Articles

Netflix's engineering team recently described a home-screen slowdown that took about four hours to diagnose. "Very little of it went into computing anything difficult," they wrote. "Most of it went into reconstructing relationships that our observability systems already knew but we had never written down anywhere a human or machine could simply read and discover."
That sentence is worth reading twice if you run anything that matters. The same gap shows up in enterprise decision intelligence.
The facts, the rules, and the approvals already exist. They are just not written down in a form a person or a system can simply read.
What is an enterprise knowledge graph?
An enterprise knowledge graph is a shared, machine-readable map of entities in a business and how they relate. Netflix built one for operations: an Observability Ontology on open W3C standards, so engineers can read those links instead of rebuilding them by hand. The same idea applies to the evidence, rules, and approvals behind a decision.
Prasanna Vijayanathan and Renzo Sanchez-Silva describe the usual fix in "Ontology-Driven Observability: Building an End-to-End Knowledge Graph at Netflix Scale." They write: "The instinct here could have been, around 2023, to reach for more data or better dashboards, and while both help at the margins, neither addresses what makes these investigations slow."
Their diagnosis is blunt: "Each of our systems is internally coherent and externally isolated." The same thing gets called an application, a service, or a cluster depending on which tool you open. An engineer has to carry the mapping in their head and becomes the bottleneck.
"What we lacked," they write, "was an explicit, machine-readable agreement about the entities in our operational world."
That agreement is the Observability Ontology. Every application is validated against a catalog of roughly 49,000 entries. The graph also models how engineers reasoned during incidents, alongside the live operational data.
A correlation engine on top proposes likely causes and discards weak chains. Each chain is stored with the graph records that produced it and carries its confidence score.
In their words, "The role of the engine here is to propose with evidence, not to make conclusions." They are frank about the early state of the work: "We are early enough that the ranked hypotheses are sometimes wrong, which is why each one carries the evidence that produced it." Humans stay the approvers.
Over time, they write, "the ontology gradually stops being a schema and starts being a memory."
Why do enterprise decisions have the same gap?
Enterprise decisions have the same gap because the evidence, the governing rule, and the sign-off each live in a different system. Each system makes sense on its own, and none share a readable map of how a decision was made. When someone later asks why, a person has to rebuild the answer by hand.
Netflix's system is about operations, and I don't want to stretch the comparison. But the structural problem it solves is the one we see in enterprise decisions every week at data².
Think about how a material decision gets made. The evidence sits in documents, ERP records, field reports, and email. The rule that governs it sits in a contract, a policy, a permit condition, or a regulation.
Conflicts between sources get resolved in someone's head. The sign-off is an approval in a workflow tool.
Each system is internally coherent and externally isolated. The person who connects them becomes the bottleneck and the single point of failure.
So when an auditor, a regulator, or opposing counsel asks why the decision was made, someone rebuilds the answer by hand. That's the decision-side version of those four hours, and the rebuild can take anywhere from days to weeks, depending on how much data the question needs. It often happens months later, after people have moved on.
Why doesn't adding AI on top of disconnected systems fix this?
Adding AI on top of disconnected systems does not close the gap. A model that summarizes unlinked sources can sound just as sure when the evidence is thin as when it is solid. Generic retrieval finds relevant passages, but it does not know which rule governs or where two records disagree.
Dashboards show the outcome. Scores with no path back give you a number, not a record you can defend.
The same gap is visible in decision intelligence vs business intelligence. Business intelligence reports outcomes, while decision intelligence keeps the logic, evidence, and approvals behind a decision.
How does reView™ apply this idea to decisions?
reView™, the decision intelligence platform from data², applies an ontology and an enterprise knowledge graph to the decision. It connects data where it already lives, models the entities and the rules that govern them, and keeps those relationships explicit so a person can inspect the work. The output is a Decision Record™ that has to stand up outside the building.
The design principles will look familiar to anyone who read the Netflix post: propose with evidence, attach the confidence, and keep a human as the approver. The difference is the object.
Netflix's chains explain an event in operations. Ours support a business decision that has to stand up when someone who was not in the room checks it.
Here is the same problem, side by side:
Netflix operations | Enterprise decision intelligence | |
|---|---|---|
What is broken | Each system is coherent on its own and isolated from the others | Evidence, rules, and sign-off sit in separate tools |
