Evolving BI for AI

A few surprises along the way

A big realization this week is that Generative AI in CX is not as big of a “big data” problem as I expected. We can now create contextual data for just about anything you want to know about an interaction, and even with automating a full QA form, it’s topping out at around 80 elements per call. So maybe you double your call detail table. Big, but not massive.

Even the “marketing” leaders touting 100M calls per month would not be scarily huge, as they’re mostly generating a fixed amount of insights per call. Existing cloud architectures can easily handle this.

The challenge I’m still bemused by is how to get all this new “any structured” data translated into insights on the fly. Giving a user the ability to generate anything they want and output it into any structure means that someone on the other side of that is hand crafting that output back into a data model that can support combining it with everything else for operational analysis.

I was originally going to hire Analytics Engineers to hand roll all these charts and dashboards, but my younger self reminded me that we want the frontline team to get the dashboards they want without having to call in the geek squad or having customers pay for a lot of professional services.

Therefore, we have no choice but reinvent BI for the world of Gen AI :) It turns out that we can use the power these LLMs to help us with that, but governance and multi-tenancy is still a mountain.

I think we’ll all be surprised with what we come out with…