Bob Pierce, PhD is co-founder and Chief Science Officer at DecisionNext. His work has brought advanced mathematical analysis to entirely new markets and industries, improving the way companies engage in strategic decision making. Prior to DecisionNext, Bob was Chief Scientist at SignalDemand, where he guided the science behind its solutions for manufacturers. Bob has held senior research and development roles at Khimetrics (now SAP) and ConceptLabs, as well as academic posts with the National Academy of Sciences, Penn State University, and UC Berkeley. His work spans a range of industries including commodities and manufacturing and he’s made contributions to the fields of econometrics, oceanography, mathematics, and nonlinear dynamics. He holds numerous patents and is the author of several peer reviewed papers. Bob holds a PhD in theoretical physics from UC Berkeley.
DecisionNext is a data analytics and forecasting company founded in 2015, specializing in AI-driven price and supply forecasting. The company was created to address the limitations of traditional “black box” forecasting models, which often lacked transparency and actionable insights. By integrating AI and machine learning, DecisionNext provides businesses with greater visibility into the factors influencing their forecasts, helping them make informed decisions based on both market and business risk. Their platform is designed to improve forecasting accuracy across the supply chain, enabling customers to move beyond intuition-based decision-making.
What was the original idea or inspiration behind founding DecisionNext, and how did your background in theoretical physics and roles in various industries shape this vision?
My co-founder Mike Neal and I have amassed a lot of experience in our previous companies delivering optimization and forecasting solutions to retailers and commodity processors. Two primary learnings from that experience were:
- Users need to believe that they understand where forecasts and solutions are coming from; and
- Users have a very hard time separating what they think will happen from the likelihood that it will actually come to pass.
These two concepts have deep origins in human cognition as well as implications in how to create software to solve problems. It’s well known that a human mind is not good at calculating probabilities. As a Physicist, I learned to create conceptual frameworks to engage with uncertainty and build distributed computational platforms to explore it. This is the technical underpinning of our solutions to help our customers make better decisions in the face of uncertainty, meaning that they cannot know how markets will evolve but still have to decide what to do now in order to maximize profits in the future.
How has your transition to the role of Chief Science Officer influenced your day-to-day focus and long-term vision for DecisionNext?
The transition to CSO has involved a refocusing on how the product should deliver value to our customers. In the process, I have released some day to day engineering responsibilities that are better handled by others. We always have a long list of features and ideas to make the solution better, and this role gives me more time to research new and innovative approaches.
What unique challenges do commodities markets present that make them particularly suited—or resistant—to the adoption of AI and machine learning solutions?
Modeling commodity markets presents a fascinating mix of structural and stochastic properties. Combining this with the uncountable number of ways that people write contracts for physical and paper trading and utilize materials in production results in an incredibly rich and complicated field. Yet, the math is considerably less well developed than the arguably simpler world of stocks. AI and machine learning help us work through this complexity by finding more efficient ways to model as well as helping our users navigate complex decisions.
How does DecisionNext balance the use of machine learning models with the human expertise critical to commodities decision-making?
Machine learning as a field is constantly improving, but it still struggles with context and causality. Our experience is that there are aspects of modeling where human expertise and supervision are still critical to generate robust, parsimonious models. Our customers generally look at markets through the lens of supply and demand fundamentals. If the models do not reflect that belief (and unsupervised models often do not), then our customers will generally not develop trust. Crucially, users will not integrate untrusted models into their day to day decision processes. So even a demonstrably accurate machine learning model that defies intuition will become shelfware more likely than not.
Human expertise from the customer is also critical because it is a truism that observed data is never complete, so models represent a guide and should not be mistaken for reality. Users immersed in markets have important knowledge and insight that is not available as input to the models. DecisionNext AI allows the user to augment model inputs and create market scenarios. This builds flexibility into forecasts and decision recommendations and enhances user confidence and interaction with the system.
Are there specific breakthroughs in AI or data science that you believe will revolutionize commodity forecasting in the coming years, and how is DecisionNext positioning itself for those changes?
