Curated context graphs
Curated Context Graphs
Keep the judgement your business creates every day. A curated context graph keeps the case, the decision and the reason together, so the next person and the next AI system can learn from it.
The valuable part is often the exception
Imagine an invoice charge that is valid under the contract. A manager waives it because the customer relationship matters more. The system records the waiver, but usually not the judgement behind it. Later, that exception may become a new policy. This is how organisations learn, and it is what a static SOP tends to miss.
The rule
What should normally happen according to the contract, policy or operating procedure.
The exception
What was different about this case and why the usual rule was not enough.
The judgement
Who made the call, what evidence mattered and whether the decision should shape future work.
What a curated context graph does
It gives your organisation a memory of how decisions are made. “Curated” matters because people review what the system learns. Not every workaround should become a rule, and not every rule should last forever.
Keeps decisions with their reasons
A future user can see not only what happened, but why it was the right choice at that moment.
Turns corrections into learning
When a person corrects an AI answer, the useful lesson is retained instead of disappearing into a chat history.
Shows where rules are changing
Repeated exceptions become visible, giving leaders evidence that a policy or process needs to evolve.
Stays under your control
The organisational memory belongs to you, so changing an AI model or vendor does not mean losing what your team learned.
From following rules to writing better ones
AI is getting very good at repeatable work such as extraction, matching and triage. The harder and more valuable work is deciding what should happen when reality does not follow the playbook.
First, automate what is known
Let AI handle repeatable decisions where the rule and the expected outcome are already clear.
Then, retain what is learned
When the AI reaches a new case, route it to the right person and keep the answer where the organisation can reuse it.
Finally, improve the rule
Review patterns across exceptions and decide which lessons should become part of the normal way of working.
How lloops builds it with you
Start with one important decision
Choose a workflow where exceptions cause delay, rework, lost revenue or unnecessary risk.
Listen to the people doing the work
Map the written rules, the unwritten judgement, the evidence people trust and the cases that do not fit.
Connect decisions to their context
Use your existing systems and documents to keep each decision with the facts and reasons that shaped it.
Create a reviewed learning loop
Make it simple for experts to correct the system, approve new guidance and see how outcomes improve.
Where this becomes useful
Supply chain exceptions
Keep how teams respond to delays, missing documents, changed routes and customer commitments.
Commercial exceptions
Connect a commercial decision with the contract, transaction facts, customer relationship and final outcome.
Operational knowledge
Preserve the judgement of experienced people so it survives role changes, reorganisation and growth.
Customer decisions
Give teams a consistent view of previous promises, exceptions and reasons before they take the next action.
Three useful questions for your AI roadmap
Who does the AI ask?
When it meets a case it has never seen, who can make the judgement and where will that answer be kept?
What is actually learning?
Is your organisation's judgement growing, or is the useful context staying inside a model or vendor product?
What do you keep?
If you change models or vendors tomorrow, do you keep the decisions, reasons and learning your team created?
Context graph FAQ
What is a context graph?
A context graph is a living record of how your organisation makes decisions. It keeps a case with the rule, exception, evidence, people involved and final judgement so future teams and AI systems can learn from it.
Why do you call it curated?
People remain responsible for deciding which corrections and exceptions are useful. Human review prevents a one-off workaround or a poor decision from quietly becoming the new rule.
How is it different from a knowledge graph?
A knowledge graph usually describes what things are and how they relate. A context graph also keeps what happened, why a decision was made and how an exception changed the rule.
Do we need new software to build one?
Not necessarily. We start with the systems, documents and review processes you already use. The design follows the business problem rather than forcing a particular database or AI vendor.
Can it use our existing systems and documents?
Yes. It connects the useful context from existing systems, documents and approved expert input without replacing your systems of record.
Who owns it?
Your organisation does. The aim is to keep your operational memory in infrastructure you control, with models that can be changed as your needs evolve.
Start with the decisions your SOP cannot explain
Bring one workflow where experienced people regularly make exceptions. We will map what they know, what the current systems forget and whether a curated context graph is the right next step.