The Four Layers of the o9 Enterprise Knowledge Graph

Most enterprises can tell you what happened, but far fewer can explain why it happened, which decision shaped the outcome, or whether the same choice should be made again.

That gap becomes especially visible when conditions change. A supplier misses a commitment. Demand shifts unexpectedly. A promotion underperforms. Inventory builds in the wrong place. Teams respond by pulling data from multiple systems, revisiting old presentations, and relying on the people who remember what happened last time.

The information exists, but the context around it is fragmented.

Plans are separated from the assumptions behind them. Decisions are buried in meetings or spreadsheets. Outcomes are reviewed after the fact, but rarely connected back to the exact plan version, policy, or trade-off that produced them.

The o9 Enterprise Knowledge Graph, or EKG, is designed to connect these missing pieces. Let’s jump right in.

Layer 1: The Value-Chain Graph

The first layer, the Value-Chain Graph, represents how value moves through the enterprise.

Most businesses are modeled across disconnected systems. Product data may sit in one place, customer orders in another, capacity information elsewhere, and financial assumptions in spreadsheets. The Value-Chain Graph connects these elements through a common model.

It includes the relatively stable structures that define the business: products, bills of materials, suppliers, plants, distribution centers, customers, contracts, promotions, cost structures, and financial guardrails.

It also represents the facts, assumptions, and plans that change over time.

Facts may include inventory, orders, available capacity, lead times, realized costs, and margins. Assumptions may include expected demand, promotional lift, supplier availability, or future costs. Plans represent the enterprise’s time-phased intent across demand, supply, inventory, commercial, and financial planning.

These plans are not stored as isolated snapshots. They are versioned and connected to the assumptions used to create them. The graph can distinguish between a draft plan, an approved baseline, and a segment that has entered a frozen execution window.

It also captures variances and situations. A supplier delay, for example, can be linked to the products, customers, locations, plan versions, service risks, and financial impacts it affects.

Layer 1 therefore answers: What does the value chain look like, and what is its current and projected state?

Its value is not simply visibility because it gives commercial, operational, procurement, and finance teams a shared representation of the business. It also provides the context AI agents and algorithms need to reason across functions rather than within separate planning silos.

Layer 2: The Decision-Context Graph

The vast majority of companies have systems of record for transactions, but few have a system of record for decisions.

A planning system may show that the forecast changed, inventory was moved, or production was reallocated. It may not show who approved the change, which alternatives were considered, which assumptions were in force, or what outcome was expected.

The second layer, the Decision-Context Graph, addresses this gap.

It records decisions as structured objects. Each material decision can be linked to an accountable owner, a timestamp, a scope, the plan version it affected, the facts and assumptions used, and the measures that will later determine whether it worked.

These decisions may be strategic, such as opening a new distribution center; tactical, such as approving a seasonal promotion; or operational, such as reallocating stock during a shortage. They may be made by executives, planners, optimization engines, or AI agents.

The graph also separates three ideas that are often blurred together: what the business has committed to do, what it expects will happen, and what it wants to achieve.

That distinction matters. When commitments, forecasts, and targets are treated as interchangeable, accountability weakens and planning becomes vulnerable to bias. Teams may undercommit to protect performance metrics or rely on optimistic assumptions without making the risk visible.

By linking outcomes back to the exact decisions, assumptions, and plan versions that preceded them, Layer 2 creates a much stronger basis for learning.

Instead of simply asking why the business missed the plan, teams can investigate whether the assumption was wrong, the response was ineffective, another decision created a dependency, or execution diverged from the approved intent.

Layer 2 answers: What did we decide, why did we decide it, and what happened as a result?

Layer 3: The Learned-Rules Graph

Enterprises contain a great deal of experience, but much of it remains tacit.

What that means is, for example, an experienced planner may know that a supplier often misses commitments during seasonal peaks. A commercial leader may recognize that certain promotions lift revenue but dilute margin. A supply chain executive may know that inventory pre-builds only work under specific demand, capacity, and shelf-life conditions.

The third layer, the Learned-Rules Graph, turns that experience into institutional knowledge.

It uses the evidence captured across the first two layers to create summaries, insights, policies, and actionable rules.

A summary might show that inventory plan-to-actual gaps increased over the past quarter. An insight might identify that the gaps were associated with volatile assumptions or late demand changes. An actionable rule could then require additional scenario analysis when similar conditions arise.

This conditional element is important because generic lessons such as “improve forecasting” or “collaborate earlier” are difficult to apply. Rules need to specify the conditions under which a certain action is more or less likely to succeed.

For example, the graph might learn that inventory pre-build decisions frequently fail when promotional assumptions exceed historical elasticity ranges. The enterprise can then require explicit elasticity analysis before similar plans are approved.

Layer 3 helps move the organization away from relying on individual memory and toward embedding proven decision logic into future planning.

It answers: What have we learned, and how should that change the way we plan and decide next time?

Layer 4: The Connected-Compute Graph

The first three layers represent the enterprise, its decisions, and what it has learned. The fourth layer puts that knowledge to work.

The Connected-Compute Graph is the executable engine behind the o9 EKG. It connects data pipelines, forecasting models, simulations, optimization algorithms, analytics, and AI agents into a synchronized compute fabric.

Rather than operating as a collection of isolated models or scripts, it understands dependencies. It knows what needs to be calculated, which inputs and assumptions are required, the order in which calculations should run, and which downstream plans may be affected.

Compute can be triggered by a regular planning cycle, an event such as a demand spike or policy breach, or a question from an executive, planner, or AI agent.

Once triggered, the engine follows a repeatable decision-intelligence pathway built around four questions:

What happened, and why?
The system identifies variances and analyzes root causes against the plans and assumptions in force.

What is the current state?
It updates the operational and financial picture at the appropriate level of detail.

What is likely to happen next?
Forecasting and simulation models project possible outcomes under different assumptions and constraints.

What are the best actions to take?
Optimization and recommendation engines generate decision-ready responses, such as reallocating inventory, expediting supply, substituting materials, or revising a plan.

These calculations can take place in a governed planning environment or in a separate what-if context where teams compare alternatives without immediately changing the baseline. Once a response is approved, it can be promoted into the operational plan.

Because the Connected-Compute Graph understands dependencies, it can also recalculate only what has been affected by a change rather than rerunning the entire planning model.

Layer 4 answers: How do we use the enterprise’s data, context, and learned knowledge to take the next best action?

Four layers, one learning loop

As you can see, each layer solves a distinct problem.

The Value-Chain Graph represents the business and its plans. The Decision-Context Graph records decisions and accountability. The Learned-Rules Graph captures what the enterprise has learned. The Connected-Compute Graph uses that knowledge to analyze, predict, prescribe, and act.

Combined, they create a continuous learning loop.

Strategy and financial guardrails shape the available plans. Compute generates alternatives. Decision-makers approve plans and make commitments. Execution produces outcomes. Those outcomes are connected back to the decisions, assumptions, and plan versions that shaped them. The resulting lessons are then converted into policies and rules that improve the next cycle.

This is the shift from an open-loop enterprise, where plans are created and reviewed retrospectively, to a closed-loop operating model that continuously senses, decides, learns, and adapts. The ambition is to create an enterprise system capable of remembering why decisions were made, understanding whether they worked, and applying those lessons at scale.

That is what turns enterprise knowledge from a passive record into an active operating advantage.

A Guide to the o9 Enterprise Knowledge Graph

The o9 Enterprise Knowledge Graph (EKG) is a four-layer, closed-loop system designed to transform how enterprises plan, decide, and execute.

Download now