The advent of functional LLMs is a breakthrough that will take a long time to fully percolate into the fabric of business decisions. The pace of improvements in the models themselves is still breathtaking and difficult to keep up with. However, I think we are only at the beginning of the road to understanding the best ways to integrate AI into business processes. Most of the problems we encounter can be framed as optimization problems with complicated constraints. The constraints within business processes are often undocumented and contextually rather than rigorously enforced. I think this area is a huge untapped opportunity for AI to both discover implicit constraints in historical data, as well as build and solve the appropriate contextual optimization problems.
DecisionNext is a trusted platform to solve these problems and provide easy access to critical information and forecasts. DecisionNext is developing LLM based agents to make the system easier to use and perform complicated tasks within the system at the user’s direction. This will allow us to scale and add value in more business processes and industries.
Your work spans fields as diverse as oceanography, econometrics, and nonlinear dynamics. How do these interdisciplinary insights contribute to solving problems in commodities forecasting?
My diverse background informs my work in three ways. First, the breadth of my work has prohibited me from going too deep into one specific area of Math. Rather I’ve been exposed to many different disciplines and can draw on all of them. Second, high performance distributed computing has been a through line in all the work I’ve done. Many of the techniques I used to cobble together ad hoc compute clusters as a grad student in Physics are used in mainstream frameworks now, so it all feels familiar to me even when the pace of innovation is rapid. Last, working on all these different problems inspires a philosophical curiosity. As a grad student, I never contemplated working in Economics but here I am. I don’t know what I’ll be working on in 5 years, but I know I’ll find it intriguing.
DecisionNext emphasizes breaking out of the ‘black box’ model of forecasting. Why is this transparency so critical, and how do you think it impacts user trust and adoption?
A prototypical commodities trader (on or off an exchange) is someone who learned the basics of their industry in production but has a skill for betting in a volatile market. If they don’t have real world experience in the supply side of the business, they don’t earn the trust of executives and don’t get promoted as a trader. If they don’t have some affinity for gambling, they stress out too much in executing trades. Unlike Wall Street quants, commodity traders often don’t have a formal background in probability and statistics. In order to gain trust, we have to present a system that is intuitive, fast, and touches their cognitive bias that supply and demand are the primary drivers of large market movements. So, we take a “white box” approach where everything is transparent. Usually there’s a “dating” phase where they look deep under the hood and we guide them through the reasoning of the system. Once trust is established, users don’t often spend the time to go deep, but will return periodically to interrogate important or surprising forecasts.
How does DecisionNext’s approach to risk-aware forecasting help companies not just react to market conditions but proactively shape their strategies?
Commodities trading isn’t limited to exchanges. Most companies only have limited access to futures to hedge their risk. A processor might buy a listed commodity as a raw material (cattle, perhaps), but their output is also a volatile commodity (beef) that often has little price correlation with the inputs. Given the structural margin constraint that expensive facilities have to operate near capacity, processors are forced to have a strategic plan that looks out into the future. That is, they cannot safely operate entirely in the spot market, and they have to contract forward to buy materials and sell outputs. DecisionNext allows the processor to forecast the entire ecosystem of supply, demand, and price variables, and then to simulate how business decisions are affected by the full range of market outcomes. Paper trading may be a component of the strategy, but most important is to understand material and sales commitments and processing decisions to ensure capacity utilization. DecisionNext is tailor made for this.
As someone with a deep scientific background, what excites you most about the intersection of science and AI in transforming traditional industries like commodities?
Behavioral economics has transformed our understanding of how cognition affects business decisions. AI is transforming how we can use software tools to support human cognition and make better decisions. The efficiency gains that will be realized by AI enabled automation have been much discussed and will be economically important. Commodity companies operate with razor thin margins and high labor costs, so they presumably will benefit greatly from automation. Beyond that, I believe there is a hidden inefficiency in the way that most business decisions are made by intuition and rules of thumb. Decisions are often based on limited and opaque information and simple spreadsheet tools. To me, the most exciting outcome is for platforms like DecisionNext to help transform the business process using AI and simulation to normalize context and risk aware decisions based on transparent data and open reasoning.
Thank you for the great interview, readers who wish to learn more should visit DecisionNext.
Leave a comment