Short answer
A decision intelligence platform is a layer on top of your existing systems, not a replacement for them. It connects to applications and data where they live, builds shared context and a knowledge graph, composes forecasting, optimization, rules and LLM reasoning into explicit decision models, executes approved decisions back into systems of record, and governs every decision with policies, audit and monitoring. It runs in your AWS, Azure or Google Cloud account, on-prem, at the edge or air-gapped.
Design principles CTOs ask about
| Principle | What it means in practice |
|---|---|
| Layer, not replacement | No rip-and-replace. Systems of record stay the source of truth; the layer reads, decides and writes back through their APIs. |
| Technology agnostic | Runs on AWS, Azure, Google Cloud, Oracle Cloud, private cloud, on-prem Kubernetes, edge gateways or air-gapped networks. |
| Model agnostic | Any model through a gateway: Anthropic Claude, OpenAI, Google Gemini, Llama, Mistral, or your own — swapped without rewriting decisions. |
| Data stays put | Federated access and change-data-capture instead of bulk copies. Your keys, your network, your residency rules. |
| Decisions as code | Decision models are versioned, testable and reviewable like software — in Python, SQL, DMN and YAML policies. |
| Open standards | REST, GraphQL, OpenTelemetry, OPC UA, MQTT, HL7 FHIR, X12 EDI, OIDC/SAML — no proprietary lock-in. |
The eight layers
ExperienceDecision cards, twins, copilots, embedded widgets
Governance & observabilityAutonomy policies, audit, drift, access
ExecutionWrite-back, workflows, events, physical dispatch
Decision modelsOptions, constraints, objectives, thresholds, KPIs
IntelligenceForecasting, optimization, simulation, rules, LLMs
Context & knowledgeSemantic layer, entity resolution, knowledge graph, vectors
ConnectivityConnectors, CDC, streaming, OT and industry protocols
Your systems & infrastructureERP, MES, EHR, TMS, lakes, any cloud or on-prem
How one decision moves through the stack
01SignalA sensor, system or person flags a change
02ContextThe graph assembles what it means
03ScoreModels evaluate each option
04DecideThresholds and authority applied
05ApproveA human approves or policy executes
06Act & learnWrite-back; outcome measured
↺ Each outcome updates the graph and the models.
Deployment models
| Model | Where it runs | Best for |
|---|---|---|
| Your cloud | Your AWS, Azure or Google Cloud account on EKS, AKS or GKE, inside your VPC with your KMS keys | Most enterprises; fastest path |
| On-prem / private cloud | Your Kubernetes or OpenShift, VMware or bare metal, with models on your GPUs | Data-residency or latency constraints |
| Hybrid + edge | Decision services in cloud, lightweight runners at plants, depots or on robots | Manufacturing, logistics, robotics |
| Air-gapped | Disconnected network, offline model packages, signed updates | Defense and regulated environments |
Security and compliance
Identity
SSO via SAML or OIDC (Okta, Entra ID, Ping), role- and attribute-based access down to the decision.
Data protection
Encryption in transit and at rest with your keys; PII and PHI minimization per decision.
Audit
Every recommendation, approval, override and write-back logged with its evidence and model versions.
Frameworks
Designed to support SOC 2, ISO 27001, HIPAA, GxP and NIST AI RMF controls in your environment.
APIs and integration surface
| Surface | Use |
|---|---|
| Decision API (REST / GraphQL) | Request a decision, fetch evidence, approve or override from any app |
| Events (Kafka, Event Hubs, Pub/Sub, EventBridge) | Stream signals in and decisions out |
| Write-back connectors | Create work orders, holds, reroutes, tasks in systems of record — idempotent and reversible |
| Embeds & SDKs | Decision cards inside SAP Fiori, ServiceNow, Salesforce, Epic or your portal; Python and TypeScript SDKs |
| Telemetry | OpenTelemetry traces and metrics into Datadog, Splunk, Grafana or your SIEM |
Key takeaways
- It is a layer on top of your stack, not a new system of record.
- Cloud-, data- and model-agnostic by design.
- Decisions are versioned, audited and governed like code.
Frequently asked questions
Is the decision intelligence platform cloud agnostic?
Yes. It runs in your AWS, Azure or Google Cloud account, on Oracle Cloud or private cloud, on-prem Kubernetes, at the edge, or fully air-gapped.
Do we need to move our data to a new platform?
No. It connects to data where it lives through connectors, federated queries and change-data-capture, and writes decisions back through system APIs.
Which AI models does it support?
Any model through a model gateway — Anthropic Claude, OpenAI GPT, Google Gemini, Meta Llama, Mistral, cloud-hosted versions on Bedrock, Azure OpenAI or Vertex AI, or your own models.
How does it integrate with SAP, Oracle, Salesforce or Epic?
Through standard APIs (OData, REST, JDBC, HL7 FHIR), event streams and embeddable decision cards that appear inside those applications.