Boston Consultation Group: To unlock enterprise AI value, start with data you ignore

If the business building AI, some companies find the most difficult part sometimes decide what to build and how to respond different processes involved.

on VentureBeat changes 2025The quality of data and handling front and centers of companies look beyond the experimental round of AI and explore ways to overthrow agents and other applications.

Organizations faced the pain of thinking through how tech tech’s, processes and designs, said Braden Holstege, Managing Holstege, Director and Partner’s Managing Holstege Grossion Group Consultation in Boston. He added that companies should think about a range of complexities related to data exposure, per-person budgets, access to external and internal risks.

Sometimes, new solutions involve methods of using previously unused data. speaking Onstage Tuesday afternoonHolstege provides an instance of a client using multiple language models (LLMS) analyzes millions of views, product complaints – and discover information that has never been in some years ago (NLP).

“The wider lesson here is that the data is not monolithic,” Holstege said. “You have all of the transaction records in customer feedback documents to track the application made with application juice and a million data classes.”

Some of the new possibilities thanks for the development of AI-ready data, Susan Ethyling, the Senior Director of Microsoft in Azure AI.

“Once you have it, you start getting that understanding of art can,” says Ethertler. “It’s a balancing move between it and comes with a clear understanding of your tried to resolve. We seek to resolve customer experience. But you don’t always know.”

Why Ai-ready data is critical for adopting business

Ai-ready data is a critical step in adopting AI projects. In a separate Gartner surveyMore than half of the 500 midsize cios and tech leaders they expect AI infrastructure to help AI infrastructure to help AI infrastructure in the AIs to help with the AI ​​infrastructure.

That would be a slow process. Up to 2026, Gartner PROPHECY Organizations leave 60% of AI projects not supported by AI-ready data. When the research company surveyed data management leaders in summer, 63% of respondents say their organizations in place of data in place, or that they are not sure about practices.

As deployments can be more adult, it is important to consider ways to meet challenges such as AI model on Onean Bajwa, head of the Bank of America. He added that businesses do not always rush to an object to end users somewhat advanced how they think potential in chat-based applications.

“We’re all in our daily life users in chat applications there,” Sher Bajwa said. “Users have become more sophisticated. In terms of training, you do not have to push it to the end users, but also meant the elements of implementation and scale, to be challenging.”

The increasing pain and complexity of AI compute

Companies should also consider opportunities and challenges of cloud-based applications, on-prem and hybrid applications. The cloud is released AI applications Allow to try different technologies and scales in a highly motivated way, as Sher Bajwa. However, he added that companies need to take different infrastructure issues such as security and costs – and vendors like companies easier to prompt the different models and different models of deployment

Decisions around cloud providers become more complex than a few years ago, say Holstege. While new options such as neoclouds (offer GPU servers backed by GPU and Virtual Machines) with multiple clients who dwell less. But even with cheaper alternatives, Holsteege sees a trade-off with computing, cost and optimization. For example, he explained that open-source models such as Llama and Mistrinal have higher computed demands.

“Is the cost of computing important to you to get the headache to use open-source models and to migrate your data?” Holstege question. “The above front of the choices that people face today is more than three years ago.”