Enterprise software has a clutter problem, and a new academic paper wants to put a number on it.
Researchers introduce two linked ideas: Enterprise Representation Simplification (ERS), a method for stripping unnecessary complexity out of how companies structure their data and systems without losing anything actually needed, and Enterprise Representation Complexity (ERC), a way to measure and compare how complex different setups are. ERC breaks that complexity into four parts: the objects a system tracks, how those objects interact, how they behave, and the sources feeding them. The paper also argues complexity isn't free - it creates ongoing costs for maintenance, governance, and change, on top of whatever it costs to simplify things in the first place.
The real pitch is for AI. The paper's second claim is that less structural clutter means an AI system has less to identify, relate, and interpret before it can answer a question, and it cites text-to-SQL research showing that a simpler database structure can make AI more accurate at querying it. That tracks with a common complaint about enterprise AI rollouts: the model usually isn't the bottleneck, the pile of accumulated internal systems it has to make sense of is.
This is a measurement framework, not a product, and the authors are careful to note ERC measures architecture, not performance or cost outright. Still, it's a useful corrective to vendors selling AI as a drop-in fix for data nobody has bothered to clean up.