Northern Trust’s AI agent cuts client data preparation by 83%

By Rebecca Stone

Northern Trust and Microsoft said on November 9, 2026 that a jointly developed client-onboarding agent reduced data-preparation time by 83%, from about two and a half weeks to three days. The companies said the deployment begins an expanded AI collaboration and serves clients on Northern Trust Asset Servicing’s Front Office Solutions platform.

The agent takes historical investment and transaction records spanning years and reformats them into Northern Trust’s standard data structure for reporting, the companies said.

Front Office Solutions supports investment-performance reporting and analytics for pension schemes, endowments, foundations and large family offices, Northern Trust said. Onboarding can require implementation teams to consolidate historical investment and transaction records from several sources. Alternative assets add complexity, the companies said.

Those institutional histories can span multiple record sets and years. Northern Trust said implementation teams must consolidate the inputs before rebuilding a client’s performance history. The agent standardizes those inputs for reporting, while the production measurement records how long the surrounding review and validation work takes.

The companies framed the deployment around a practical financial-services test: whether AI can shorten onboarding while preserving the oversight, data stewardship and analyst validation required for institutional records.

Northern Trust Chief Operating Officer Pete Cherecwich said the agent reduces the time needed to prepare and validate client data “while maintaining the oversight and controls our clients expect,” creating capacity for both sides to focus on higher-value work.

Under Northern Trust’s broader augmented-intelligence approach, the company said AI extends human expertise while judgment and accountability remain with people.

How did the companies measure the 83% saving?

Northern Trust measured the time saving in real client-onboarding scenarios after deployment. The company compared the previous workflow’s baseline with the average completion time for agent-assisted production cases. According to Northern Trust and Microsoft, the analysis covered data transformation, analyst review, exception handling and final validation.

The companies stressed that the production measurement included surrounding analyst work rather than only the automated conversion step. The reported three-day result therefore reflects the full workflow, not a machine running alone. Northern Trust said the agent handles data conversion while implementation staff review, correct and validate the output.

The production comparison used two reference points: the previous workflow’s baseline and the average completion time for agent-assisted production cases. Northern Trust and Microsoft said each assisted case still moved through data transformation, analyst review, exception handling and final validation. The average therefore captures the time required to reach a reviewed and validated output, not just the conversion task.

According to the companies’ account, that end-to-end comparison gives the 83% figure more operational context than a conversion-only benchmark. It also separates the production timing from a controlled test of first-pass output quality. The quality test used historical client material; the timing result came from real onboarding cases after deployment.

Northern Trust and Microsoft separated the 83% timing result from the 96.5% F1 result. The timing figure covers the full production path; the quality figure records the first generated pass in a controlled test.

How accurate was the first-pass output?

Northern Trust and Microsoft reported a 96.5% F1 first-pass quality score. The companies described F1 as a metric combining precision and recall. They recorded the score before analysts gave feedback or made corrections, so it measured the agent’s initial output rather than the quality after human review.

The test set contained seven representative cases drawn from real historical client transaction workbooks. Northern Trust supplied those workbooks in anonymized form. The companies said the test measured performance against historical client work.

Northern Trust withheld analyst-prepared reference outputs during execution and compared them with the agent-generated values. The companies said this distinction kept the controlled quality test separate from the production timing measurement, which includes analyst review and validation.

The seven-case test therefore measures first-pass agreement with historical reference outputs, not the full turnaround of an onboarding file. Analyst corrections could improve the final file, but the companies excluded those corrections from the 96.5% score.

Microsoft’s optimization used only anonymized data, giving the engineering team a way to improve evaluations without bringing Northern Trust’s live client records into the process. Northern Trust continued to govern its client data while the companies measured F1 first-pass quality and other evaluation results.

How does Microsoft Azure fit into the workflow?

Northern Trust developed the system with Microsoft Frontier Company Forward Deployed Engineering teams, the companies said. Microsoft Foundry supplies the enterprise AI tooling, while Microsoft Azure provides the underlying platform. The companies said the deployment extends an existing client workflow on Northern Trust Asset Servicing’s Front Office Solutions platform.

Integrations with Excel, SharePoint and other Microsoft 365 tools let analysts inspect, adjust and approve outputs, the companies said. Azure also supports a governed intelligence layer that retains approved onboarding knowledge. Northern Trust said the layer lets the system draw on institutional expertise over time within the governed environment.

The separation between system tooling and human approval runs through the workflow. Microsoft Foundry and Azure run and connect the agent, while Excel, SharePoint and Microsoft 365 place its output in tools analysts already use. Northern Trust and Microsoft said inspection, adjustment and approval remain explicit workflow steps.

Where does human control remain?

Human oversight continues throughout the workflow, the companies said. Analysts can evaluate the agent’s reasoning, resolve exceptions and provide expert feedback. During implementation, before a client goes live, they can use the agent to identify and reconcile errors while validating starting data. Northern Trust said this lets teams check both incoming history and the agent’s output.

The companies described a specific division of responsibility: AI handles data conversion, analysts manage exceptions and validation, and Northern Trust retains control of client information. That model keeps final judgment and accountability with people, the companies said.

Microsoft’s development process used what the companies call an “eyes-off” approach. Microsoft optimized F1 first-pass quality and other evaluation measurements using only anonymized data. Northern Trust continued to govern its client data throughout that work.

The companies said the arrangement supported financial-services privacy, security and data-stewardship requirements while allowing the joint engineering team to improve the system. Under that setup, Microsoft worked from anonymized material while Northern Trust retained governance of client information and analysts retained approval of the output.

Northern Trust said the measured cycle cut roughly two and a half weeks of data handling to three days. The companies said the agent takes over conversion so analysts can focus on exceptions, validation and higher-value work.

Can the results hold across more clients?

The companies’ open operational question is whether the 83% time saving and 96.5% F1 score remain consistent across varied client histories under the same controls. More pension schemes, endowments, foundations and family offices could test the workflow with different mixes of records and alternative assets.

The next test is whether production performance stays stable without relaxing Northern Trust’s data governance or analyst approval.