Your AI pilots are stalling because agents live outside the workflows that run your business. BAW Agent Factory puts agentic AI inside IBM Business Automation Workflow. Agents call your real processes, service flows, and human tasks as tools. They handle sync and async execution natively. No new platform, no retraining your team, governance from day one.
Enterprise BAW Projects
IBM Gold Partner
Days cycle time (claims)
IBM AI for Business Award
The Problem
Standalone agent platforms sit outside your real workflows. That creates a specific set of problems that don't go away with better prompting.
Every new tool creates adoption burden and IT lift your team didn't budget for: a separate governance model, a separate deployment pipeline, a separate skill set.
Glue code connecting agents to your BAW processes breaks. It's expensive to maintain, hard to version, and impossible to audit cleanly.
Compliance teams can't approve what they can't trace. Audit trails that live outside BAW don't meet regulated-environment standards.
When agents can't access real process data in real time, they stay in sandbox environments indefinitely. The demo looks great. Production never arrives.
The most common cause isn't a shortage of good ideas. Agents live outside the workflows that run the business.
BCG, 2024
What BAW Agent Factory Does
BAW Agent Factory is a BAW-native agent layer. You model, run, and govern AI agents directly inside IBM Business Automation Workflow, using BPMN patterns your team already knows. Agents can call BAW processes, service flows, human tasks, and MCP-connected tools, giving them access to the full toolkit your environment already has.
Agents participate in long-running BAW processes, wait for human decisions, and resume automatically. No polling loops, no external orchestration.
Agents call your existing BAW processes, service flows, and human services directly as tools. The infrastructure you've already built becomes the agent's toolkit.
BAW's workflow logic controls when agents run, what they can access, and what happens at low confidence. Adaptive AI outputs inside the compliant process shell your organization already depends on.
SLA controls, audit trails, and workflow policy configurations, all within the framework your compliance team already approves.
Governance boundary: inside BAW throughout
Capabilities
Five capabilities that let BAW shops deploy governed AI agents without standing up a new platform.
Design agents as BPMN subprocesses using patterns your team already knows. Agents map requests to tools, execute steps in parallel where useful, evaluate outputs, and complete or escalate, all in the same environment your BAW developers work in every day.
Run Claude, GPT, Granite, Llama, or any model your organization approves, without rewriting your integration layer. MCP (Model Context Protocol) support lets you consume tools from other systems or publish BAW services as tools for external agents. No custom wrappers.
Every prompt, input, and output is logged. BAW workflow configurations control PII handling, model selection, and escalation rules. Confidence and grounding checks trigger automatic human-in-the-loop paths. The full audit trail lives inside your BAW governance boundary, not a separate system.
Agents don't just fire and return. They participate in long-running BAW processes, wait for approvals or human decisions mid-flight, and resume automatically. BAW's deterministic workflow layer controls exactly when agents run and what triggers escalation. Adaptive AI inside a compliant shell.
Agents connect to your existing BAW processes, service flows, and enterprise systems: ERP, CRM, content stores, APIs. Tool adapters for LLM, RAG, vector, OCR, translation, and search are included. No new integration platform to stand up, license, or govern.
Architecture
Six steps from agent design to production, all inside BAW.
Design the agent subprocess using standard BPMN patterns. Define what tools the agent can call, what parameters they accept, and what confidence thresholds trigger escalation.
The agent subprocess calls your existing BAW processes, service flows, and human services directly, using the infrastructure you already have.
Policy decisions live in BAW workflow configurations: model selection, PII handling, escalation thresholds, confidence gates. These sit alongside your existing process controls. Business owners can update thresholds without touching the agent subprocess.
Prompts, inputs, outputs, confidence scores, latency, and cost telemetry are captured in the observability store. Nothing lives outside your BAW governance boundary.
Below-threshold cases route to a task UI automatically. The reviewer gets full context, makes a decision, and the process continues. Designed in from the start, not bolted on later.
Agents, prompts, tools, and policies move through dev, test, and production with standard BAW lifecycle management. No separate deployment pipeline.
Use Cases
BAW Agent Factory is designed for regulated, high-volume environments where governance and auditability aren't optional.
Agents classify inbound claims, extract key fields, enrich from systems of record, and provide a decision recommendation with documented rationale. Low-confidence claims route to a human reviewer with full context.
