Microsoft says its hands-on AI work spans nearly 500 projects and over 200 customers

Microsoft said on November 12, 2026, that its Frontier Company had engaged in nearly 500 AI projects with more than 200 customers across 28 countries. Alongside those totals, the company published customer outcomes spanning financial data preparation, telecom networks and materials research—putting specific business workflows behind its pitch for hands-on AI engineering.

The disclosure came from Rodrigo Kede Lima, president of Microsoft Frontier Company, an operating business of 6,000 industry and engineering experts. Its forward-deployed engineers work alongside customer teams to build systems for particular business processes. The examples describe engineers working with bank analysts, network specialists and researchers, rather than simply supplying access to an AI tool.

Microsoft also said most customers whose initial projects were complete had returned for additional engagements, generating hundreds of follow-on projects. Kede Lima framed those repeat engagements as evidence that customers had measured value and found another workflow worth redesigning. That is the company’s interpretation of the repeat business, alongside its disclosure of the project totals.

A bank’s time savings come with a separate quality test

Northern Trust provides a detailed example. Its onboarding agent converts clients’ historical investment transactions into the bank’s standardized format for its Front Office Solutions platform. The companies’ November 9 announcement said data preparation fell from about two and a half weeks to three days, an 83% reduction. Analysts use Excel and SharePoint to review, refine and approve the work.

The timing measurement covered real client onboarding cases after deployment. It included data transformation, analyst review, exception handling and final validation—not just the agent’s processing time. Northern Trust compared the average time for those agent-assisted production cases against the previous workflow. Human oversight remained part of the process, including checking outputs and resolving exceptions before clients went live.

The quality figure needs a different reading. Although the November 12 post called it 96.5% first-pass accuracy, the detailed announcement identifies a 96.5% F1 score, a measure combining precision and recall. It came from seven controlled test cases using anonymized historical transaction workbooks. Outputs were checked against analyst-prepared references before human corrections; it is not a general production accuracy rate.

Not every example is a finished result

Other examples in Microsoft’s disclosure describe different stages of work: a workflow under construction, research leads awaiting laboratory tests and a factory simulation supporting operators. Their reported benefits should be read at those stages, rather than treated as equivalent deployment results.

• 3M: Engineers are building an AI-driven order-management workflow for credit checks, delinquency assessments and system updates. Human approvals and a monitoring dashboard are part of the design. Microsoft describes reduced manual effort and faster cash flow as expected benefits, not measured outcomes.
• Saint-Gobain: Microsoft says its engineers assembled 62,000 patents and 44,000 research abstracts into a knowledge base in hours. Saint-Gobain estimates that work would otherwise have taken a year. The team identified cement-additive hypotheses in about six weeks; scientists are now testing them in the lab.
• Unilever: A digital twin—a simulation of a real factory—at its Pondicherry site uses live data to monitor performance, anticipate problems and guide operators through fixes. Microsoft says the system is built on NVIDIA Omniverse and hosted on Azure.

Microsoft also describes knowledge transfer as part of the engineering work. At Publicis Groupe, its teams are co-building AI processes that use business context and improve through real-world outcomes and human feedback. Kede Lima’s stated longer-term test is whether customers can run, govern and improve their own AI systems after the engagement—not only what Microsoft delivers during it.