Mike Watson teaches industrial engineering students at Northwestern through the Client Project Challenge, a project-based course centered on real clients, live data, and operational problems. His work sits at the intersection of higher education, analytics, and applied decision-making, which gives him a close view of how students develop judgment when they are pushed beyond classroom exercises and placed inside real client situations.
The class is designed around the idea that students should work on problems that matter, but not problems so urgent that failure would damage the client. Watson looks for what he calls a client’s fourth-most-important project. It should be important enough that the client shows up every week, shares data, and cares about the result. But it should not be so critical that a 10-week student project carries too much organizational risk.
Students lead the client meetings themselves. Faculty stay in the background. Watson does not sit in on the calls, even though that makes him nervous. The point is to create a setting where students are treated like working professionals, not like students waiting for a professor to step in. They have to ask better questions, manage ambiguity, make assumptions, and carry the work forward.
One of Watson’s lessons is that failure is not always the best teacher for inexperienced teams. Experienced operators can usually diagnose why a project went wrong because they have seen enough successful projects to compare against. Students often do not have that pattern recognition yet. When a project fails, they may learn the wrong lesson. Students need to experience what good progress, client trust, and useful delivery feel like before they can properly interpret failure.
AI has changed what those students can produce. A few years ago, a student without a coding background could not realistically build a working prototype in the time available. Today, LLMs can act as just-in-time tutors, helping students learn inventory models, optimization methods, manufacturing concepts, or software development in the context of a specific client problem. That has made prototypes easier to build and has raised the ambition of what a student team can deliver.
Building something is no longer the main bottleneck. The tougher work is deciding what to build, identifying the real problem behind a messy client request, choosing among possible product directions, and getting people to trust and use the output. As AI lowers the cost of execution, judgment becomes more valuable.
That shift also changes the role of educators and teaching assistants. When students can get technical explanations from an LLM, the human support layer cannot simply answer routine questions. Its value moves toward coaching, critique, framing, and helping students reason through uncertainty. In client projects, the most important questions are rarely generic. They come from incomplete data, shifting priorities, unclear ownership, and stakeholders who may not agree on what the problem actually is.
For Watson, project-based education becomes more important in an AI-enabled world. A case study frames the problem. A client project forces students to discover it. AI does not replace experiential learning. It exposes why experiential learning matters. When students can build faster, they need stronger judgment about what deserves to be built. The scarce skill is no longer only technical execution. It is problem finding, client understanding, operational judgment, and the ability to turn messy reality into useful work.
Market Pulse
Grant Thornton's 2026 AI Impact Survey reports that most organizations are keeping agentic AI away from high-stakes calls. Only 5% allow agents to execute high-stakes decisions without human review, and 60% limit agents to moderate-risk task automation. Organizations with fully integrated AI are nearly four times more likely to report revenue growth than those still piloting, at 58% compared with 15%. Nearly three in four organizations now give agentic AI access to their data and processes, while only 20% have a tested AI incident response plan for when it fails.
Wolters Kluwer advanced its Expert AI strategy with an architecture designed for high-stakes professional decisions in healthcare, tax, accounting, legal, and compliance. The approach combines domain-specific reasoning, proprietary expert content, and validation layers to support professionals who remain accountable for their decisions. Early deployments reduced manual work by 20-30% while helping maintain the quality and consistency of expert advisory services.
Forrester's 2026 Technology & AI Predictions suggest that enterprises are becoming more selective in their AI investments as leadership teams focus on measurable business outcomes. The firm projects that 25% of planned AI spending will shift to 2027, while only 15% of AI decision-makers reported EBITDA improvements in the previous 12 months. Organizations that define success metrics before deployment are likely to be in a stronger position when investment decisions come under closer scrutiny.
Resources and Events
📅 AeraHUB 26 (New York City, NY - October 27-28, 2026)
AeraHUB returns for its annual global summit, bringing together business and technology leaders working on decision intelligence across supply chain, finance, operations, and commercial functions. Speakers include leaders from AstraZeneca, Harvard Business School, and The Hershey Company Details →
📅 The AI Summit (New York City, NY - December 9-10, 2026)
One of North America's largest enterprise AI conferences, returning to New York for its 2026 edition. The program focuses on production AI systems, commercial deployment, and operational outcomes. Sessions span applied machine learning, enterprise infrastructure, AI strategy, and the organizational changes required to scale AI initiatives across large organizations. The summit features 350+ speakers across multiple tracks. Details →
📊 Report Spotlight: The State of Organizations 2026 (McKinsey)
McKinsey's second annual State of Organizations report draws on responses from more than 10,000 senior executives across 15 countries and 16 industries. The research identifies three forces reshaping organizations: advances in technology, changing workforce dynamics, and a growing focus on long-term productivity. A central theme is that leading organizations are redesigning work around what technology makes possible. For analytics and decision intelligence leaders, the report offers useful insight into how organizations are scaling technology adoption while maintaining a focus on measurable business outcomes. Read →
The DecideWise Edge
Benjamin Baer, Founding Member of the DecideWise community, argues that decision velocity, the time between recognizing an opportunity and acting on it, is one of the most overlooked performance metrics in enterprise AI. Organizations routinely measure model accuracy, dashboard usage, and pilot success rates. Far fewer measure how long it takes for an insight to become a recommendation, a recommendation to become a decision, and a decision to become an action. Baer describes decision velocity as the combined effect of four stages: time to insight, time to prediction, time to prescription, and time to execution. Each stage introduces latency, and those delays compound. Read More →
For the Commute
Building a Community for Decision Intelligence (Decision Intelligence Lab)
Benjamin Baer and Linda Crowe, the founders of DecideWise, join Vijay Mehrotra and Michael Watson to explain why the decision intelligence market needs a dedicated community. The conversation traces why the field matured first in financial services, where credit and insurance decisions had to be auditable and defensible, and why it is now spreading into retail, manufacturing, utilities, and government. Baer and Crowe make the case for starting with the business decision the organization needs to make and then working back to the data, and they walk through the governance problems AI introduces, including data leakage, compliance limits under rules such as PCI and HIPAA, and the difficulty of auditing how a neural network reaches an answer.