Outcomes: Fewer manual touches. Faster cycle time. Documented decisions that survive audit.
Agents validate documents, run KYC enrichment checks, build a risk profile, and gate approval with configurable confidence thresholds. Human review steps are built into the workflow for edge cases.
Outcomes: Shorter onboarding loops. Fewer exceptions reaching the back office. A documented decision trail for every case.
Agents evaluate clinical documentation against coverage criteria, flag missing information, and route incomplete cases back to submitters automatically. Reviewers receive a pre-built summary with relevant policy citations.
Outcomes: Manual review time drops. Denial rationale is traceable. Staff handle more cases with the same headcount.
Agents classify purchase requests, validate against policy, check vendor risk scores, and route to the right approver with a pre-built brief. Routine approvals clear without human intervention.
Outcomes: Throughput increases. Rework drops. Every approval decision is documented.
Proof
We've been building on IBM BAW in regulated environments since 2011. BAW Agent Factory comes out of that work.
days claims cycle time
2025 IBM AI for Business Award, North America
Salient won IBM's top AI for Business award in North America for a claims process built on IBM BAW, IBM RPA, and watsonx.ai. The only North America winner. That architecture is the foundation BAW Agent Factory extends.
enterprise projects delivered
Salient has delivered BAW implementations across financial services, insurance, healthcare, and manufacturing. We know where governance gaps appear, where agents add the most value, and where deterministic control needs to stay deterministic.
IBM BAW UI toolkit
Salient created SPARK, the UI toolkit that powers IBM Business Automation Workflow interfaces across enterprise deployments. When we say BAW Agent Factory is built native, that depth of platform knowledge is what's behind it.
including Eli Lilly
BAW Agent Factory is running in production today across regulated enterprise environments, including active deployments in life sciences and financial services. This is production-grade software, not a prototype.
Accelerator Benefits
The outcomes that matter to the teams that own BAW environments.
If you already have BAW processes built, you can expose them as agent tools and have something running today. Reusable BPMN patterns accelerate everything else from there.
Every existing process, service flow, and integration becomes a tool the agent can call. If you're already running BAW, you're closer to production agents than you think.
Audit trails, workflow policy controls, and human-in-the-loop paths are part of the agent pattern. Compliance teams can review the design, not just the outputs.
Agents run inside your BAW governance boundary. No new middleware layer, no new vendor, no new security review to reach production.
Cost and usage telemetry are built into the observability layer. You know what every agent run costs before it becomes a budget problem.
How to Get Started
Whether you're renewing CP4BA or want guided hands-on support, there's a direct path to your first production agent. Included with CP4BA Renewal
Included with CP4BA Renewal
Renew your CP4BA license through Salient and receive BAW Agent Factory at no additional cost.
Managed Service
20 hours per month with Salient process engineers and BAW specialists.
You can use any model: Claude, GPT, Granite, Llama, and others. MCP support means you can swap providers without rewriting wrappers. Model selection logic lives in workflow configurations alongside your other governance controls.
If you renew your CP4BA license through Salient Process, BAW Agent Factory is included at no additional cost, with updates and support for as long as the renewal continues. Available in the US, Canada, UK, and Colombia.
Standalone platforms require you to build governance on top: audit logs, approval workflows, compliance controls. BAW Agent Factory runs inside the governance structure you already have. Your existing BAW controls and audit infrastructure apply to every agent run.
Your BAW developers can start immediately. Agent subprocesses use standard BPMN patterns. There's no new language to learn. The Agentify Managed Service option provides Salient support for use-case selection, prototype development, and process assessment if you want it.
Confidence thresholds configured in your workflow catch low-quality outputs before they reach downstream decisions. Below-threshold cases route to human reviewers automatically. Every run is logged with inputs, outputs, and confidence scores so you can audit what happened and tune thresholds over time.
Yes. That's the core design principle. BAW's deterministic controls handle routing, SLAs, and compliance enforcement. AI runs at the specific steps where it adds value: classification, extraction, summarization, decision support. You define the boundary.
Start with a 60-minute use-case workshop. We'll map one workflow, identify the right AI insertion points, and sketch the agent design, using your real process.
Already on CP4BA? Talk to us about your renewal and get BAW Agent Factory at no additional cost.
