This article was also published on LinkedIn
Data analytics performed by GenAI agents means the end of BI software as we know it.
Rapid progress in Generative AI technologies, especially large language models (LLMs) and the natural language agents that use them, has sparked a lot of talk about the end of certain human jobs, such as call center workers. But what about the end of software categories as we know them today?
Companies and the executives who run them have been attempting to gain insights into their businesses since long before computers or databases existed. The modern software category we know as “BI” (Business Intelligence) is the culmination of over 75 years of software and hardware attempts to do this job faster and more effectively through computer-based automation. Advances in processing power, storage, graphics, and especially database theory have culminated in solutions like Looker and PowerBI that are used by hundreds of thousands of executives, revenue managers, and business analysts on a daily basis. In addition, data lakes perform trillions of calculations every year as they crank out queries and reports on a recurring basis.
But despite all these advances, business analytics remains challenging. A typical executive is skilled in operating their business and reading reports, not in crafting accurate SQL queries, let alone understanding how to optimize those database queries for speed and cost efficiency. “Cube” and graphics-based systems like Looker that try to simplify BI can still be complicated to learn and use, even for business school graduates. And there’s a good reason for this – BI is a fundamentally hard problem. It’s a perfect example of a sparse solution in a large search space: Change a database query even a tiny bit and the answer probably won’t reflect what the user was trying to find out. It’s also a translation problem: regardless of how appealing a UI or templated approach might be, it’s difficult for any static approach to simultaneously offer expressive power (you can ask anything you want) and ease of use (there’s a guaranteed simple path to follow to get your answer).
Classic approaches to BI require a database expert to write the queries and prepare reports. “Self service” BI software often tries to limit the type of query (of course, attempting to select for the most common and useful ones) in order to make the problem more approachable. Some SaaS applications offer “precanned” reports that reflect the most common or typically requested kinds of analyses. But unfortunately, all of these approaches are inherently limited – sometimes, a business owner really needs to know something that can only be discovered by crafting a SQL query that isn’t already sitting in a template somewhere or produceable with a couple of mouse clicks. And that’s just fundamentally difficult…Or at least it was, until GenAI came along.
LLMs have proven to be very, very good at both understanding and generating languages. Converting a human executive’s requests for BI-style analyses into one or more SQL queries that can be run on a database or data lake is essentially a translation problem – and it turns out that LLMs are pretty good at this kind of translation (and getting astonishing better at a rapid rate). Unlike a list of precanned templates or a UI that can only generate a certain subset of queries, an LLM isn’t constrained – it can formulate an infinite variety of SQL queries to answer an infinite variety of requests, just like a highly trained database operator could do. And unlike UI gestures that don’t know what you’ve tried previously, GenAI agents have the advantage that they can understand (at least to a degree) what you’ve asked for previously, how you’d like to modify the results relative to what you’ve tried before, and even what your “usual” reports or queries are like. When attempting to answer ambiguous or only partially understood business problems, this gives GenAI a huge leg up relative to every prior attempt to build easy-to-use BI software.
There’s a more subtle advantage to GenAI as well: UI- and template-based approaches to BI generally have no way of knowing if the results are meaningful. For example, if you’re trying to figure out the average value of sales contracts in the western region of the US and the answer is usually around $100,000 but today it comes back as $0.05, existing BI tools won’t flinch…they’re not able to comprehend that that’s not a reasonable result, the way a human data operator could. But GenAI can perform similar sanity checks to a human, especially if it has historical context for the queries the business analyst has performed in the past. And unlike conventional BI tools, GenAI based ones can apply intelligent analysis 24×7, alerting humans or even taking action themselves if they spot unusual or concerning trends in the data they’re evaluating.
The Future of BI is ABI
Template- and UI-driven BI won’t necessarily disappear completely, but it will almost certainly be supplanted for most of us by AI-driven solutions that offer a conversational approach to generating both one-off and recurring business reports.
Even more exciting, these capabilities will no longer be just the province of specialized BI software or services – because the marginal cost of attaching an LLM to an existing solution is rapidly dropping to zero, every SaaS application and even internal apps that utilize a database or data lake at any level will likely start to include the ability to generate arbitrarily complex reports through human language commands. This is already happening, and improvements in LLM-generated SQL and training for this specific purpose will undoubtedly accelerate these outcomes in the very near future. Specialized cap table reports in Carta, candidate reports in Greenhouse, payment analyses in Stripe…there’s really no limit to where and how ABI (and custom report generation in general) will become available to every user of enterprise applications. Even consumer experiences, such as banking, will benefit from ABI capabilities that rival the top of the line enterprise BI tools today, because it will be nearly free to offer these services.
Under the hood: MCP as an ABI Enabler
Part of what will power the ABI revolution is a new flavor of middleware designed to give AI agents better access to business data, known as Model Context Protocol, or MCP. Originally designed and open sourced by Anthropic, MCP helps AI agents access enterprise systems, including SaaS apps, operational databases, data lakes, and more. While originally designed to make real-time information more readily available, MCP is just as good at exposing tables in data lakes – exactly the sort of information historically used by BI software to formulate reports. Operational data can also be exposed and combined, leading more real-time insights.
What’s especially exciting is that AI agents can increasingly also discover patterns and insights, just like a human being, and report those back to busy executives. So, not only are business analytics getting easier, the more tedious and standard of those analyses are going to increasingly be delegated to AI agents to monitor, with executives only needing to take a look at unusual patterns or novel information that the agents discover.
Importantly, MCP isn’t just a one-way, “read only” path – it also potentially allows AI agents to connect to operational systems to take action as well. Unlike conventional BI, ABI won’t be limited just to “reporting the news”, it will be able to do something about it (under human control and guidance, of course). Rebalancing sales teams across geographies, implementing dynamic pricing, formulating and executing churn reduction strategies – there’s really no limit to what AI agents will ultimately be able to help a business with, given their increasingly detailed understanding of the business and their ability to process vast amounts of BI intelligence, not just once but on a continual, 24×7 basis.
Conclusion
Our world is changing rapidly thanks to AI, and no corner of the software world will be the same again. The real winner in the ABI revolution will be the busy executives and business analysts who no longer need to think about how (or how quickly) they can get the right report to gain insights and take appropriate actions. Instead, they will delegate routine BI activities to AI agents, generate and receive custom reports from their ABI-enabled services and applications, and focus their precious time on LLM-provided insights and outcomes that surpass even today’s most sophisticated BI software!