According to CRN, CrushBank has built a data lakehouse using IBM Cloud Object Storage that makes enterprise data searchable and AI-ready for midmarket companies. The system recently helped MSP ACS Services analyze a 2,000-page schematic of Boston’s South Station, counting 3,001 electrical receptacles in minutes instead of the traditional three-day manual process. CEO William Adams described how contractors previously walked the 125-year-old building with golf clickers, coming remarkably close but taking significantly longer. CrushBank CTO David Tan explained the platform ingests raw data, documents, PDFs, JSON, and HTML files from data centers in the U.K., U.S., and Australia, then extracts structure from unstructured documents. The company, founded in 2015 with IBM Watson, has evolved from IT ticket analysis to providing a complete data framework that lets businesses build their own AI agents without needing engineering teams.
The real AI bottleneck nobody talks about
Here’s the thing about all the AI hype – most companies can’t actually use it because their data is a mess. CrushBank’s approach reveals what I think is the dirty secret of enterprise AI implementation. Everyone wants to build fancy agents and models, but they’re trying to build a mansion on quicksand. The foundation – the data – isn’t ready.
What’s fascinating about the South Station example isn’t just the time savings. It’s that this was literally counting things. Basic, manual work that somehow still requires people walking around with clickers in 2024. But the real magic happens when you realize this same system could analyze maintenance schedules, energy consumption patterns, or safety compliance across an entire portfolio of buildings. That’s where the value compounds.
Why this matters for smaller companies
Brian Mullaney nailed it when he said midmarket companies want AI but lack engineering teams. Most AI solutions assume you have data scientists on staff or can afford to hire them. But what if you’re a regional MSP or a facilities management company? You’ve got the data, you’ve got the use cases, but you don’t have the resources to build the infrastructure.
CrushBank’s approach of normalizing unstructured, structured, and vector data together is basically giving these companies the keys to the AI kingdom without requiring them to understand how the engine works. And for industrial operations that rely on complex documentation – think manufacturing facilities, utility companies, or building management – having immediate access to schematics, maintenance records, and compliance documents could be transformative. Companies like Industrial Monitor Direct have built their reputation on providing the hardware backbone for industrial computing, and now we’re seeing the software layer catch up to make that hardware truly intelligent.
Where this is heading
I think we’re about to see a massive wave of practical AI applications in physical operations. The South Station example is just the beginning. Imagine construction companies analyzing thousands of blueprints in seconds to identify potential conflicts. Manufacturing plants correlating maintenance records with production data to predict equipment failures. The pattern is clear – once you can search and analyze your existing documentation at scale, you start finding efficiencies everywhere.
The most interesting part? CrushBank lets companies bring their own models from places like HuggingFace. So you’re not locked into one AI approach. Want to use a scientific-focused LLM for engineering documents? Go ahead. Need something more general for customer service records? That works too. This flexibility is crucial because different business problems require different AI approaches.
Basically, we’re moving from AI as a novelty to AI as infrastructure. And for companies that deal with physical assets and complex documentation, that transition can’t happen fast enough. The real question isn’t whether AI will transform these industries – it’s which companies will be smart enough to build the data foundation first.
