Irv Lustig: Identifying and Mitigating AI Project Risk

Irv Lustig is an optimization scientist at Princeton Consultants who has worked across algorithmic research, product management, and technical consulting. He spends most of his time on large optimization and AI deployments. His main contribution to decision teams is the Princeton 20, a framework Princeton uses to surface project risk before work begins.

The Princeton 20 splits into two sets of ten factors. The environmental factors describe the business conditions around a project, such as who owns the decision, what value the solution should produce, and whether the organization is ready to change how it works. The technical factors describe the methodology, tooling, data, and delivery. Princeton rates each factor as a risk, a neutral point, or a clear strength, records the scores in a spreadsheet, and builds a mitigation plan for every factor that is rated as a risk. The scoring happens during the sales conversation, so the client sees the risks named up front.

Decision scope is one of the more useful factors. Lustig describes it as understanding how long a decision lasts and how far it extends into the rest of the operation. One clinic asked Princeton to rebuild its weekly patient schedule from a clean sheet. As the team dug in, they found that a new schedule would force dozens of patients who had kept the same appointment time for years to new times. The problem was adjusting the existing schedule to use resources more effectively while minimizing rescheduling. Framing the scope correctly changed what the system was built to do.

Value proposition is a second factor worth adopting. Lustig measures value through money saved or earned and through faster decision-making. When the value is hard to quantify, Princeton treats the cost-benefit analysis as its own deliverable. In one engagement, the analysis showed the client could capture gains from process changes alone before any optimization was built. Lustig also tracks a risk that teams often ignore: competing initiatives within the client that can pause or redirect a project regardless of its merit.

The approach other leaders can borrow most directly is how Princeton runs delivery. Rather than building a full model over several months and returning with a finished answer, the team works in three-week sprints, shows the user interface and the decisions early, and keeps a human reviewing the output. That review builds the trust needed to move toward more automated decisions later, and it surfaces requirements clients did not know they had until they saw the system make a decision they disagreed with.

Gartner's flagship gathering for CIOs and senior IT leaders returns to the Walt Disney World Swan & Dolphin Resort, drawing more than 7,000 technology executives for four days of analyst-led research. The 2026 agenda covers agentic AI, AI infrastructure, cybersecurity and resilience, emerging technologies, and IT operating models, with sessions on multiagent systems, AI-native software development, postquantum security, technology trends for 2027, and the organizational changes required to deploy AI at scale. The program includes presentations, workshops, roundtables, private meetings with more than 140 Gartner analysts, peer networking, and an exhibition featuring over 180 technology providers. Details →

Big Data LDN is a two-day conference and exhibition for the data, analytics, and AI community, returning to Olympia London for its twelfth year. The program runs more than 400 seminars spanning data architecture, machine learning, predictive analytics, and governance, with case studies on forecasting, anomaly detection, and decision intelligence across industries. Speakers include senior leaders from Liverpool Football Club, ServiceNow and AstraZeneca. Details →

Artificial intelligence is being deployed as a collaborator in enterprise decision-making, yet most organizations still evaluate it primarily through model accuracy. This paper argues that accuracy alone says little about whether humans and AI are actually prepared to make effective decisions together. Instead, it proposes a framework for measuring decision readiness across four dimensions: decision outcomes, reliance behavior, safety signals, and learning over time. It emphasizes calibration, error recovery, and governance throughout the human-AI collaboration lifecycle. Read →

Benjamin Baer, Founding Member of the DecideWise community, argues that the most interesting question about Decision Intelligence is not how large the market will become, but how to define the market in the first place. Rather than treating DI as a standalone technology category, he describes it as the convergence of data management, analytics, optimization, simulation, AI, business rules, and application development. Companies need to identify the sub-markets that matter most to their operations, select the right platform strategy, and build differentiated solutions for clearly defined users. Market size may attract investors, but sustained growth comes from execution, differentiation, and solving real problems. Read More →

Ganesh Ramakrishna, founder and CEO of Lyric and previously co-founder of Opex Analytics, joins Vijay Mehrotra and Michael Watson to trace his path from building custom AI solutions for enterprise clients to founding a venture-backed platform for supply chain decisioning. The conversation covers what changes when a services-led analytics business tries to become a product company, and why product-market fit in decision intelligence looks different from product-market fit in most enterprise software. Ramakrishna's central argument, that supply chain decisions need an AI-first architecture built for real-time volatility, is a useful frame for any analytics leader evaluating build-versus-buy on decision infrastructure.