Dive Brief:
- Most organizations in banking, insurance and government agencies believe the data feeding their AI could use improvement, and 43% think their data needs significant improvement or a complete overhaul, according to a Tuesday IDC study commissioned by AI software company Quantexa. IDC surveyed 366 technology leaders at large banks, insurers and government agencies for the study.
- Amid data quality and connectivity issues, banks and insurers are deploying agents within workflows faster than they’re able to explain the context behind agents’ reasoning, the study found. Context is the complete set of information such as instructions, rules, knowledge and memory that agents need to reason and make decisions.
- While agents are expected to take on nearly one-third of work within the next year, only about 9% of companies have broadly embedded context within decision-making and AI workflows, according to the report. “Organizations have invested in the building blocks of context, but those capabilities sit in pockets,” Stewart Bond, VP of data intelligence and integration software at IDC, said in a press release.
Dive Insight:
Data silos are hindering banks’ ability to compete with third-party tools consumers are using to navigate financial tasks. A lack of context is also hindering broader rollouts of AI capabilities.
Nearly three in four consumers using AI for financial advice and tasks are turning to OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini before seeking out their bank’s chatbot, according to McKinsey & Company’s Global Banking Annual Review 2026. Agents are already affecting money movement, with Meta’s Muse agent reading consumers’ balances and transactions through Plaid, and PayPal adding its wallet to ChatGPT’s checkout, the report found.
Although banks and insurance agencies are getting the building blocks of context in place to support agents and AI tools, only about half of organizations implement multiple context capabilities, and mostly in specific use cases rather than broadly, the Quantexa study found. Only around 17% of companies reported broad implementation of knowledge graphs — which outline reasoning paths for AI — the lowest of any context capability.
Integrating data across silos is one of the biggest challenges facing organizations when building a context layer, according to the study. Without the right data, particularly lineage and decision data, banks can’t answer how agents reached a decision, the study found.
“Lack of data and decision lineage, not model complexity, is what most limits trust in AI decisions, and a real-world relationship view is the context organizations are missing most,” Bond said.
IT executives will need to make data lineage and decision data a design requirement to improve trust and to move proven context capabilities beyond isolated use cases to broad availability, the study said. Meanwhile, business leaders should define the decisions agents will own with a human in the loop, as well as the context agents need to operate within the organization.
Vendors are zeroing in on context as a key issue for enterprises looking to scale agentic AI. During Microsoft Ignite on Tuesday, Salesforce and Microsoft launched Copilotforce to bring key Salesforce customer context and AI capabilities into the Microsoft software environment.
“AI is only as powerful as the data and context that grounds it,” Judson Althoff, CEO of Microsoft Commercial Business, said in a press release.