Usual fix that falls short | More data and better dashboards | More AI on top of disconnected sources |
Shared agreement | An Observability Ontology, stored as a knowledge graph | An ontology and an enterprise knowledge graph applied to the decision by reView™ |
What it produces | Evidence-backed chains, each with a confidence score | A Decision Record™ for each material decision |
Who decides | Humans stay the approvers | A person approves, qualifies, or escalates |
What does a Decision Record™ hold?
A Decision Record™ is the lasting output of a material business decision, with its proof attached. It holds the evidence linked back to source, the governing rule, any conflicts between records, a confidence level, and a human sign-off. If an auditor, a board, or a regulator asks why the decision was made, the Decision Record™ is the reply.
For each material decision, the Decision Record™ holds:
The evidence: the specific records and documents behind the conclusion, linked back to source.
The governing rule: the contract term, policy, or regulation that applies.
Conflicts: where sources disagree, surfaced for review rather than silently resolved.
Confidence: how far the evidence supports the conclusion. Where it's too thin, reView™ says so.
Human sign-off: a person approves, qualifies, or escalates. Machines reason, people decide.
Decision intelligence examples across industries
These decision intelligence examples come from the sectors we work in. The subject changes. The requirement does not: the decision has to hold up outside the building.
Energy: a well or capital decision has to hold up for partners and regulators.
Financial services: a payment or risk finding has to survive audit.
Supply chain: a vendor call depends on agreed terms that rarely sit next to the transaction.
Defense and intelligence: a finding has to show its sourcing before anyone acts on it.
Why does a decision need to show its work?
A decision needs to show its work because it has to hold up when someone outside the room checks it. You cannot dashboard your way out of a modeling problem. Done right, the Decision Record™ stops being paperwork and starts being the organization's memory of why it acted.
That's the standard we build to at data². Connect the facts, keep the rules explicit, surface the conflicts, be honest about confidence, and let people decide. Then show the work.
Key Takeaways
Netflix's Observability Ontology writes down operational relationships so people and machines can read them.
Enterprise decision intelligence has the same problem: evidence, rules, and approvals live in systems that are isolated from each other.
Adding AI on top of disconnected sources does not create a record you can defend.
reView™ applies an ontology and a knowledge graph to the decision itself and keeps a human as the approver.
A Decision Record™ holds the evidence, the governing rule, conflicts, confidence, and human sign-off. It becomes the organization's memory of why it acted.
FAQ
What is decision intelligence?
Decision intelligence is the practice of connecting the evidence, the rules, and the approvals behind a business decision so the result can be checked and defended. data² is a decision intelligence company. Its platform, reView™, applies an ontology and a knowledge graph to the decision and enables AI to scale, with a human as the approver.
What is a Decision Record™?
A Decision Record™ is the lasting output of a material business decision, with its proof attached. It holds the evidence linked to source, the governing rule, any conflicts, a confidence level, and a named person's sign-off. If an auditor, a board, or a regulator asks how you know, the Decision Record™ is the reply.
What are examples of decision intelligence?
Examples of decision intelligence include a well or capital decision that has to hold up for partners and regulators. A payment or risk finding has to survive audit.
A vendor call depends on agreed contract terms. A defense or intelligence finding has to show its sourcing before anyone acts on it. In each case, the Decision Record™ holds the evidence, the governing rule, conflicts, confidence, and human sign-off.
How does decision intelligence differ from business intelligence?
Decision intelligence differs from business intelligence in what each one records. Business intelligence reports on what happened and organizes data into dashboards.
It does not show which rule applied or where sources disagree. Decision intelligence connects the evidence, the governing rule, and the approvals, and produces a Decision Record™ a person can check and defend.
Does decision intelligence replace human approval?
No. Decision intelligence puts the evidence, the rule, and the conflicts in front of a person who can approve, qualify, or escalate.
Netflix makes the same point for operations: the engine proposes with evidence and does not make conclusions. At data², the standard is the same. Machines reason, people decide.
Source: Netflix Technology Blog, Prasanna Vijayanathan and Renzo Sanchez-Silva, "Ontology-Driven Observability: Building an End-to-End Knowledge Graph at Netflix Scale," 16 October 2026.
Last Posts

Articles
Decision Intelligence Needs the Same Links as Netflix
Read More

Articles
After Check-the-Box: What Your Examiner Will Ask For Instead
Read More

Articles
What Energy Companies Need to Scale the AI They Already Own
Read More

Articles
data² Launches Asset Performance & Optimization for Energy
Read More
©2026 Data Squared USA Inc. | All rights reserved | US Patent US012339839B2