Bob Pierce, PhD is co-founder and Chief Science Officer at DecisionNext. His work has introduced superior mathematical evaluation to completely new markets and industries, enhancing the best way corporations interact in strategic resolution making. Previous to DecisionNext, Bob was Chief Scientist at SignalDemand, the place he guided the science behind its options for producers. Bob has held senior analysis and growth roles at Khimetrics (now SAP) and ConceptLabs, in addition to educational posts with the Nationwide Academy of Sciences, Penn State College, and UC Berkeley. His work spans a variety of industries together with commodities and manufacturing and he’s made contributions to the fields of econometrics, oceanography, arithmetic, and nonlinear dynamics. He holds quite a few patents and is the creator of a number of peer reviewed papers. Bob holds a PhD in theoretical physics from UC Berkeley.
DecisionNext is an information analytics and forecasting firm based in 2015, specializing in AI-driven worth and provide forecasting. The corporate was created to handle the constraints of conventional “black field” forecasting fashions, which frequently lacked transparency and actionable insights. By integrating AI and machine studying, DecisionNext gives companies with better visibility into the components influencing their forecasts, serving to them make knowledgeable selections primarily based on each market and enterprise danger. Their platform is designed to enhance forecasting accuracy throughout the provision chain, enabling clients to maneuver past intuition-based decision-making.
What was the unique concept or inspiration behind founding DecisionNext, and the way did your background in theoretical physics and roles in numerous industries form this imaginative and prescient?
My co-founder Mike Neal and I’ve amassed quite a lot of expertise in our earlier corporations delivering optimization and forecasting options to retailers and commodity processors. Two main learnings from that have had been:
- Customers have to consider that they perceive the place forecasts and options are coming from; and
- Customers have a really arduous time separating what they suppose will occur from the chance that it’ll truly come to cross.
These two ideas have deep origins in human cognition in addition to implications in how you can create software program to resolve issues. It’s well-known {that a} human thoughts isn’t good at calculating chances. As a Physicist, I realized to create conceptual frameworks to have interaction with uncertainty and construct distributed computational platforms to discover it. That is the technical underpinning of our options to assist our clients make higher selections within the face of uncertainty, that means that they can not know the way markets will evolve however nonetheless must resolve what to do now with the intention to maximize income sooner or later.
How has your transition to the function of Chief Science Officer influenced your day-to-day focus and long-term imaginative and prescient for DecisionNext?
The transition to CSO has concerned a refocusing on how the product ought to ship worth to our clients. Within the course of, I’ve launched some each day engineering duties which might be higher dealt with by others. We all the time have an extended record of options and concepts to make the answer higher, and this function offers me extra time to analysis new and revolutionary approaches.
What distinctive challenges do commodities markets current that make them notably suited—or resistant—to the adoption of AI and machine studying options?
Modeling commodity markets presents a captivating mixture of structural and stochastic properties. Combining this with the uncountable variety of ways in which folks write contracts for bodily and paper buying and selling and make the most of supplies in manufacturing ends in an extremely wealthy and complex area. But, the mathematics is significantly much less nicely developed than the arguably easier world of shares. AI and machine studying assist us work by means of this complexity by discovering extra environment friendly methods to mannequin in addition to serving to our customers navigate advanced selections.
How does DecisionNext steadiness using machine studying fashions with the human experience essential to commodities decision-making?
Machine studying as a area is continually enhancing, but it surely nonetheless struggles with context and causality. Our expertise is that there are facets of modeling the place human experience and supervision are nonetheless essential to generate strong, parsimonious fashions. Our clients typically take a look at markets by means of the lens of provide and demand fundamentals. If the fashions don’t replicate that perception (and unsupervised fashions usually don’t), then our clients will typically not develop belief. Crucially, customers is not going to combine untrusted fashions into their each day resolution processes. So even a demonstrably correct machine studying mannequin that defies instinct will turn into shelfware extra possible than not.
Human experience from the client can be essential as a result of it’s a truism that noticed information isn’t full, so fashions characterize a information and shouldn’t be mistaken for actuality. Customers immersed in markets have necessary data and perception that isn’t out there as enter to the fashions. DecisionNext AI permits the consumer to enhance mannequin inputs and create market situations. This builds flexibility into forecasts and resolution suggestions and enhances consumer confidence and interplay with the system.
Are there particular breakthroughs in AI or information science that you just consider will revolutionize commodity forecasting within the coming years, and the way is DecisionNext positioning itself for these adjustments?
