JBS Dev: Imperfect data won’t block enterprise AI

Joe Rose, president of JBS Dev, told attendees at AI & Big Data Expo that imperfect data should not stop firms from adopting generative and agentic AI. He urged companies to prioritize cost control, portability and human review when deploying models.

Rose rejected the notion that organizations must wait for pristine data or build large data lakes before using generative AI. He called it “a common misconception that your data has to be perfect before you do any of these types of workloads” and added that “the tooling has never been better than it is now to deal with poor quality data.” He noted that large language models often interpret incomplete prompts and extract useful information from mixed-format records.

He described a healthcare client that needed to migrate billing and reconciliation to a new system while records existed as PDFs, images and inconsistent text fields. Generative models handled optical character recognition and text extraction, and agentic techniques compared customer records with insurance contracts to check billing rates. The rollout began with partial automation and human review for exceptions, increasing from about 20% automated to 40%, then 60% and higher over time.

Rose warned that model output can be unpredictable and that organizations need processes to detect and correct bad output. Human reviewers, he said, remain necessary for edge cases and ongoing quality checks.

On future priorities, he said conversations will move from pure model capability to cost and portability. He asked how models can run on smaller devices: “The last mile is how do we get these things to run on a laptop or a phone instead of having to run in a data centre?” He noted much training data already comes from large public corpuses and suggested future gains may come from optimization rather than far larger datasets.

He encouraged firms to consider building their own deployments instead of defaulting to third-party SaaS vendors. “It’s not as hard as it sounds,” he added, citing existing cloud tools from major providers that can run agentic workloads without new software licenses. JBS Dev offers services to help clients adopt those tools and scale implementations.