An LLM agent can read a customer email, route it, and explain itself. It cannot reliably solve a fleet-routing problem, apply 400 underwriting rules in a deterministic order, or own a multi-day claim across three departments. Each of those is a different kind of work — and each has a purpose-built engine in the IBM stack.
Routing, scheduling, allocation, blending — problems with millions of feasible answers and one optimal one. LLMs hand-wave; CPLEX solves.
Underwriting, pricing, eligibility, KYC — hundreds of policy rules that must apply identically every time and survive an audit. ODM stores them as code, not prompts.
A claim that spans intake, adjudication, payout, and post-payment review takes days, not seconds. BAW is the case file the agents and bots write into.
Each product handles a kind of decision agents shouldn't try to make alone. We compose them — agent calls Decision Optimization for the math, ODM for the rules, BAW for the long-running case.
Mathematical Optimization
Industry-standard solver for linear, mixed-integer, and constraint programming. Common use cases:
Problems with hard constraints and a quantifiable objective — the prescriptive answer is provably optimal, not vibes.
Business Rules · Decision Tables
The deterministic decision service every agent and bot calls when the answer must be repeatable, auditable, and editable by the business. Common use cases:
Authored as decision tables, versioned, simulated, and deployed as a service.
Long-Running Cases · Human-in-the-Loop
Case-management spine for work that takes days or weeks (claims, onboarding, loan origination, dispute resolution). Key capabilities:
The thing an LLM context window cannot hold.
Discovery · Conformance · Simulation
Reads event logs from your ERP, ITSM, and core systems to reconstruct how the process actually runs, surfacing:
The evidence layer that says “automate this step, not that one” before any agent or bot is built.
From first event-log scan to a fully composed decision plane your agents and bots call into.
An evidence-based starting point.
Deliverable: process map + automation roadmap + tool recommendation
Deliverable: live decision service + business-authored rules
Deliverable: deployed solver + benchmark vs current method
The case file becomes the system of record.
Deliverable: live workflow + SLA dashboards + integration hooks
The IBM business-automation suite is the prescriptive engine room behind Decision Stack. Agents handle intent and judgment; ODM, CPLEX, and BAW handle the math, the rules, and the long-running case.
Agents reason and route. ODM enforces rules. CPLEX solves optimization. BAW orchestrates multi-day cases. RPA executes the structured steps. Five engines, one prescriptive plane.
Process Mining surfaces the as-is process — feeding the knowledge plane with what really happens, not what the SOP says.
CPLEX consumes Analytics-Stack forecasts as inputs — “given this demand projection, what's the optimal plan?”
CPLEX has been the gold-standard MIP solver for thirty years. ODM evolved from JRules, the rules engine that powered banking and insurance decisioning long before LLMs existed. BAW carries forward decades of BPM patterns. We pick IBM here precisely because these are battle-tested engines with decades of production proof — and because each exposes a clean API the rest of the stack can call.
