Short answer
Decision intelligence deploys in weeks instead of quarters because most of the platform is pre-built: industry use-case templates, connectors and APIs, a KPI and threshold library, decision logic, forecasting and optimization algorithms, AI guardrails, a decision API with write-back, a frontend kit with decision cards and digital twins, and infrastructure-as-code for any cloud. A typical first decision connects data in week one, runs in shadow mode by around week three, executes approved actions by around week six, and is in production by around day 90.
What’s already built
| Accelerator | What’s included | Built with |
|---|---|---|
| Business use cases | Templates for maintenance timing, quality holds, scheduling, prior-auth triage, claim routing, exception rerouting, fleet task allocation, production forecasting | Decision model templates |
| Connectors & APIs | SAP, Oracle, Dynamics, Maximo, historians, MES, Epic, Facets, TMS/WMS, Salesforce, ServiceNow, files and email | OData, REST, JDBC, OPC UA, MQTT, FHIR, EDI, CDC |
| Logic | Options, constraints, objectives, escalation and authority rules | Python, DMN, YAML policies |
| Math & algorithms | Forecasting, anomaly detection, survival curves, optimization, Monte Carlo, SPC | Python, scikit-learn, PyTorch, OR-Tools |
| KPIs & thresholds | OEE, MTBF, MTTR, OTIF, first-pass yield, denial rate, turnaround time, cost of delay — with default alarm and confidence thresholds | SQL, dbt metrics |
| AI logic | RAG over SOPs and records, tool-using agents, abstain rules, evaluation sets, prompt and model versioning | Model gateway, vector store |
| Backend | Decision API, event bus, write-back with idempotency and rollback, audit log | FastAPI, Kafka, PostgreSQL |
| Frontend | Decision cards, approvals, scenario views, 3D digital twins with heatmaps, embeds for Teams and Slack | React, three.js, web components |
| Cloud properties | Networking, identity, secrets, autoscaling and monitoring for AWS, Azure, GCP or on-prem | Terraform, Helm, Kubernetes, OpenTelemetry |
A typical first 90 days
- Week 1 — ConnectRead-only connectors to 3–8 systems; semantic layer mapped; entities resolved.
- Week 2 — Model the decisionApply the use-case template; set KPIs, thresholds, authority and escalation with the decision owner.
- Week 3 — Shadow modeRecommendations run alongside today’s process; accuracy and value measured without risk.
- Week 6 — Approved actionsOwners approve recommendations in their tools; write-back creates work orders, holds or reroutes.
- Day 90 — ProductionBounded, reversible decisions automate inside policy; the next decision reuses the same layer.
Why it’s faster
No rip-and-replace
It sits on top of your systems, so there is no migration project on the critical path.
Templates, not blank pages
Each use case starts with a working decision model, KPIs and thresholds you tune.
Reuse compounds
The second decision reuses connectors, the graph and the frontend — it goes faster than the first.
Your cloud, day one
Infrastructure-as-code deploys into your AWS, Azure, GCP or on-prem environment.
Timelines are illustrativeActual timelines depend on data access, security review and the decision chosen. The console above is an illustrative example, not a guarantee.
What a decision looks like as code
# decision: maintenance timing (illustrative)
decision: maintain_or_run
owner: reliability_engineer
signals: [vibration_mm_s, bearing_temp_c, rul_days]
options: [run_to_turnaround, fix_in_planned_slowdown, controlled_shutdown]
objective: minimize(expected_cost_of_downtime + repair_cost)
thresholds:
vibration_alarm: 7.1 # mm/s
min_confidence: 0.80
approval_over_usd: 100000 # escalate to engineer
autonomy: augmented # human approves
writeback: cmms.work_order(draft=true)
kpis: [unplanned_downtime_h, mtbf_days, decision_latency_h]Key takeaways
- Use cases, APIs, logic, math, KPIs, AI, backend, frontend and cloud are pre-built.
- The first decision goes live in weeks; the next ones go faster.
- Everything is configurable in Python, SQL and YAML — no black box.
Frequently asked questions
How fast can decision intelligence be deployed?
A first decision typically connects data in the first week, runs in shadow mode by about week three, executes approved actions by about week six, and reaches production around day 90, depending on data access and security review.
What is pre-built in the platform?
Use-case templates, connectors and APIs, decision logic, algorithms, KPI and threshold libraries, AI guardrails, backend services, frontend components including digital twins, and cloud infrastructure-as-code.
What programming languages does it use?
Decision logic and models are written in Python and SQL with YAML policies; the frontend uses React and three.js; infrastructure uses Terraform and Helm.
Do we need a data lake first?
No. The platform connects to data where it lives. A lake or warehouse helps but is not a prerequisite.