it: “Queen”. This really proves that we are so far that the model does not just memorize words, but it really understands concepts like gender and royalty as mathematical directions.

This works because AI embeddings have so many dimensions, and every dimension combined allows the AI to create those unique mathematical signatures for every concept. Why does it matter, you might ask? Embeddings are the foundation of everything we do with AI today. They are the hidden underlying power of recommendation systems, like when amazon.com again suggests you a great new product? Yeah, that´s based on embeddings. Also, they are used for many more topics like image recognition or fraud detection. Most importantly, they are the first step in the retrieval-augmented generation pipeline. Without embeddings, the AI could not search through a company´s SharePoint to find the right answer because it would not know which chunks of text are relevant to the actual user question.

You now understood the importance of embeddings which take the integers of tokenization and give them semantic meanings. In my next post, you´ll learn how businesses use these embeddings to build “RAGS to RICHES” systems that can chat with their own private data.

[businessanalyticsenthusiast]

Why AI Fails at Counting “Strawberries”

Have you ever asked your favorite AI to count the number of “r”´s in a word like strawberry? No? You might want to give it a shot and you´ll be surprised what the answer will be like.

We expect those AI´s to be those ever knowing machines that you can ask anything. However, if you take a closer look at the tokens behind it, you will soon begin to understand something called the “jagged edge” of AI intelligence.

Let´s dive into it. Firstly, large language models do not read text the way humans do. They process tokens, which are basically integers assigned to chunks of texts. Because the model sees the integer representing a whole chunk, the individual characters inside that chunk are mathematically invisible during the model´s processing. In the word strawberry, a model might tokenize it into two chunks: “STRAW” and “BERRY”. If we now ask it to count the letter “r”, it isn´t looking at the letters but only at the tokens. Since “r” is not a standalone token in that sequence, the model might guess it wrong.

The modern tokenization, however, took a long time. Earlier, in the bag of words era, text was crudely split by whitespace and punctuation. Also, stop words were removed in order to save space and punctuation was discarded. The flaw? This method resulted in out-of-vocabulary errors. So if a model had not seen a specific word during its training, it did not know what to do with it. Now, more modern LLM tokenization use sub-word units. This means that modern models use fragments of words, allowing them to construct any word from smaller pieces. Further, unlike legacy methods, LLMs retain stop words and punctuation because they are essential for understanding syntax, nuance and logic.

Now, let´s dive even deeper into the technicalities of byte-pair encoding (BPE). It starts by treating every individual character as a token. It then scans the data to find the most frequent pair of adjacent characters and merges them into a single new token. This process repeats until a target vocabulary size is reached. For example, GPT-4 iterated until it reached exactly 100,277 unique tokens. This comes to the efficiency balance which I will discuss in the following. Character-level tokenization has a tiny vocabulary but creates sequences that are too long for the model´s memory. Word-level tokenization keeps sequences short but results in a massive and unmanageable vocabulary matrix. BPE provides therefore the best solution as common words become single integers while rare or complex words, for example long German compound words, are broken into meaningful sub-units.

You might ask yourself, why would the vocabulary size even matter? Well, as explained, frequent words are compressed into a single integer which makes the model faster and more efficient. For handling complexity like in long German words like “Eierschalensollbruchstellenverursacher”, sub-word tokenization is needed and it allows the model to see the smaller words inside the one big word, which is not possible using other methods.

So, now you learnt about tokenization, but you might ask yourself: “Is that all there is to know?” and I can tell you that there clearly is more about the ever-knowing AI´s that make them so smart. If you want to learn more about embeddings, which is the next step and allows a model to understand semantic similarities between words, make sure to read my next article which will be published later today!