From Concept to Production-Ready, Agentic AI POC
ABOUT THE COMPANY
Williams handles approximately one-third of the United States’ daily natural gas supply– used to heat our homes, cook our food, and generate our electricity. Williams works closely with customers to provide infrastructure that safely delivers natural gas products and reliably supports the clean energy economy.
OPPORTUNITY: AI-Driven Innovation for Improved Field Safety & Maintenance
Driven by a growing backlog of maintenance tickets, dependency on a small set of SMEs, and increased demand for continuous improvement, Williams wanted to design an AI-driven application that would allow their contractor workforce to submit information about field-based safety or operational concerns in a fast and easy way. This information needed to be high quality so that site managers wouldn’t need to contact the person who submitted the concern for additional details. Williams sought 27Global’s guidance on how AI could support:
- Scaled access to knowledge
- Issue diagnosis & troubleshooting
- Faster repair cycle times
- Improved documentation
- App-embedded smart assistance
SOLUTION: Guided AI Innovation; “Foundations” Workshop ⟶ Agentic AI POC Build
27Global convened a three-day AI workshop with approximately 25 of Williams product and IT team members.-
Day 1: AI Foundations
Before embedding AI broadly, Williams set out to establish a unified foundation of AI knowledge to ensure their organization would be aligned on terminology, architecture patterns, and governance guardrails. Having this baseline understanding would reduce risk, accelerate decision-making, and ensure that the appropriate tools and models were chosen for specific use cases—particularly important in a safety-critical, regulated environment where data privacy, model provenance, and deployment standards matter.
Day 2: Proof of Concept – Ideation & Selection
In preparation for the workshop, 27Global prepared a list of potential AI use cases for Williams to consider. To identify which use case would provide the quickest win as a POC (Proof of Concept) the team:
- Categorized each use case by value area
- Estimated value vs. level of effort
- Clarified scope and requirements so the group was aligned on the intended functionality
This approach not only identified “quick win” options suitable for sub-week POC builds, but also revealed larger ideas that could be split into smaller incremental future projects.
After reviewing a myriad of possible use cases the team chose to prototype an agentic chatbot for contractors to submit concerns for review by a site manager. The chatbot’s core job entailed:
- Parsing user input
- Mapping input to predefined fields that would provide context for the large language model (LLM)
- Prompting the user for missing details
- Preparing a submission the user can send
Before beginning development, the group co-created a concise, collaborative workflow that would allow for cross-team task delegation.
Day 3: Proof of Concept – Build
Once the Williams teams were aligned, 27Global moved directly into developing the production-ready POC on Microsoft Azure, which was guided by several key directives:
- LIGHTWEIGHT APPLICATION A simple chat interface backed by a workflow engine orchestrating agents and integrations.
- HOSTING & AI SERVICES Deployed in a controlled Microsoft Azure cloud environment; No sensitive data processed. AZURE AI SERVICES: Azure AI Foundry, Azure Copilot Studio
- WORKFLOW VALIDATION Validate app & business process alignment
- DATA MODEL Application anchored to existing master data (concerns types, activities, assets, locations) with periodic updates sufficient for pilot
- SAFETY & GOVERNANCE Maintain narrow POC scope; Contractors access via company-specific links; Mandatory human-in-the-loop for safety-critical decisions (full auth) & data transformations
- IMPROVEMENTS & ITERATIONS Improve concerns prioritization; Basic image assistance; Lightweight workflow reporting
RESULTS: Successful POC Launch + Foundation Laid for Future Innovation
Williams partnered with 27Global to establish a shared AI foundation and rapidly build a safety-aware POC in less than a week. The workshop aligned ~25 team members on AI vocabulary, architecture, and governance, then guided the selection and development of an application with agentic chatbot for contractor concern submissions.
Following the workshop, Williams engaged 27Global to harden the POC into an enterprise-ready application and to use the pattern as a template for future AI initiatives across the organization.
