AI is moving from experimentation to everyday business operations. Companies are using it to reduce routine work, make internal knowledge easier to access, accelerate decision-making and give employees more time for higher-value tasks.
Microsoft’s research on real-world workplaces, based on more than a dozen studies, shows that generative AI is already improving productivity — although the impact varies significantly by role, organization and level of adoption. In one of the largest real-world experiments, researchers studied more than 6,000 employees across over 60 organizations using Microsoft Copilot.
But the most interesting question for managers is not whether AI works.
It is where it creates the most value — and how to turn individual productivity gains into business results.
Here are five examples.
1. Morgan Stanley: turning internal knowledge into an AI assistant
Financial advisors work with enormous amounts of information: market research, investment reports, internal documents and client data. Morgan Stanley introduced an AI assistant to help advisors search this knowledge base, summarize information and prepare more relevant insights for clients.
More than 98% of advisor teams adopted the assistant, while the share of documents accessed by advisors increased from around 20% to 80%. The important lesson is not simply that AI makes search faster. It changes the economics of organizational knowledge. Information that previously existed somewhere inside a company’s systems becomes much easier for employees to find and use.
What managers can take from it
Look for processes where employees repeatedly ask:
- “Where can I find this information?”
- “Has anyone dealt with this before?”
- “Which document contains the answer?”
- “Can you summarize this for me?”
These are often strong candidates for an internal AI assistant.
The goal is not to create another chatbot. It is to make the company’s existing knowledge usable at the moment employees need it.
2. KOHLER: making AI adoption part of the management system
KOHLER took a different approach. Instead of simply giving employees access to Microsoft 365 Copilot, the company built an AI change-management and training program around it. The result was striking: 98% adoption among 5,000 employees within six weeks.
Employees reported saving 2–5 hours per week. KOHLER also reported a 25% improvement in code releases and a 75% reduction in error defects with GitHub Copilot. The company has since created around 3,000 AI agents. The key lesson is about adoption. Giving employees an AI license does not automatically create value. Companies need training, internal champions, leadership support and practical examples of how AI can be used in specific roles.
What managers can take from it
Instead of saying:
“Everyone now has access to AI.”
Give teams a more specific challenge:
“Find three recurring tasks in your role that AI could make faster or easier.”
Then share successful use cases across the organization.
This turns AI adoption from an IT project into a management practice.
3. Novo Nordisk: letting employees build solutions themselves
Novo Nordisk demonstrates another model of AI adoption: putting the technology in employees’ hands and allowing them to solve their own problems. In one of the company’s AI initiatives, around 25,000 employees created chatbots for more than 2,500 use cases.
The result was not one universal AI solution. It was thousands of small solutions addressing very different workflows. This matters because managers often know exactly where their teams lose time — but central IT teams cannot automate every small inefficiency. AI changes that equation. An employee can identify a repetitive process, build a simple AI workflow around it and test whether it creates value.
What managers can take from it
Don’t ask only:
“What enterprise-wide AI product should we deploy?”
Also ask:
“What problems are employees trying to solve themselves?”
The best AI use cases may come from the frontline rather than the technology department.
This approach also requires guardrails: employees need clear rules around data, security, accuracy and which decisions can or cannot be delegated to AI.
4. Mitsui: attacking the hidden cost of knowledge work
At Mitsui, one AI use case focused on a very specific bottleneck: reviewing bidding documents. The process could take 30–40 hours. Generative AI reduced review time by around 40%, and in some cases by as much as 80%. This is exactly the type of workflow where AI can create immediate value: large volumes of documents, repetitive comparison, information extraction and summarization.
What managers can take from it
Don’t start by asking:
“How can we use AI in procurement?”
Break the workflow down instead:
Receive documents → read → compare → extract information → identify risks → prepare recommendation.
Then identify which steps actually require human judgment. AI may be able to handle the first four stages while the manager remains responsible for the final assessment. This is often a better approach than trying to automate an entire business process at once.
5. Orion Health: from searching 500,000 records to getting an answer
Orion Health faced a familiar problem: critical knowledge was distributed across a large amount of internal documentation and support history. The company built an internal generative AI chatbot that can search more than 500,000 records in under a minute.
The system is expected to reclaim around 50 staff hours every day by reducing manual search effort. But again, the value is not simply “50 hours saved.” The bigger benefit is that employees can use that time for resolving customer issues, analyzing problems and making decisions rather than searching through documentation.
What managers can take from it
When evaluating an AI project, don’t measure only:
How much time did we save?
Also ask:
What did people do with that time?
If employees simply fill the saved hours with more administrative work, the business impact may be limited.
If they spend that time with customers, on strategy, product development or solving complex problems, AI creates much more value.
The common pattern
These companies operate in very different industries, but their AI use cases have something in common.
They don’t start with:
“Where can we put AI?”
They start with a business problem:
Too much information → AI helps find and summarize it.
Too much repetitive work → AI automates parts of the process.
Too much time spent reviewing documents → AI handles the first analysis.
Too many fragmented sources of knowledge → AI connects them.
Too much administrative work → AI gives employees time back.
Microsoft’s research confirms that AI productivity gains are real, but uneven. The impact depends on the role, workflow and how actively people use the technology. Simply providing access to an AI tool is not enough.
From individual productivity to organizational impact
This is where many AI initiatives face their biggest challenge. An employee can save an hour a day with AI. But that does not automatically mean the company has become more productive. McKinsey’s 2026 research makes a similar point: AI does not create enterprise value simply because more people use it. Companies need to redesign workflows and operating models around the technology.
For managers, this means moving through three stages:
1. Automate
Find repetitive tasks AI can handle.
2. Redesign
Once those tasks become faster, rethink the entire workflow.
3. Reallocate
Use the time and capacity created by AI for work that creates more value. This third step is often overlooked. If AI saves a manager two hours a week, the goal shouldn’t be to schedule two more meetings. It could mean two more hours for strategy, employee development, customer conversations or solving complex problems.
What AI still cannot do for managers
There is another important lesson behind these cases. AI can analyze information, prepare a presentation, summarize a meeting or recommend an option. But it cannot take responsibility for the organization’s direction.
McKinsey argues that as AI becomes more capable, distinctly human leadership becomes more important: setting ambitious goals, making difficult decisions, building trust and recognizing opportunities that go beyond existing patterns.
Where should a company start?
The best starting point is usually not a large AI transformation program. Take one team and map its weekly workload. Find the five activities that consume the most time. Then ask:
- Can AI automate part of this task?
- Can it make the task significantly faster?
- Can it improve quality or accuracy?
- What human judgment still needs to remain?
- What will employees do with the time we free up?
That last question is crucial. The real value of AI is not measured only by how much work it removes. It is measured by what better work becomes possible because of it. And that is where the role of the manager begins to change: from coordinating more tasks to creating the conditions for people and AI to produce better outcomes together.
