In just a few months, the conversation about data has changed. ChatGPT launched on 30 November 2022 and, this very week, GPT-4 has arrived (14 March 2023). What looked a few months ago like a laboratory experiment is now in the hands of millions of people, and the world of Business Intelligence will not stay on the sidelines. It is worth looking calmly at what all this means for data analysis.
The turning point
The novelty is not artificial intelligence itself —we have been using predictive models in BI projects for years— but generative AI: models able to understand a question in natural language and produce coherent text, code or explanations from it.
The jump from ChatGPT to GPT-4 in such a short time gives a sense of the pace. And although today these tools live mostly inside a general-purpose chat, it is reasonable to expect that same capability to reach the data platforms we use every day before long. It is worth understanding the ground before the wave arrives.
What generative AI promises for BI
Applied to data analysis, the potential is clear and moves along three fronts:
- Asking in natural language. Instead of navigating a dashboard, a business user could type “how did sales evolve by region last quarter?” and get a direct answer.
- Summarising reports. Turning a dense dashboard into a paragraph that explains, in plain text, what is happening and what deserves attention.
- Generating queries and code. Helping the analyst write SQL, calculation expressions or data transformations from a description.
Taken together, the promise is to bring data closer to the people making decisions and to free the analyst from part of the mechanical work.
What is realistic today and what is hype
It is worth being honest. Today, in March 2023, these capabilities are very promising but still nascent inside BI tools. A generative model writes fluently, but it does not know your business: it does not know what “active customer” means in your company, nor which table holds the truth about revenue.
That is why the hype needs filtering. Generative AI does not replace a data strategy: it amplifies it, for better and for worse. On tidy data, it speeds up analysis; on chaotic data, it speeds up mistakes.
The risks not to ignore
Before running, you have to understand the limits:
- Hallucinations. These models can produce answers that sound confident but are wrong. In a business report, an invented figure is worse than a gap.
- Governance. Who validates what the model answers? Which definition of each metric does it work with? Without governance, every user can get a different number.
- Security and privacy. Sending sensitive data to an external service carries legal and confidentiality implications that must be resolved before using it in production.
None of these risks cancels the opportunity, but they do demand that it be approached with judgement.
How to prepare right now
The best news is that what you need to ride this wave is the same thing that already underpins a good BI project:
- Tidy your data. Reliable, clean, accessible sources. A generative model does not fix bad data.
- Define your semantics. A single, clear, shared meaning for each key metric. It is the foundation that makes any answer —human or automated— consistent.
- Document and govern. Clear names, correct relationships and well-defined permissions. The better described your model is, the better any future assistant can lean on it.
Put another way: preparing for generative AI starts with getting the data fundamentals right. Whoever has that foundation in order will be ready to bring in these capabilities as soon as they mature; whoever does not will have to start there anyway.
At Digital Fox Data we help companies build that solid foundation with our Business Intelligence consulting and choose the right data tools, so that the arrival of generative AI adds value rather than noise. If you want to prepare your data strategy for what is coming, let’s talk.