
A contract arrives. Someone has to turn its contents into usable data, check the details and get the work to the right place. At Premera Blue Cross, one such workflow used to occupy 30 to 45 minutes. The insurer says it now takes about three. An afternoon has the same number of hours; considerably more of them become available for something else.
The result appears in Microsoft’s April 17, 2026 customer story about Premera’s use of Microsoft 365 Copilot and Copilot Studio. It is a neatly bounded claim: a particular contract workflow, a before-and-after time, an approximate finish. Give an AI project a stopwatch and the conversation becomes pleasantly specific.
At those reported timings, the reduction works out to roughly 90–93%. That is arithmetic from the stated range, rather than a separately measured company-wide productivity result. A business case grows stronger when its numbers know their place. Three minutes is impressive enough without asking it to stand for every process in the insurer.
Less time at the desk
Bars use the 45-minute upper baseline. Approximately 90–93% less time, calculated from Premera’s reported range. Source: Microsoft customer story.The person stays in the process
The contract exhibit agent tackles manual entry across systems. It reads incoming documents through intelligent document processing, offers extracted details for a person to verify, then sends the work to the appropriate queue. The machine prepares; a human checks. That division of labour explains more than a heroic claim about automation ever could.
Consider where the value sits. Repeatedly moving information consumes attention before judgment even begins. Removing some of that copying gives the reviewer a different job: deciding whether the proposed details are right. A fast extraction becomes useful when the next person can inspect it, correct it and carry the task forward.
For a buyer studying this example, the natural follow-up is to time the complete workflow, including review and exceptions. How often does a document need correction? What happens when its format changes? Those are questions this account prompts, rather than results it supplies. The elegant demo is the beginning of the inquiry. The ordinary, awkward document is where the inquiry earns its keep.
Nine weeks to a first act
Premera brought together a pod spanning product, platform, cybersecurity and software development. Initial agents took nine weeks to build. Training accompanied the effort. The arrangement put people who understood the work beside people who understood the platform and its boundaries.
There is a practical appeal to that small cast. Product can describe the nuisance. Platform can identify the available machinery. Cybersecurity can examine access. Development can help the pieces behave together. Each has a reason to be in the room before a prototype becomes somebody else’s daily dependency.
The reported count eventually reached 900 agents built. That figure describes the scale of creation; it does not establish that 900 agents are active, equally useful or deployed throughout the business. A catalogue can fill faster than an organisation learns what deserves a permanent place in it.
Here the editorial lesson is straightforward: creation and stewardship need different scorecards. An agent count shows participation. Repeated use, dependable output and a named person responsible for maintenance would show something else. The next chapter of a large agent programme should make those distinctions visible.
A key is still a key
Premera’s account says existing Microsoft 365 permissions constrained its early agents. Carolyn Schauble, Senior Manager of Product Management, puts the principle crisply: “Your right to your documents follows your footprint.” An agent does not need a grand tour of the organisation to do a modest job well.
The distinction becomes clearer in Microsoft’s Agent Builder documentation. Sharing an agent gives people access to that agent. The documentation separately addresses permission to its underlying SharePoint files and folders, which may require action by file owners or a site administrator. A useful interface and permission to the material behind it are two separate matters.
That separation should shape an access review. Start with a user, identify the task and examine the documents the task needs. Then test the experience with that user’s actual permissions. A demonstration performed by someone with broad access can conceal precisely the problem a colleague will encounter on Monday morning.
Microsoft’s security guidance also addresses overshared information: administrators can assess sharing risks and restrict access while fixing them. Existing permissions are a foundation to inspect, not a reason to stop inspecting. If a door was left open years ago, a faster route to the door does not close it.

Photo: Microsoft customer story
“Your right to your documents follows your footprint.”CAROLYN SCHAUBLE · MICROSOFT CUSTOMER STORY
The rules need a working address
Premera paired its training programme with a Data, AI and Ethics Committee. For a regulated health insurer, that gives questions about acceptable use a place to land. Dr. Adrian Mayers, Vice President, IT and Chief Information Security Officer, states the boundary in five words: “There is no shadow AI.”
The setting explains the seriousness. Premera is headquartered in Mountlake Terrace, Washington. Its published confidentiality policy covers members’ protected personal, health and financial information and prohibits unauthorised disclosure. A tool that makes work easier still operates inside those responsibilities.
At scale, policy needs practical controls. Microsoft’s agent-management documentation describes administrator control over agent availability, including Copilot Studio agents shared with specific users or submitted to an organisational catalogue for approval. Its Agent Builder guidance describes restrictions on organisation-wide sharing. These are platform capabilities, not a list of controls confirmed as deployed by Premera.
For an organisation with hundreds of creations, the inference is plain: somebody needs to know who owns each useful agent, who can use it and what happens when its purpose or source material changes. A committee can settle principles. Named owners turn those principles into working habits.

Photo: Microsoft customer story
“There is no shadow AI.”DR. ADRIAN MAYERS · MICROSOFT CUSTOMER STORY
Give the minutes somewhere to go
A shorter task creates an opportunity. Whether that opportunity improves service depends on what happens next: how work is scheduled, how exceptions are handled and where the recovered attention goes. The stopwatch measures a process. Management must decide what to do with the space it opens.
That is why the modest contract example carries more weight than the spectacular agent count. It connects a recognisable piece of office work with a changed routine and a retained checkpoint. Other teams can examine the shape of that problem without pretending their documents, permissions or review standards are identical.
Premera has offered a concrete place to begin: one repetitive task, a measurable duration and a human decision that remains in the sequence. The ambition can be large. The first question can be small. Which thirty minutes of someone’s day would you most like to give back?
The practical questions
How many AI agents has Premera built?
Premera reports 900 agents built in Microsoft’s April 17, 2026 customer story. The story does not specify how many are active.
What contract task became faster?
A contract exhibit workflow fell from 30–45 minutes to about three. The agent reads incoming documents, presents extracted details for human verification and routes the work to the appropriate queue.
Which Microsoft tools did Premera use?
The case study describes Microsoft 365 Copilot and Copilot Studio, working within Premera’s existing Microsoft environment.
How long did the initial agents take to build?
A pod spanning product, platform, cybersecurity and software development built initial agents in nine weeks.
How did Premera govern its AI effort?
Premera paired training with a Data, AI and Ethics Committee. Its early agents operated within existing Microsoft 365 permissions, and the contract workflow retained human verification.