Dr. Rich Staats spent 41 years in the U.S. Army, retiring as a Major General after leading the Army Reserve’s 75th Innovation Command. A West Point graduate with a PhD in operations research from MIT, he also spent two decades at MITRE. Across those roles, he approached problems through the same questions: What are we trying to accomplish? How will we measure success? What choices do we have, and what limits do time, people, and money impose? He found that answering those questions often revealed a way forward.
That approach gradually changed how his staff brought him problems. Instead of presenting an issue and waiting for instructions, they arrived with an objective, measures of success, and constraints. Staats recalls that meetings that initially lasted two hours had fallen to 15 minutes after a year in command. People understood how he would assess a proposal, which also gave them a basis for acting without waiting for him to make every decision.
For that independence to succeed, people needed both clear expectations and permission to take reasonable risks. Staats published three principles in every command: become competent in your responsibilities, use your strengths to address needs beyond your assigned tasks, and create an environment where others want to cooperate with you. At MITRE, he reinforced the last two through a monthly Red Pen Award for the biggest mistake. The protection applied to honest attempts to accomplish something useful, within legal and ethical boundaries. Staff displayed the pens proudly, and he saw greater initiative and creativity.
The same willingness to act mattered when information was incomplete. Staats recalls General Rodriguez describing how, early in an Iraq deployment, he waited until he was roughly 80% confident in the data and analysis but accomplished his mission only 20% of the time. Later, he acted with about 20% of the information he previously wanted and reported succeeding 80% of the time. Staats uses the anecdote to explain that gathering more data may reduce uncertainty while increasing the risk of acting too late.
At MITRE, Staats taught new analysts to understand the underlying problem and the choices the executive actually had authority to make. Even a technically correct analysis has little value if its recipient cannot act on it. He encouraged analysts to present ranges and options so leaders could judge which fit their circumstances, rather than expecting one calculated answer to settle the decision.
His concern about generative AI follows from that emphasis on judgment. Staats sees its usefulness, but worries that young analysts who skip the fundamentals will struggle to detect errors or develop their own approaches. If they have never learned how to conduct the analysis, they may also lack the knowledge to check an AI-generated result. He wants analysts to keep proposing and examining their own solutions, especially in military settings where the next problem may involve circumstances nobody has encountered before.
The DecideWise Insider
Tim Varelmann, Founding Member of the DecideWise community, argues that discomfort with algorithmic decisions reflects the role emotion plays in human judgment and commitment. Optimization can compare scenarios, apply constraints consistently, and show the consequences of different choices, but people still need to decide which trade-offs are acceptable. A model might show how accepting 5% less profit improves resilience or sustainability. Whether that cost is justified depends on the organization’s priorities and circumstances. Varelmann proposes using models to clarify the options and their consequences while leaving judgments about values and strategy with people. That gives decision-makers a stronger basis for choosing an action and explaining why it fits their goals.
Market Pulse
Oracle introduced Fusion Claw, adding 25 applications that can handle tasks such as accounting reconciliation, staffing plans, and shipping consolidation. The system combines AI reasoning with deterministic computation and, Oracle says, enforces company policies, permissions, approval requirements, and escalation limits during execution. Each completed run produces an audit record of the evidence, decisions, authority, and actions taken.
Autonomize made Context AI generally available, a shared context layer that lets healthcare organizations define clinical terminology, operational rules, and institutional expertise once and reuse them across agents and workflows. Every agent-supported decision, from prior authorization routing and payment integrity review to coverage determinations, can be traced to the concepts, policies, evidence, and sources behind it. The layer runs on top of existing data platforms, and customers keep ownership of their knowledge and decision logic while every agent shares a governed understanding of it.
OpenAI’s Decisions API is in public beta and returns classifications, probability estimates, and rubric-based scores from text and images. OpenAI says it runs about 10 times faster than the Responses API. Its supported model, GPT-6 Luna, costs $0.10 per million input tokens, with no output-token charges. Applications can use the results to route requests, prioritize issues, or flag cases for review. The guide recommends setting thresholds using labeled examples and the costs of false positives and false negatives.
Liquid AI introduced d1 with text and image support. It can sort support tickets, identify defective parts, and choose an agent’s next action from a set of options. Liquid AI reports text responses in 200-300 milliseconds, at $0.04 per million input tokens. In its tests, d1 matched or beat GPT-6.1 Sol on four of six applications and cost 19-200 times less than the models compared.
Resources and Events
📅 INFORMS Annual Meeting 2026 (San Francisco, CA - November 1-4, 2026)
The European edition of the TechEx series returns to RAI Amsterdam as one of eight co-located events, alongside shows on physical AI, cybersecurity and cloud, IoT, edge computing, intelligent automation, digital transformation, and data centers. Organizers expect more than 8,000 attendees and over 250 speakers across the events. The program covers generative and agentic AI, machine learning, AI governance, data and analytics, and enterprise infrastructure. A free pass and a paid Gold Pass are available, and senior leaders can request a free Enterprise Leadership Pass.
📅 AI & Big Data Expo Europe 2026 (Amsterdam, Netherlands - October 19-20, 2026)
Celonis' annual conference returns to Messe München, drawing AI and process leaders. The conference will focus on using process intelligence and AI to improve business operations, with sessions covering IT, supply chains, finance, and procurement. Speakers include leaders from Hitachi Energy, Mondelēz International, AkzoNobel, and Avis Budget Group. The program also includes Celonis platform updates, while the pre-conference AI Lab gives participants a day to build functional prototypes using the Celonis Context Model.
📅 Catalyst New York (New York, NY - November 18, 2026)
Pigment’s Catalyst New York brings finance and planning leaders together at ARTECHOUSE to explore AI in business planning. The program includes a keynote, a live demonstration of AI agents building models and analyzing data, and a panel on preparing teams and systems for these tools. Speakers include Datadog’s AJ Ljubich and Snowflake's Prince Ayinde. Attendees can see Pigment’s tools in action, hear how peers are approaching adoption, and connect with other practitioners.
📊 Report Spotlight: Structured Elicitation in AI-Assisted Enterprise Decisions (arXiv)
This paper tests whether asking users to clarify the problem, choose a strategy, and explain their reasoning before AI drafts a plan improves the result. Across 400 sessions involving a fictional business crisis, average quality scores rose 32%, from 15.5 to 20.5 out of 40, while off-topic plans fell from 34% to 5%. The extra steps added about 2.4 minutes and improved how clearly plans explained their strategy and trade-offs, but did not improve technical detail.
Join the conversation: Is OpenAI's new Decisions API a decision tool or a constraint-based query?
Join the DecideWise discussion on whether the new API is closer to a constraint-based query, how it compares with TypeSafe AI’s Jev, and where it could be useful in decision-making.
For the Commute
Solving the Meta Problem (Decision Intelligence Lab)
Zohar Strinka joins the Decision Intelligence Lab podcast to discuss how analytics teams choose which problem to solve. She explains the difference between selecting a worthwhile problem and defining it clearly, and how client expectations can push consultants toward solving a problem that misses the intended outcome. The conversation covers setting scope, budgeting for repeated attempts, and revising the approach as new information emerges.
