OCO Global needed to enrich and reclassify a rapidly expanded dataset of company information for supplier selection, due diligence, and prospecting, but manually reviewing 13 million new rows would have taken roughly four years and an earlier OpenAI-based approach was still too slow.
Reported impact
The watsonx.ai workflow reduced processing from years to weeks, delivered much faster row-by-row enrichment than the earlier API design, and created a reusable platform OCO Global can extend to other internal classification and signal-detection use cases.
Working with Prolifics and IBM Client Engineering, OCO Global built a right-sized generative AI classification pipeline on IBM watsonx.ai instead of relying on a larger, slower API workflow. The team starts by selecting and cleaning subsets of company data, then passes those rows through watsonx.ai so a smaller third-party reasoning model can review industry assignments, suggest additional categories, and return structured results for downstream processing. IBM and Prolifics tuned the approach around latency, accuracy, and cost, ultimately cutting round-trip performance to about 0.1 seconds per row. OCO Global is using the implementation as an MVP platform that it can iteratively refine over a year-long classification program and reuse for adjacent workloads such as internal file classification and market-signal analysis, giving the business a reusable gen-AI foundation rather than a one-off script.
OCO Global selects and cleans batches of company records, sends each row through watsonx.ai for industry reasoning and category suggestions, and then processes the returned classifications back into its company-information dataset for supplier, due-diligence, and prospecting use cases.
“IBM Client Engineering and technical specialists worked with Prolifics to design a solution based on the IBM watsonx.ai AI studio that offers round-trip performance of 0.1s per row—50 times faster than the previous approach.”
“Today, we have a working platform for building gen-AI-based solutions.”
“Using the OpenAI API produced good results ... [but] we were still looking at 2 years of solid processing time.”
“for 250,000 records, it takes 4 hours to select and clean data, pass it through watsonx.ai for classification and process the results.”
“there were 13 million new rows of data”
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
IBM Sample Sep 24, 2026 StarLink StarLink needed to eliminate manual effort in procurement processes such as email intake, purchase-order validation, document extraction, and quote comparison without losing governance or visibility across enterprise… 11,500 hours projected annual time savings across purchase order and quote processes Automation Anywhere · Docling · IBM Bob IBM Sample Dec 10, 2025 Tech Mahindra Tech Mahindra needed a way to extract value from long, complex, unstructured enterprise documents while also proving that generative-AI outputs could be governed, explained, and trusted in regulated customer… 30% productivity improvement IBM Deep Search · IBM Granite 13b · IBM watsonx.ai IBM Sample Dec 9, 2025 LDS Łazewski Depo & Partners LDS needed to reduce the time, cost and risk involved in manual patent searches and drafting work so its attorneys could protect clients' intellectual property more efficiently without sacrificing accuracy or data… over 40 IP professionals IBM watsonx.ai 