Are we replacing “Excel Hell” with “AI Hell”? 

For decades, our industry has been plagued by the phenomena of 3 different reports having 4 different occupancies (or whatever metrics the report contains). This is the natural and logical consequence of analytical processes where multiple reports are downloaded from the system(s) of record (often the PMS) and then stitched together with VLOOKUP or INDEX/MATCH statements. Each analyst may have their own definition, e.g. whether a down unit counts as inventory in the denominator or not (heck, the same analyst could inadvertently use different definitions in different reports). 

At REBA, we referred to this as “Excel Hell,” and we committed ourselves to solving that problem. We built a single, governed semantic data model that makes it easy for users of our BI “platform as a service” to access validated metrics and dimensions. We replaced the frustrations and distrust that come from different values for the same metric with a new world where all dashboards and reports have the same occupancy number, no matter who created them. Our official mission statement is “We’re on a mission to change how multifamily housing uses data,” but our unofficial mission statement may be even more apropos, “We’re on a mission to eliminate VLOOKUP and INDEX/MATCH statements!” 😊 

I now fear the siren call of “AI DIY BI” (how’s that for a set of alphabet soup?!) is drawing many in the industry to the rocky shores of what I can only refer to as “AI Hell.” And in many ways, it could be worse. If Excel Hell has three reports with four different occupancies, AI Hell has eight agents with ten different occupancies. And the same agent could report different occupancies at different times! 

I get the allure of AI DIY BI. Claude can do some amazing things, and we all want to be judicious in our vendor spend. However powerful AI may seem, it doesn’t change the laws of physics…or the Gartner Hype Cycle. The latter guarantees that all tech gets to the “Peak of Inflated Expectations” before eventually finding its “Plateau of Productivity,” and the former means AI doesn’t magically make dirty data clean or bring context to raw data. 

That’s why every article/blog from McKinsey, Harvard Business Review, Gartner and others talks about how good AI projects are all built on a foundation of clean, governed data. It’s why, in a recent Walker Webcast, McKinsey’s head of global real estate answered his own question, “Should I build, buy or partner?” with “Yes, all three!” 

And we’re already beginning to see real-world experience surfacing this very issue. A few weeks ago, I mentioned one owner/operator reporting that turning their asset managers loose on enterprise Claude just resulted in multiple agents with multiple occupancies (and other metrics), none of which matched their governed BI data warehouse. After years of work to go from “Excel Hell” to governed reporting, they dropped into AI Hell in days! 

Let’s not fall into that trap! Sure, AI is powerful; but it doesn’t render moot the GIGO principle (“garbage in, garbage out”). And why would we want to use our limited IT and data science resources to solve problems that have already been solved. We can avoid AI Hell the same way we got away from Excel Hell—with the great foundation of a governed, large semantic data model. 

We have the power…we just need to apply the wisdom of past lessons!