‘The data lake is like the SQL database, in my opinion. We’re throwing everything into it, right? And CrushBank is the wrapper that’s making it searchable for us,’ says William Adams, CEO of MSP ACS Services.
Earlier this year, William Adams fed a 2,000-page schematic of Boston’s famed South Station into a data lake created by CrushBank to count electrical receptacles for a customer.
Ingesting and reading the thousands pages of diagrams and unstructured data in minutes, the data lake provided an answer that previously took humans three days of walking the floors of the 125-year-old neoclassical building, said Adams, CEO of Boston-based MSP ACS Services.
“They walk around with a clicker like you play golf and they walk around with that around the building, and they count, OK ,one, two, three, four, five, six, seven and then they go back at the end and they say, ‘OK, 3,000.”
It was 3,001, according to the plans.
“The guy was saying, ‘Look how close we were,’” Adams said of the conversation he had with the contractor. “I’m saying, ‘I get it. But that’s one minute. It takes your guys three days.’ We scan it and we ingest it into the data lake, and then we search it through CrushBank and put the symbol in and say, ‘Search for this.’”
CrushBank CTO David Tan said when it comes to returning value from AI in the SMB and midmarket, the challenge most users face is getting a handle on the data they have, normalizing it, applying the correct permissions, getting it into a data lake and making it available in a logical format.
“We get you to that point if people want to bring their own agents, bring their own models, they can, absolutely,” he said. “If someone wants to just go to HuggingFace and find a scientific-focused LLM and tie it to their data, we can just tie it right into their instance [so] that LLM will have access to this data. And essentially, the way it works is when you ask a question, we do the retrieval, we find the documents, we shove the documents, or the chunks, along with that question into the context window of the LLM, and it generates the answer. Bottom line: It’s just automating and streamlining the process of getting access to that data.”
Tan said the company can ingest an organization’s files into IBM’s Cloud Object Storage—with data centers in the U.K., U.S. and Australia—including raw data, documents, PDFs, JSON, HTML files, all of the enterprise data that a company uses.
“The most interesting thing that IBM enables for us is that we can take structure out of those unstructured documents and create a structured table as well,” Tan said. “Let’s say you have a bunch of invoices or purchase orders; we’ll ingest those. We’ll store them in object storage. We’ll store them in the vector database, but we’ll also read the documents and pull the structure out of them.”
CrushBank CRO Brian Mullaney said the data lake the company has created is built specifically to address the challenges it saw with midmarket companies that wanted to buy into AI but lacked an engineering team to build the systems needed to make the data performant.
“So what we’ve done with IBM is normalize that whole thing. So essentially it’s unstructured, structured and vectored [data] all in together, all normalized that you can then build real agents off of,” he told CRN. “So we’ve got a very small hit list of agents that are kind of cool that we built for the MSP marketplace. But the broader story is the data is there. The data is normalized. The data is secured.”
CrushBank started in 2015, building a machine learning model using IBM Watson that ingested and understood type and subtype tags in IT tickets used by the popular IT service management tools like ConnectWise, Autotask and ServiceNow. It could use those tags to find similar closed tickets inside an MSP’s ticket library. This way when a new ticket arrived, the system read it and directed the Level One tech to a previous solution.
The early work in AI has paid off for the Syosset, N.Y.-based company, which has parlayed those lessons into providing a return on AI investment with its data lake framework.
“If you get this framework, this data lakehouse for midsize businesses in place, then you can actually do the work that’s fun, where you’re building agents and such,” Mullaney said.
ACS Services’ Adams told CRN the early work with the electricians working on South Station has paid off and has led to more engagements throughout its larger corporate network. He said the data lake alone would not be valuable if it wasn’t for the integration and inference layer that CrushBank built on top of it.
“The data lake is like the SQL database, in my opinion,” he said. “We’re throwing everything into it, right? And CrushBank is the wrapper that’s making it searchable for us.”