I dispatched “research” for this article to Claude, as is normal, to uncover links and data to support my writing. In its humble, honest and (typically) hubristic demeanor, it spat back other similar recent research, which was based on even more other “research” from earlier. It gave me links to 6 articles talking about the evolution of BI tools and the how it has been underscored primarily by low adoption rates at 29% (25%?), per Gartner.

I’ll approach this story differently. While I agree with the premise of these articles, I’ve always found that bare statistics without any real mechanistic explanation are essentially astrology. Having worked with several early stage startups that have adopted various BI tools, I want to share what I have personally watched go wrong.

At some point your business outgrows vibes based decision making. There comes a stage where you have a sizable revenue and want to deep dive into what is contributing to it positively, and what drives it the other way. Or you want to find what to do to become profitable. Or possibly something more specific to a business process.

The typical BI path looks like this:

In my last couple of engagements, it took the companies close to a year, to roll out the dashboards for the intended use, after paying for external BI implementation consultants. It took even longer to realize that most of the intended users were either unwilling or not skilled enough to create their own utility dashboards.

To compound this tragedy further, both the companies then decided to cut cost by discontinuing the BI consulting engagement and, no surprises here, transfer the burden of maintaining and building dashboards to their respective engineering teams, in addition to their core role of managing the data pipeline, and model definitions.

The final outcome somehow makes this worse. The starting point of this is usually (and it should be) the intention to optimize key business metrics, which the outcome should exactly map to. However, most BI tools end up being used like fancier versions of spreadsheets. This result is a combination of the user not knowing which threads to pull and the platform not being able to articulate, or being incapable of, converging to the intended outcome.

There has been an initial assumption, particularly in the self-serve BI community, of their platforms being easy to use. But this has largely been a result of human hallucinations about what is easy to use. The symptom here is unfortunate and the most dreaded one in tech, where learning from actual user behavior and fixing their actual problems instead of retrofitting imagined solutions, gets deprioritized over a compulsive feature rollout cycle.

The convenient solution has been to create an ecosystem of “experts” who can now help you use a product, which is supposed to help you make sense of your business. That this is also an additional revenue stream is, of course, a coincidence.

Enter AI and 2026. AI driven query building, inferences and explainability present a formidable alternative to traditional BI, especially in business segments where we can still find early adopters. A company that has invested heavily in implementing a PowerBI across their large organization, may not be very keen on switching over to a Claude wrapper with domain definitions, vibe coded over a weekend, today. However, for industries where the data exists but the friction of implementing a BI tool has kept them away, or new enterprises that are yet to make a call, any alternative that can take away or reduce the pains of money cost, time cost or opportunity cost from low adoption, would be a strong contender.

This is what we’re building at Lumen.