o9 Solutions has been an innovator in the supply chain planning market. They held their user conference in Chicago last week. I spoke with Chakri Gottemukkala, o9’s CEO. I told him o9 seemed prescient for adopting the knowledge graph so early. Yes, he said, when large language models emerged, other supply chain planning software vendors realized they needed to capture more context to make LLMs effective.

o9 Solutions is a leading provider of supply chain planning solutions. They have grown quickly. This year alone, they added 40 new customers. They generate over $500 million in annual revenue, up from ongoing subscription revenue in the $200 million range in 2023 . They are recognized as both one of the largest supply chain planning vendors and a leader for their comprehensive planning capabilities.

The knowledge graph is one of the three core technologies at the heart of a modern supply chain planning platform.

The supply chain planning market got started when supply planning models were put into in-memory databases in the early 1990s. Without this technology, an advanced plan that could account for complex constraints across a supply chain could not be generated quickly enough to be useful. This is still a core planning technology.

For demand planning, the other core SCP application, the solution has grown from statistical forecasting to forecasting using machine learning, genetic algorithms, and other algorithms and methods.

But in recent years, the key advance has been adding knowledge graphs to SCP platforms. o9 was the first to do this. Other leading SCP vendors now understand the need for a knowledge graph and are working to catch up.

When you think knowledge graph, think “flexibility.” Knowledge graphs support flexibility by using a graph‑structured model that can integrate diverse data sources and be updated without the rigid restructuring required by relational databases that rely on fixed tables and schemas. This technology makes it much easier to update planning models and integrate demand and supply planning.

I asked o9’s CEO: if other major SCP suppliers were working to introduce their own knowledge graph, how would o9 continue to differentiate itself in the market?

“The biggest problem in companies is not just the technology; it’s about change,” Mr. Gottemukkala explained. “And the knowledge graph that we had built earlier was essentially a way to bring all the data and the knowledge of the enterprise and how they make decisions into a single model, so that all the silos get connected. So, we focused on that, and we had really solved the problem of bringing the data and knowledge together. But we hadn’t solved the problem of change. That’s really what we are focused on now.”

The knowledge graph is now being extended, Mr. Gottemukkala explained, not just to capture the business knowledge of an enterprise - how products are made, market intelligence, customer preferences, supplier capabilities, and many other things – but also to include how decisions are made in the organization. “What we call the operating model of the business.” How does the company analyze situations? Who makes decisions? How should decisions be routed? What is driving planning decisions? This includes a huge quantity of tacit knowledge. “The problem is: how do you drive change in an organization? And to capture that, we need to capture a lot more of the context of the enterprise and how they make decisions and what drives decision-making behaviors.”

For example, demand can shift in the market, and a company has multiple ways to respond: they could change pricing, launch a new product, kill some SKUs, and more. All of those choices must be analyzed. “All the context that led to that decision - what was the situation, what analysis was done, what was considered- can all be captured in a decision trace object. So, after six months, if I want to ask the question: Hey, why did you make that decision? What panned out? What did not?” Those questions can be answered.

But knowing who made the decisions, what decisions were made, and the outcomes also adds context. “Today, all that context is lost. We store some final numbers, but we don’t store the entire context related to decisions.” o9 speaks of “Decision Trace Objects”; this is a method to capture the full context of a decision. This would include the choices considered, the analysis performed, and unstructured text/numerical data surrounding decisions. This allows for natural language queries about past decisions.

As the boundaries between planning and execution have broken down, supply chain planning can be quite agile. If something unexpected happens, the planning engine can be run to see how to best take advantage of a new opportunity or mitigate a disruption in the supply chain.

o9 believes SCP solutions also need to be “adaptive.” Adaptive is their term for supporting continuous improvement of the planning process.

In short, “while others are coming and trying to catch up with where we already were,” Chakri explained, o9 is moving toward what they believe will be the next big innovation in the market.

The Enterprise Knowledge Graph

o9 refers to its knowledge graph as the Enterprise Knowledge Graph.

When you begin to model a supply chain, you begin with a business model. These products roll up to this product family; this business unit serves these customers. The business model captures products, markets, customers, and suppliers.

Supply planning, in particular, relies on a decision model. By modeling the constraints (how fast factory machines can be set up, the time it takes a supplier to ship their products to a manufacturer, the amount of inventory a warehouse can hold, and hundreds of other elements) and the policies (the service level target for our best customers is 95%, for other customers it is 90%, and again there can be many, many policies in a supply chain), a company can create an optimized plan.

Several other leading SCP companies are in various stages of trying to put the business model and decision model into their own knowledge graph. But o9 alone is talking about adding an operating model as a key tool for building an adaptive supply chain.

Gartner is a leading industry analyst firm. They influence what solutions companies buy. They are a big proponent of Agentic AI. The result has been a good deal of “agent washing” by supply chain solution providers hoping to curry favor with Gartner and the companies that listen to Gartner.

If by Agentic AI, you mean that the SCP stack is composed of many small, autonomous agents - little pieces of code - that work together to orchestrate what will happen in a supply chain, that will not work. SCP requires really big pieces of code .

But if you think of agents as something that will operate on the edge, which is o9’s view, this really can move supply chain planning forward. So, for example, a planning engine might generate a plan that includes how much of which components to buy from various suppliers. Over time, agents may be trusted to execute these purchase orders autonomously. Thus, executing the plan is one area where agentic AI can help.

But o9 also talks about using agents to survey the external environment and make sure the situation that led to the plan hasn’t been overturned by unexpected events.

o9 uses a distinctive vocabulary. o9 speaks of managing external signals through a “world model” to monitor the entire value chain. The world model monitors external factors (competitors, consumers, economic data), while the enterprise model monitors internal performance. This external monitoring will be conducted by AI agents. AI will monitor diverse feeds like social media, demographic data, or local zip code sales and pricing trends.

The system will use “rough cut” impact analysis to determine if an emerging signal (such as an influencer’s post - which could lead to soaring demand for a particular SKU) warrants a full-fledged scenario analysis or should be discarded to avoid overwhelming human and compute resources. This intelligence layer will perform initial assessments of whether a situation is important, which then determines if it enters an active queue for further analysis. The impact analysis helps to separate signal from noise.

In practice, if an external signal leads to a new forecast with much higher sales for a particular item, the supply plan is run to see whether and how that new demand can be met. If macroeconomic or industry signals indicate lower demand, the knowledge graph is used to see what inbound purchases can be delayed or canceled.

This World Model is not a completely new idea. Vendors in the market provide real-time supply chain risk solutions, including risks that may occur multiple tiers up the supply chain. Other vendors use transportation network solutions to determine if inbound or outbound goods will arrive on time. Networked procurement solutions improve the intelligence received from suppliers. But o9 is taking this concept and extending it to create signals for monitoring the entire value chain.