A working definition
Decision intelligence treats every recurring decision, a price change, a supplier switch, a stock reorder, a payment approval, as an object that can be designed, instrumented, and improved. It layers analytics, forecasting, optimization, and machine learning underneath a business workflow so the recommendation lands with the person (or system) who actually decides.
How it differs from business intelligence
Traditional BI answers "what happened?" through dashboards and reports. Decision intelligence answers "what should we do, and did it work?" It closes the loop:
- BI reports metrics. DI recommends actions and tracks their outcome.
- BI is consumed on-screen. DI is embedded in the workflow (pricing tool, ERP, procurement system).
- BI is governed at the report level. DI is governed at the decision level, inputs, model versions, approvals, and audit trails.
- BI success is dashboard adoption. DI success is measured in margin, cost, cycle time, or risk.
The three layers of a DI system
1. Data foundation
Integrated internal data, reliable external signal (competitor prices, tariffs, marketplaces), and documented KPIs everyone trusts.
2. Decision models
Forecasts, optimizations, scenario simulations, and business rules that turn signal into a recommendation, with an evaluation set, not a vibe.
3. Execution & feedback
The recommendation reaches the right owner in the right tool, and the outcome flows back so the model gets sharper over time.
Where decision intelligence pays off first
- Pricing and margin intelligence, daily competitive-aware price recommendations.
- Procurement and landed-cost decisions, supplier, route, and duty-optimized sourcing.
- Margin-leakage recovery across contracts, rebates, and billing.
- Document-heavy workflows (invoices, POs, claims) with bounded AI review.
- Marketplace and compliance monitoring against tariffs, HS codes, and MAP policies.
What to build before you buy a "DI platform"
Most decision-intelligence platforms fail because the underlying data foundation isn't ready. Before evaluating tools, make sure you have (1) integrated operational data with owned definitions, (2) reliable external signal with SLAs, and (3) one named decision, one baseline metric, and one target metric per use case.
That's what our Data Readiness Score measures in 10 questions.