Building a Production Order Orchestration Agent Across Enterprise Distribution Networks

Motion is one of the largest industrial parts distributors in the United States, supporting a massive network of customers across manufacturing, logistics, and industrial operations.

At that scale, even routine operational workflows become deeply complex systems problems.

One of Motion’s largest operational burdens was customer order intake. Purchase orders arrived through email in inconsistent formats and required employees to manually interpret requests, identify customer accounts, correlate products against internal catalogs, validate information across multiple systems, and create downstream orders. What appeared to be a straightforward intake workflow was actually a coordination problem spread across fragmented enterprise infrastructure.

Focused partnered with Motion to design and build a production-grade agent system using LangGraph that could reason through unstructured customer requests, orchestrate actions across existing enterprise systems, and support employees throughout the order intake process without requiring Motion to replace the infrastructure already powering the business.

The result was a production multi-agent orchestration system integrated directly into Motion’s operational workflows across its national branch network.

The Challenge

The complexity of the problem extended far beyond parsing incoming emails.

Orders arrived in inconsistent formats with varying levels of detail, ambiguous product references, incomplete customer information, and branch-specific operational nuances. Successfully processing an order often required employees to coordinate information across multiple internal systems before downstream execution could occur.

Motion’s infrastructure reflected the realities of a large enterprise environment. Customer data lived in legacy APIs that were never designed for modern agent workflows. Product correlation depended on Elastic Search alongside internal systems with varying data quality and retrieval patterns. The variability of customer communication meant the system needed to reason through ambiguity while maintaining reliability and traceability throughout the workflow.

Focused also identified a critical operational requirement early in discovery: Motion needed strict ownership and control over evaluation, tracing, and runtime telemetry data. Any production AI system would need to satisfy enterprise governance requirements while operating securely inside Motion’s existing infrastructure.

Those requirements directly influenced how the platform was architected from the beginning.

The Solution

Focused designed the platform as a coordinated LangGraph workflow composed of specialized reasoning and execution paths.

The system independently handled customer identification, order interpretation, product correlation, ambiguity detection, and downstream execution across Motion’s operational systems. This architecture allowed the workflow to maintain state across complex processes while keeping employees actively involved in validation and decision-making throughout the order lifecycle.

The LangGraph architecture operated as a reasoning and orchestration layer across the systems Motion already relied on every day. The platform integrated directly into Motion’s internal APIs, Elastic Search infrastructure, and operational tooling while maintaining visibility and traceability throughout the workflow lifecycle.

The user experience was also designed around how Motion employees actually work. Rather than presenting a single opaque recommendation, the system surfaced exact matches, close matches, and “z-coded” alternatives distinctly so experienced operators could review decisions quickly and move through workflows efficiently. The objective was to accelerate operational workflows while keeping employees in control of critical decisions.

Focused worked side by side with Motion’s engineering team to build the evaluation and observability infrastructure alongside the agent system itself. Evaluation was treated as a core part of the platform from the beginning, with shared ownership around testing, tracing, runtime visibility, and long-term operational reliability.

Using LangSmith, the teams established structured evaluation pipelines, golden datasets for regression testing, runtime tracing, and experimentation workflows that allowed Motion engineers to better understand system behavior under production conditions and improve the platform safely over time.

Because of Motion’s governance requirements, LangSmith was deployed in a self-hosted Kubernetes environment rather than through a third-party cloud deployment. This allowed Motion to retain ownership over operational traces, evaluation data, and runtime telemetry while still providing the visibility required to operate production agent systems responsibly at enterprise scale.

The rollout began with a controlled deployment across a small subset of branches before expanding incrementally across Motion’s broader distribution network. That phased rollout allowed the teams to validate escalation behavior, operational reliability, evaluation thresholds, and workflow performance under real-world conditions before broader deployment.

Results

The LangGraph-based platform is now operating across Motion’s branch network, supporting customer order workflows that previously required manual coordination across multiple systems and operational teams.

More importantly, the engagement demonstrated what it takes to operationalize agent systems inside large enterprise environments. Success depended on coordinating reasoning across fragmented systems, establishing measurable evaluation practices, introducing visibility into agent behavior, and operating reliably inside infrastructure that existed long before modern AI frameworks arrived.

By introducing a reasoning and orchestration layer on top of Motion’s existing systems, Focused and Motion transformed a heavily manual operational workflow into a production agent platform capable of supporting one of the country’s largest industrial distribution networks.

The engagement reinforced a pattern Focused continues to see across enterprise AI adoption: successful agent systems are deeply integrated operational platforms built around orchestration, evaluation, observability, and the realities of existing enterprise infrastructure.