The proposed action
A machine-readable action, target, segment, threshold or intervention appropriate to the business decision.
Decision API
Put the decision inside the workflow.
Graphite Note can deliver model outputs as machine-readable decision objects so another application can consume a recommendation, expected impact, confidence and supporting evidence without forcing users into a separate analytics screen.
Machine-readable output
A recommendation your product can consume.
The exact contract depends on the use case and integration. A decision object can carry the fields required by the receiving system, including decision context, recommendation, expected impact, confidence, evidence and governance state.
The example shown here is illustrative, not a fixed public API contract.
Illustrative decision objectJSON
{
"decision": "reduce_discount",
"segment": "high_value_repeat_customers",
"recommended_change": "-5%",
"expected_margin_impact": "+180000 EUR",
"confidence": 0.88,
"evidence": ["causal_effect", "uplift_test"],
"review_required": true
}Integration pattern
Keep your interface. Add a decision layer.
Your application
→Business context + candidate actions
→Graphite Note Decision API
→Recommendation + impact + evidence
The receiving system can be an ERP, CRM, BI product, workflow platform, vertical SaaS product or AI agent.
What can be returned
More than a score.
What is expected to change
An estimated business effect, uncertainty or confidence value when the underlying method supports it.
Why the action was selected
References to model outputs, causal evidence, constraints or other decision context required for audit and review.
Policy-bound execution
A decision object can carry the guardrails with the recommendation.
For agentic or automated workflows, the useful contract is not just an action. It is an action that has been checked against the relevant business policy and can state what the receiving system is allowed to do next.
Eligibility and constraints
Candidate actions can be screened against product, customer, commercial and operational rules before recommendation.
Permissions and approval state
The object can distinguish actions that are safe to execute automatically from those that require review or escalation.
Evidence and outcome record
Keep the recommendation, expected impact, evidence, governance state and eventual outcome together so the decision can be reviewed later.
Outcome feedback loop
The decision should not disappear after execution.
A governed decision system can keep the recommendation and the eventual result connected so teams can review whether the expected effect happened, where performance drifted and which decision rules deserve another look.
Record what actually happened
Keep the action taken, approval state, target and execution context tied to the original decision object.
Compare expected vs actual outcome
Attach the realised business outcome when it becomes available and compare it with the expected effect and uncertainty.
Use outcomes to improve the next decision
Surface systematic gaps, changing conditions or weak decision rules for model, policy or workflow review rather than assuming every recommendation remains valid forever.
Direct answers
How the API fits.
Does the Decision API replace our application UI?
No. It is designed to let an existing product keep its own interface and workflow while consuming the decision layer programmatically.
Can the output require human review?
Yes. A workflow can keep approval with the responsible user and use the API to provide the recommendation, expected impact and evidence needed for that review.
Can Graphite Note work behind an AI agent?
Yes, where the integration is appropriate. The agent can use a governed decision object as a structured tool output instead of inventing an action from free-form text alone.
Can the API enforce business rules and escalation?
The integration can include policy, eligibility and operating constraints before a recommendation is returned, plus a governance state that tells the receiving workflow whether the action may proceed, needs approval or should be escalated.
Can we evaluate a decision after it was executed?
Yes. The workflow can attach the action taken and the eventual business outcome to the original decision object so expected-versus-actual performance can be reviewed later. That supports audit, monitoring and improvement without treating every recommendation as permanently correct.
Decision IntelligenceFrom signal and prediction to a governed business action.AutoMLAutomate repetitive model-building work without losing business context.Causal Machine LearningSeparate drivers from correlations and estimate intervention effects.Predictive AnalyticsForecast demand, score customers and anticipate business outcomes.Decision OptimisationCompare feasible actions and rank them by expected business impact.Forecasting vs OptimisationSee where prediction ends and the business decision begins.Decision APIDeliver recommendations, expected impact and evidence into workflows.Embedded Decision IntelligenceAdd a decision layer to ERP, CRM, BI, workflow and vertical software.Build vs BuyA practical framework for deciding whether to build, buy or embed.
Your product already has the workflow. Add the decision.
We can scope the decision object and integration around one high-value use case.