From Business Intelligence to Generative AI: a retrospective view

What about if machines can generate content, like human-like responses, pictures, videos and music? Good news, we are already living that moment right now. ChatGPT and LLM are the new buzzwords in the tech world. This kind of AI (Artificial Intelligence) was released and improved since 2018, but it was not until the end of 2022, when ChatGPT gained momentum and visibility, reaching one million users just in 5 days. Now is a hot topic for companies trying to take the advantages and benefits of these technologies, providers developing solutions to speed up implementation, conferences' agenda with numerous keynotes and presentations about it.

The path to getting here was full of highs and lows, expectations fulfilled, and more than once failures. What yesterday was a solution and the new holy grail, today is obsolete and needs to evolve and improve. From the beginnings of Business Intelligence and Data Warehouse to Generate AI, the market witnessed different approaches to manage data and?analytics. Throughout the decades, companies have seen value on data to generate insights and support decision-making.

Gartner Analytics Maturity Model

Gartner, a globally well-kwon consultancy company, in 2012 defined what today still be a map or guidelines referred by companies driving the cross organisational adoption of data and analytics. Titled Analytics Maturity Model, it defines a series of stages, taking in consideration the value of a specific analytics practice delivers to companies at the same time of measuring how complex or the difficulty of implement it. This model has been used to assess the current stage where companies are, and the work that should be done to breach the gap to move to the next stage. In its essence, it determines that to deliver more business value, companies should implement more and more complex analytics practices.?

Here are the key characteristics of each stage:

Descriptive Analytics

Here are the beginnings and foundations of Analytics as a formal, enterprise function. At this stage, company's analytics capabilities are limited to simply measure business KPI (Key Performance Indicators), at the most, in real time, but is a reactive way of decision making. At the same time the difficulty of implement such practice is not high, actual value is also low. Ingesting historical structured and transactional data to generate insights, it focus at the nature of the problem or questions, but does not offer go to actions guidelines on how to improve business tasks/functions.

Diagnostic Analytics

At this stage, companies learnt how to manage data, and now are capable to understand the business behaviour using analytics. Expanding ingestion to more data sources - from both internal and external sources - it focus on detecting and analysing the root cause and reasons that led to bad or good business performance. This maturity level requires more effort, sometimes investment as well to add additional technical tools, but increases the value that is delivering. It still a reactive way of decision making, but now the business functions are able to understand things like the customer journey, purchase patterns, and uncover the drivers that are leading to those behaviours.

Predictive Analytics

As companies move forward on the data and analytics practices, they are able to increase the value delivered to the business. At this advanced stage of the Maturity Model, the main goals is to predict or anticipate future behaviours, intervening at the early stages. As the result of understand the root cause of how business operates, it is possible to extrapole this analysis by leveraging more and more data (structured and unstructured) using advanced algorithm and techniques looking for patterns to identify risks and opportunities. Most used practices are, but not limited to, Data Mining, Machine Learning, Data Science, Deep Learning, Statistical Modelling and so on. Clearly this is a transition from reactive to a proactive management approach.

Prescriptive Analytics

The keyword for this stage is actionable insights. While predictive analytics can predict future outcomes and the underling reasons, it's not sufficient to avoid or cause any impact on the business performance. Storing and analysing even more data, companies are able to implement real time actions to influence their customer behaviour and improve their experience. This actions are triggered, measured and analysed during customers interactions with the services and products provided by a company. Combining the predictions and the actionable guidelines, at this stage analytics is delivering a huge value for the company, moving forward to a real proactive way of managing the business.

Generative AI and the Maturity Model

From Descriptive to Prescriptive, the driver is always to finding answers to more complex questions. Questions are the basis of any analytics practice, and data is the raw material used to build those answers. As more advanced analytics the companies are capable to implement, more data is required to ingest, store and analyse - and also secure. In our current era, digitalisation is a widely industry initiative to build "as-a-service" business models, integrating technology across organisational functions - sales, marketing, finance, and so on - to capture all those metrics and interactions. Those touch points are generating more data than ever before.

The promise behind Generative AI is to be able to answer any type of question, on both corporate and public data, which will require handle even more data, but also, all the previous data companies have been storing for years. There is no pre-defined scope for GenAI. The only way to respond to any question is ingesting all possible data available. This is, perhaps, the next level of Big Data required to support learning and intelligence from data.

Generate AI features:

• All data (or at least all data related to a scope/topic)
• All data formats (structured, unstructured and semi-structured data)
• Internal/corporate and external/open data
• Question scope: any question
• Human-like interactions and responses

Besides the boom LLM is bringing and revolutionising the market, there's still some challenges to overcome. The main one is related to data privacy. Some of the LLM solutions requires the share the data to third-party companies - SaaS solutions -, which is not accepted by many organisations and industries. Companies are now looking for solutions to implement an Enterprise LLM approach, keeping control and sovereignty of the full process, including data and compute processing. Second is how to operate and maintain LLM solutions. LLMOps will be the next capability companies need to develop internally to operationalise the end to end LLM lifecycle.

Even tough ML/AI and now LLM solutions are still complex technologies, it is true that tech providers are making them easier to adopt. The latest progress on?frameworks, languages and tools to support analytics projects is making possible to shorten the learning curve. Pre-trained algorithm and models, MLOps, AutoML, API-ready and low-code/no-code tools are all focused on making the adoption easier and faster. This is also expected to be part of the LLM landscape.

Last but not least, is how to take advantages of LLM for the enterprise. More natural conversations, content generation (from text to images and videos) and search are some of the use cases or implementations companies are working on. In the end, Generative AI will be able to respond to a much broader type of questions and address a variety of use cases, compared to the previous stages of the Maturity Model. AI is moving to the direction to be a "co-pilot" of human intelligence, incorporated at any level of decision making.