OCO Global’s clients rely on its know-how to help them get business done anywhere in the world. One of its service offerings is the provision of detailed company information, which its clients use to choose strategic suppliers, run due diligence on potential partners and identify prospective clients and deals. 

Back when the database underpinning this information service was relatively small—covering a few hundred thousand organizations—it was feasible for OCO Global’s in-house team to manage the data. However, when the team decided to source additional data from an external provider, it ran into a significant scaling challenge. OCO Global wanted to enrich the new data, and while this new process would only take around 10 seconds per row for a skilled employee, there were 13 million new rows of data. This massive number implied a solid 4 years of round-the-clock work.

Yeray Callero, CTO at OCO Global, explains: “To add value to the dataset, we planned to review the classification by industry sector. We wanted to understand if some organizations might fit into 2 or more categories, and if it would be appropriate to expand our standard set of industries. We judged that it would take an unacceptable amount of time to review and tag the data manually.”

Intrigued by the possibility of using gen AI to review its data and suggest different or additional industry categories, Callero and his team experimented with an OpenAI API.

“Using the OpenAI API produced good results,” says Callero. “However, from making the API call to processing the response took about 5 seconds per row, so we were still looking at 2 years of solid processing time.” 

Seeking a much faster outcome, OCO Global approached IBM and Prolifics, a global IBM Business Partner with more than 3,000 employees and a 30-year track record of success. 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. 

The wide availability of models on watsonx.ai enabled OCO Global to find the best-fit model for its use case, balancing performance, accuracy and cost. The joint team chose a third-party model that achieves strong ratings in reasoning and classification but is also relatively small and, therefore, efficient in terms of resource consumption. 

“IBM and Prolifics offered the capability and resources to achieve our business objectives in a much more manageable time: from 2 years to a couple of weeks,” says Callero.