The arrival of useful LLMs is a breakthrough that can take a very long time to totally percolate into the material of enterprise selections. The tempo of enhancements within the fashions themselves remains to be breathtaking and tough to maintain up with. Nonetheless, I feel we’re solely in the beginning of the highway to understanding one of the best methods to combine AI into enterprise processes. A lot of the issues we encounter will be framed as optimization issues with sophisticated constraints. The constraints inside enterprise processes are sometimes undocumented and contextually fairly than rigorously enforced. I feel this space is a big untapped alternative for AI to each uncover implicit constraints in historic information, in addition to construct and resolve the suitable contextual optimization issues.
DecisionNext is a trusted platform to resolve these issues and supply easy accessibility to essential data and forecasts. DecisionNext is creating LLM primarily based brokers to make the system simpler to make use of and carry out sophisticated duties inside the system on the consumer’s course. This can permit us to scale and add worth in additional enterprise processes and industries.
Your work spans fields as numerous as oceanography, econometrics, and nonlinear dynamics. How do these interdisciplinary insights contribute to fixing issues in commodities forecasting?
My numerous background informs my work in 3 ways. First, the breadth of my work has prohibited me from going too deep into one particular space of Math. Relatively I’ve been uncovered to many alternative disciplines and might draw on all of them. Second, excessive efficiency distributed computing has been a by means of line in all of the work I’ve completed. Most of the methods I used to cobble collectively advert hoc compute clusters as a grad scholar in Physics are utilized in mainstream frameworks now, so all of it feels acquainted to me even when the tempo of innovation is speedy. Final, engaged on all these completely different issues conjures up a philosophical curiosity. As a grad scholar, I by no means contemplated working in Economics however right here I’m. I don’t know what I’ll be engaged on in 5 years, however I do know I’ll discover it intriguing.
DecisionNext emphasizes breaking out of the ‘black field’ mannequin of forecasting. Why is that this transparency so essential, and the way do you suppose it impacts consumer belief and adoption?
A prototypical commodities dealer (on or off an alternate) is somebody who realized the fundamentals of their trade in manufacturing however has a ability for betting in a unstable market. In the event that they don’t have actual world expertise within the provide facet of the enterprise, they don’t earn the belief of executives and don’t get promoted as a dealer. In the event that they don’t have some affinity for playing, they stress out an excessive amount of in executing trades. In contrast to Wall Avenue quants, commodity merchants usually don’t have a proper background in chance and statistics. In an effort to achieve belief, we’ve to current a system that’s intuitive, quick, and touches their cognitive bias that offer and demand are the first drivers of enormous market actions. So, we take a “white field” strategy the place every little thing is clear. Often there’s a “relationship” part the place they appear deep underneath the hood and we information them by means of the reasoning of the system. As soon as belief is established, customers don’t usually spend the time to go deep, however will return periodically to interrogate necessary or shocking forecasts.
How does DecisionNext’s strategy to risk-aware forecasting assist corporations not simply react to market circumstances however proactively form their methods?
Commodities buying and selling isn’t restricted to exchanges. Most corporations solely have restricted entry to futures to hedge their danger. A processor would possibly purchase a listed commodity as a uncooked materials (cattle, maybe), however their output can be a unstable commodity (beef) that usually has little worth correlation with the inputs. Given the structural margin constraint that costly amenities must function close to capability, processors are compelled to have a strategic plan that appears out into the long run. That’s, they can not safely function completely within the spot market, and so they must contract ahead to purchase supplies and promote outputs. DecisionNext permits the processor to forecast the whole ecosystem of provide, demand, and worth variables, after which to simulate how enterprise selections are affected by the total vary of market outcomes. Paper buying and selling could also be a element of the technique, however most necessary is to know materials and gross sales commitments and processing selections to make sure capability utilization. DecisionNext is tailor made for this.
As somebody with a deep scientific background, what excites you most concerning the intersection of science and AI in reworking conventional industries like commodities?
Behavioral economics has remodeled our understanding of how cognition impacts enterprise selections. AI is reworking how we will use software program instruments to help human cognition and make higher selections. The effectivity positive factors that will likely be realized by AI enabled automation have been a lot mentioned and will likely be economically necessary. Commodity corporations function with razor skinny margins and excessive labor prices, so that they presumably will profit vastly from automation. Past that, I consider there’s a hidden inefficiency in the best way that almost all enterprise selections are made by instinct and guidelines of thumb. Choices are sometimes primarily based on restricted and opaque data and easy spreadsheet instruments. To me, essentially the most thrilling end result is for platforms like DecisionNext to assist remodel the enterprise course of utilizing AI and simulation to normalize context and danger conscious selections primarily based on clear information and open reasoning.
Thanks for the nice interview, readers who want to be taught extra ought to go to DecisionNext.