How to apply AI for Business Intelligence (an ideal first GenAI use case).

The realm of business Intelligence (BI) offers a prime opportunity to start applying Generative AI to exploit the value of your data and augment intelligent insights. CIOs and Data Leaders looking for low-hanging fruit in applied AI should consider starting with this value-driven use case for Large Language Models. Aiimi CTO Paul Maker explains how to get started and prove ROI.

We all know business intelligence reporting is only as good as the breadth and depth of data you can feed it. Maybe there’s some insights you know you don’t know about; certainly, there are others you don’t even realise you’re not getting.

BI dashboards thrive on structured data, but the typical enterprise is made up of up to 90% unstructured content and information. The benefits of unstructured data discovery are well documented; but how can we turn more of this unstructured content into the structured formats our data-hungry BI reporting tools need?

This is where Generative AI and Large Language Models (LLMs) come in. In this post, I’ll be explaining exactly why they offer such great potential, and ROI, with very little downside. If there’s an ideal ‘let’s get started with generative AI in business’ use case, this might just be it.

Generative AI for Business Intelligence – How does it work?

Generative AI, as its name suggests, is excellent at creating and understanding text. With careful prompt engineering, it’s relatively easy to get sophisticated outputs from an AI model. This presents a great opportunity for turning unstructured information that’s not at all BI-friendly, into structured data that can be fed into your business reporting.

Most Business Intelligence tools can’t handle unstructured data directly. Without a complementary solution, businesses are missing out on understanding their unstructured content and, most importantly, bringing together the worlds of structured data and unstructured content to gain more complete insights derived from their entire data picture – not just a structured slice of it.

AI can make the task of ‘structuring the unstructured’ faster, simpler, and more cost effective. The approach is simple – by prompting a large language model to turn unstructured data into a structured data format (like JSON, a standard text-based format), you can quickly and accurately create a machine-readable output which can be fed into downstream BI tasks.

Of course, the key here is accurate information retrieval to first find the relevant unstructured data. If we can’t find the right data, we can’t pass it to an AI model to work its magic.

Creating structured data for BI using Large Language Models

Here’s an example to bring this concept of AI for business intelligence to life.

Take a typical event experienced by a water company, one that’s rich with data from both structured and unstructured sources – like telemetry from sensors on the network, calls and messages from customers, responses from service agents, reports from on-site engineers. How can we use AI to give our BI reporting tools the most complete insights into what happened? For arguments sake, we’ll assume that all the relevant data is discoverable via our search or information retrieval solution – whether that’s an Insight Engine or something else.

Large Language Models are exceptionally good at understanding text and extracting facts, so if you point your AI model at a set of retrieved data and provide it with a prompt, such as “Tell me the main causes of delay that have occurred when responding to this event”, it will return a text summary. But we can then prompt the model further to transform this into the structured format we need. Finally, provide the model with an example of the JSON structure we want, in this case a list containing objects like ‘Event Type’, ‘Event Description’ and ‘Date/Time’, and then simply ask it to “Complete the following JSON”.

In a matter of minutes, you have a richer set of structured data ready to plug in to your BI system and feed into your downstream reporting tasks.

This can be applied across event management data, customer records – any unstructured areas of your data landscape you need to gain insight into and report on.

Why is AI for business intelligence such a valuable AI use case?

Using generative AI to augment your BI reporting is a powerful and readily measurable use case for AI in business. It’s a clear-cut thing to associate positive return on investment with improved business intelligence and the impact this has on decision making. Whereas productivity-related use cases – like AI co-pilots attached to your office applications – can be difficult to measure and take time to reveal their impact, applied AI for business intelligence offers a straightforward ROI which leaders can build a business case around.

We’re seeing this challenge play out for senior data and AI leaders through our own ongoing research into AI readiness – over 43% of leaders we surveyed don’t know how to measure the ROI associated with their AI activities. And almost half of leaders don’t know how much their organisation actually plans to invest in AI in 2024. When cost is so intangible, and ROI is key to securing future budget, smart leaders need to select use cases that are controlled and measurable.

Plus, there are some important practical considerations associated with generative AI – like cost, security, and toxicity – that compare favourably when we consider AI for business intelligence over other enterprise AI use cases. With a vast array of available AI models at different sizes and compute requirements, marrying up the right model with the right use case is key to managing your AI deployment costs.

With AI for BI reporting, you don’t need to use huge models – options like Llama 70B are more than capable and can be run privately to give you enterprise-grade security. With BI-related tasks, we can even run this work at scale overnight, potentially reducing the costs associated. Toxicity and copyright issues are also of minimal concern as the AI model is tightly constrained by prompts to only retrieve facts from your defined set of information.

So, AI for business intelligence is a great starting point for leaders who want to show tangible returns on their investment in Generative AI use cases. What are you waiting for?

About the author:

I’m Paul Maker, CTO at AI-led data insights company Aiimi. I lead Aiimi’s R&D team, with a particular focus on AI and the capabilities of new and emerging technologies. Follow me on LinkedIn where you’ll find my weekly videos on AI for business and data leaders.