A Bad Process With AI Is Still a Bad Process. Microsoft’s Own Playbook Says Fix the Work First

For two years the standard enterprise AI strategy has been a licensing strategy: buy seats, push adoption, count active users, and wait for productivity to show up. Microsoft, which sells more of those seats than almost anyone, has now published the results of running that experiment on itself, and its conclusion is blunt. “A bad process with AI is still a bad process,” Kathleen Hogan, Microsoft’s Executive Vice President and Chief Strategy and Transformation Officer, said of the findings, as reported by Channel Insider. The results that moved business numbers came from teams that tore a workflow down to its steps before any agent touched it.

In a September 17 post on the Official Microsoft Blog, Hogan summarized the lesson this way: “Access to AI will not be the differentiator. The advantage will come from an organization’s ability to empower and engage employees, redesign work, govern AI responsibly and operationalize what works.” The accompanying playbook draws on more than 100 internal transformation case studies across corporate functions, sales, and engineering, and one line in it should be pinned above every operations leader’s desk: “A tool licensed and rolled out to 100,000 employees does not change how the work gets done.”

The Supply Chain Did Not Get Faster by Adding Agents

The most instructive case is Microsoft’s cloud supply chain, because it is the one most operations teams will recognize. According to the Microsoft post, the organization built more than 100 purpose-built agents across planning, sourcing, fulfillment, and logistics (Konsulteer’s summary of the playbook puts the count at 111). The agents evaluate demand shifts, model capacity, and compare transportation options across air, land, and sea on cost, timing, and carbon impact.

The order of operations is the point. Channel Insider reports that the team mapped and stripped down its procedures first, then introduced the agents. The result was an average planning cycle that fell from roughly 10 business days to under 2.5 across five monthly planning cycles, a reduction of up to 75% in select workflows. Demand plan investigations that used to take five to seven days now take a few hours, and some finish in under 20 minutes.

Just as telling is how the human role changed. Planners began by using agents to answer questions. They now update or cancel purchase orders directly, within defined permissions and approval thresholds. That is not a faster version of the old job. It is a different job, with the control points designed in rather than bolted on afterward.

Usage Was Not the Same as Impact

The sales numbers carry a similar warning. CFO Dive reports that account managers who used Microsoft 365 Copilot at least half the time between January and June 2024 generated 9.4% more revenue per manager and closed deals at a 20% higher rate than low-usage colleagues. But Microsoft did not get those gains by handing out a chat window. Its sales organization deployed specialized agents for pipeline analysis, deal packages, and customer research, then ran weekly peer-led huddles to spread what worked. Adoption of priority use cases tripled.

Konsulteer’s reading of the playbook makes the same point in operating terms: simply deploying copilots or agents across existing business processes is unlikely to deliver meaningful transformation on its own. The sellers who gained were working inside a redesigned motion, with purpose-built agents and a peer cadence for spreading what worked. Counting active users would have captured none of that.

That gap shows up in finance data too. A Gartner study of CFOs, covered by CFO Dive on September 24, found that 55% reported positive overall returns from their 2025 AI initiatives, yet 57% said returns were unclear when they looked at individual use cases. Gartner puts the typical time to value for straightforward finance use cases at nine to ten months, and Marco Steecker, senior director analyst in Gartner’s finance practice, warned that if a project has not delivered by that point, “you’re probably just putting money into something that isn’t going to be delivering value.”

Managers, Not Models, Set the Ceiling

The playbook’s most underrated finding is about people. Microsoft reports that when managers actively model AI use, reported value from agentic AI rises 17 points and trust rises 30 points. Employees on teams where managers create psychological safety are 1.4 times more likely to be high-frequency users of agentic AI. And 58% of AI users say it helps them do work they could not do before, rising to 80% among advanced users.

Channel Insider highlights a related strategic claim: Microsoft recommends treating foundation models as interchangeable and prioritizing proprietary evaluations and institutional knowledge over vendor lock-in. For operations leaders, that is encouraging. The asset that matters most is one they already own, which is a detailed understanding of how their processes actually run.

The Redesign-First Playbook

• Start with one outcome, not a tool: Microsoft’s first lesson is to begin with the business result. Pick a metric such as planning cycle time or close rate and design backward from it.
• Map and strip the process before automating it: The supply chain gains followed a procedure teardown. Remove redundant handoffs and approvals first, or agents will simply run a bad process faster.
• Design the control points in advance: Planners can cancel purchase orders because permissions and approval thresholds were defined up front. Decide where human judgment is mandatory before granting agents authority.
• Measure outcomes, not seats: Report adoption and business results separately, and judge each redesigned workflow on the metric it was meant to move so early usage numbers are not mistaken for returns.
• Make managers go first: A 17-point lift in reported value when managers model AI use is cheaper than any license. Build manager-led demonstrations and peer huddles into the rollout.
• Set a kill date: Use Gartner’s nine-to-ten-month benchmark as a review point. Projects that have not moved an outcome metric by then should be redesigned or stopped.

Microsoft’s experience is not proof that every company can cut planning cycles by three quarters. It is proof that the company with the easiest possible access to AI still had to do the unglamorous work of redesigning processes before the numbers moved. Most organizations have not started that work. That is the real gap, and no license closes it.