Northern Trust AI Agent Cuts Client Onboarding Time by 83%

By Olivia Hart

Northern Trust and Microsoft said on November 9, 2026, that their jointly developed AI agent cut client data-preparation time by 83%, from about 2.5 weeks to three days. The deployment expands the two firms’ AI collaboration and serves clients on Northern Trust Asset Servicing’s Front Office Solutions platform.

Institutional clients, including pension schemes, endowments, foundations and large family offices, use Front Office Solutions for investment-performance reporting and analytics. The agent converts historical investment and transaction records spanning years into Northern Trust’s standard data structure. Previously, implementation teams had to consolidate records from several sources to reconstruct a new client’s performance history; the companies said alternative assets added complexity.

How did Northern Trust measure the time savings?

Northern Trust measured the reduction in real client-onboarding scenarios, comparing the previous workflow’s baseline with the average completion time for agent-assisted production cases. The calculation covered data transformation, analyst review, exception handling and final validation. It therefore measured the surrounding human work, not just the automated conversion.

Because the calculation includes analyst review, exceptions and final validation, the 83% figure is an end-to-end result rather than a conversion-only benchmark. Northern Trust and Microsoft did not disclose how many cases entered the average, how much time each step consumed or whether every client type achieved the same reduction.

What did the quality test actually establish?

The companies reported a 96.5% F1 first-pass quality score, using a metric that combines precision and recall. Their test set contained seven representative cases drawn from real historical client transaction workbooks that Northern Trust supplied in anonymized form.

Northern Trust compared the agent-generated values with analyst-prepared reference outputs withheld during execution. The result was recorded before analyst feedback or corrections, so it measures initial output rather than quality after human review. The controlled test thus cannot establish the final error rate after intervention; the separate production timing does include analyst review and validation.

Where does human control enter?

Northern Trust developed the system with Microsoft Frontier Company Forward Deployed Engineering teams. Microsoft Foundry supplies the enterprise AI tooling, with Microsoft Azure providing the underlying platform. Integrations with Excel, SharePoint and other Microsoft 365 tools let analysts inspect, adjust and approve outputs.

Azure also supports a governed intelligence layer that retains approved onboarding knowledge so the system can draw on institutional expertise over time. Analysts can evaluate the agent’s reasoning, resolve exceptions and provide expert feedback. Before a client goes live, they can use the agent to identify and reconcile errors while validating the client’s starting data.

That workflow checks both the incoming history and the output produced from it. It also keeps final judgment and accountability with people, matching Northern Trust’s broader augmented-intelligence approach, in which AI extends human expertise rather than replacing it.

How did the partners protect client information?

Microsoft’s development process used what the companies call an “eyes-off” approach. The team optimized F1 first-pass quality and other evaluations using only anonymized data, while Northern Trust client data remained governed under the firm’s control. The companies said the arrangement was designed around financial-services privacy, security and data-stewardship requirements while the joint engineering team improved the system.

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.” Cherecwich said the time savings would create capacity on both sides to focus on higher-value work. Under the firm’s stated model, judgment and accountability remain with people.

What remains unproven?

The deployment assigns AI to data conversion, analysts to exceptions and validation, and Northern Trust to control of client information. The open operational question is whether the measured savings and first-pass quality hold across varied client histories under the same controls, especially where alternative assets complicate the records.

A seven-case quality test is narrower than the production workflow, and the 83% reduction remains an average without a disclosed case count. For broader use, the central test is whether those savings and quality levels persist across more client histories without weakening analyst control.