Introducing the Concept

Business Intelligence, while sounding like another unnecessarily used buzzword, is a practical tool which uses modern technology to support decision-making. Methods include processing vast data sets, helping businesses make smarter and more calculated choices. Ultimately, leading to better and more meaningful outcomes.

Its Vital Role

Traditionally, business decisions were entirely based on human intuition alone, a practice which is now recognised as sub-optimal. Today, advancements within the technology sector mean that we are able to process information at a rate unfathomable to the human mind. A more obvious example of such innovations: Artificial Intelligence (AI), an umbrella term for technologies whose purpose it is to complete tasks and solve problems that humans would have otherwise, but instead; rapidly, at scale and with consistent accuracy.

Data has long been the lifeblood for many industries now, and is beginning to assist professionals in all corners of commerce. The process of transforming vast and various data sources into actionable insights is essential for business growth. A journey in which data professionals play a vital role.

The Typical Workflow

BI in Action:

Real-World Application

While we emphasise the limitations of human intuition alone, this doesn't diminish the value of human ingenuity. It's generally known that technologies such as Chatbots, Generative AI or Large Language Models lack the common sense and human-spark needed to navigate the physical world in which we live. So we might consider at least for now, in Q4 of 2024, AI-driven solutions a perfect productivity enabler to our daily routines and professional work lives.

To provide an example, let's say we have some customer review data on Cornish Airbnb stays. Review data, which is often regarded as ambiguous or unhelpful at times, remains important for building and maintaining brand reputation. Service providers rely heavily on star ratings(or similar) to provide them with a quantified perspective of how well, or not so well their services are doing.

Review Data at a Glance

Typically, review data consist of the service or product being reviewed, the reviewers given score and some textual justification for the scoring. It's important to note at this stage, that our aim is to highlight the level of abstract and potentially helpful information going amiss within textual data. Although our dataset consists of ~36,000 Cornish Airbnb reviews, let's take just 5 properties at random, excluding any that contain direct entities such as property or host name.

By plucking just 5 at this stage, we can see that all properties within our very small sample score highly on the given star system, it would also seem as though language being used is mostly positive with named attributes such as 'Exceptional location' and 'lovely and clean'. However, there are detractors mentioned throughout, some of which may currently be flying under the radar.

Augmenting Human Intelligence

An important consideration is that, manually reviewing vast amounts of textual data, checking for nuances in language that note specific issues is sometimes unfeasible, requiring some level of human effort and resource. However, with text classification and processing tools, we can quickly identify and summarise both positive and negative feedback. Even highly-rated properties within this space occasionally receive negative sentiment in their reviews and detecting such nuances remains key for businesses to address hidden issues and improve. See for example, how we might classify our selected reviews using a simple sentiment-based model.

Going one Step Further

Digging deeper, by using language embedding models *(similar to what is at the very core of technologies such as ChatGPT)* we can group all reviews into consistent themes based on their semantic meaning. This helped us to identify seven key positive themes from Cornish Airbnb reviews, revealing to us what guests value the most. While these exact findings might not be *too* surprising or insightful, it gives us an accurate summary of the positive experiences lived by consumers, and perhaps even provides a more accurate framework to consider when improving offerings within this space.

Figure 1: A donut chart showcasing the distribution of 7 Positive themes across ~97,000 promotors

The Other Side of the Coin

Similarly, we have grouped any negatively classified language, but this time it has has helped us uncover nine common themes which highlight areas for improvement, such as; noise, spaciousness and temperature. Again though, while we weren't exactly blown away by the most voluminous topic of this category, that being directly tied to property comfort and aesthetic appeal, we *were* surprised to see such a huge proportion (19%) of negative language used, noting property uncleanliness, yikes!

Figure 2: A donut chart showcasing the distribution of 9 Negative themes across ~4,900 detractors

Just Scratching the Surface

The techniques shown here are just the beginning of how Business Intelligence, and more specifically Natural Language Processing (NLP) can be applied to process large quantities of information and summarise insights from customer feedback. By analysing over 36,000 reviews efficiently, we save substantial amounts of time and resources that can now be invested elsewhere. But such techniques are just the beginning, by expanding these methods to larger data sets, like social media or survey responses, this offers an even broader insight into various brand perceptions, and can help to identify emerging trends. If that's not enough, businesses can go even further by developing models of their own that detect nuanced signals like ‘loyalty’ or ‘churn,’ predicting customer satisfaction or retention based on linguistic patterns. In future articles, we’ll explore Emotion Classification models and Named Entity Recognition (NER) tools that can pinpoint how customers feel about specific brands or features, whether it’s; joy, anger, disgust, fear, sadness or surprise.

Final Note

Uncovering insights from unstructured text is no longer a complicated process but rather an invaluable step toward achieving a competitive edge. With the rapid growth in Generative AI and Transformer-based models such as those that have gained notoriety in the previous two years, these very techniques are now more accessible and impactful than ever. This article offers just a glimpse into how Business Intelligence can transform large amounts of data into more concise strategies. We’ve only touched on one approach here, classifying and segmenting text data, but many more powerful tools await. Stay tuned for upcoming articles where we’ll dive deeper into BI techniques and uncover their practical applications!