## Overview This case study, presented at an AWS conference (likely re:Invent 2024), features a joint presentation between AWS and Jabil, the world's second-largest contract manufacturing company. Jabil operates with approximately $29 billion in revenue, 140,000 employees across 100+ locations in 35 countries, and manages $25 billion in global procurement spend with 38,000 suppliers. The presentation, led by May from Jabil's technology leadership team, details how the company is leveraging Amazon Q Business to transform their manufacturing and supply chain operations through generative AI. The presentation provides valuable insights into how a large manufacturing enterprise approaches LLMOps at scale, particularly focusing on governance, data strategy, and practical deployment considerations rather than just proof-of-concept demonstrations. ## Strategic Foundation and AI Governance Jabil's approach to AI and GenAI deployment is notably structured around five key pillars that underpin their strategy: The first pillar emphasizes **value creation** over experimentation. May explicitly states they are "over the stage of POC" and focuses on building solutions with clear business justification. This represents a mature approach to LLMOps where the emphasis shifts from proving technology works to delivering measurable business outcomes. The second pillar addresses **new ways of working**, specifically around empowering their diverse workforce of 140,000 employees. Jabil has developed persona-based approaches for different roles—from operators and technicians to finance controllers, supply chain officers, and buyers—ensuring that GenAI solutions are tailored to specific user needs rather than deployed as one-size-fits-all tools. The **digital foundation** forms the third pillar, acknowledging that AI success depends on solid infrastructure, proper data handling, data curation, and establishing semantic layers. This foundational work is often overlooked in discussions of GenAI deployment but is critical for production systems. **Continuous improvement** as the fourth pillar reflects their understanding that transformation isn't a one-time event but an ongoing process involving people, process, and systems working together in new, more efficient ways. Finally, **responsible AI use** addresses governance concerns. A particularly insightful revelation from the presentation was Jabil's journey in recognizing that IT doesn't "own" the data—functional business units do. This led to the establishment of a Data and AI Council with senior VP and SVP-level representatives from each function who collectively make decisions, define roles and responsibilities, and design AI and data policies covering acceptable use guidelines, do's and don'ts, and data safeguarding requirements. ## Technical Architecture and Amazon Q Business Implementation Jabil uses Amazon Q Business as the primary GenAI platform, leveraging its capabilities for document retrieval, natural language querying, and integration with existing enterprise systems. The choice of Amazon Q Business appears driven by several factors: The built-in security and permission handling was emphasized as critical for their enterprise deployment. Q Business maintains document-level permissions, ensuring employees can only access information they're authorized to see. For a company operating across 100 locations with sensitive manufacturing and procurement data, this enterprise-grade security was essential. The platform's ability to connect to 40+ data sources, including custom connectors for legacy systems, enables Jabil to integrate their existing document repositories, supplier databases, and operational systems without requiring data migration or consolidation. Multilingual support proved particularly valuable given Jabil's global footprint. While documentation is standardized in English, many operators across their locations don't speak English fluently. Q Business's translation capabilities allow operators in Poland, for example, to query in their native language and receive translated responses. ## Production Use Cases ### Ask Me How: Manufacturing Shop Floor Assistant The first production use case, humorously named after an operations VP called "Me How," addresses the challenge of supporting operators on manufacturing floors. Jabil operates SMT (Surface Mount Technology) lines, CNC machines, and injection molding equipment that generate thousands of error codes. The problem is straightforward: operators encounter error codes like "nozzle error 001" but may not know what it means or how to resolve it, especially new operators who cannot memorize thousands of potential errors. Previously, they would need to stop the line and wait for a supervisor or technician. With Q Business, operators can query the system in natural language, receive explanations of error codes, and get resolution steps—all without stopping production. The system indexes documents and tracks query patterns, building a knowledge base that captures not just the original documentation but also the solutions that have been applied historically. The results include reduced downtime, increased efficiency, and elimination of dependencies on supervisors for routine troubleshooting. The multilingual capability means operators can receive answers in their native language, further reducing friction. A key insight from this deployment is May's vision for the future: moving beyond a reactive "library" model to a predictive, closed-loop system that can anticipate problems before they occur based on patterns in equipment behavior. ### Procurement Intelligence Platform (PIP) with Q Integration The second use case addresses Jabil's massive procurement operations—$25 billion in global spend across 350,000 suppliers. The PIP system, enhanced with Amazon Q, helps buyers and category managers make informed purchasing decisions. The genesis of this use case came from Jabil's chairman, who recognized the potential of combining internal demand signals from 400 customers with external market intelligence from traders and brokers. By marrying these data sources, the system can identify supply-demand imbalances and recommend proactive purchasing decisions. For example, if the system detects increasing demand for capacitors while supply-side intelligence suggests an impending shortage, it can alert procurement officers to stock up before prices rise or availability drops. The system can consolidate internal and external information, demand signals, and supplier data to provide comprehensive market intelligence. The demonstration showed users querying for "latest news in the electronic industry for October 2024" and receiving consolidated reports with supplier-specific insights and customer trend analysis. This transforms procurement from a reactive function to a strategic, intelligence-driven operation. May noted that the chairman's imagination was sparked to consider going even deeper—understanding the raw materials (sand, iron ore) needed to produce components like capacitors—though this remains aspirational rather than implemented. ### V-Command: Supply Chain Services Platform The third use case, V-Command (Virtual Command), supports Jabil's emerging business line of providing supply chain services to other manufacturers, both customers and non-customers. This platform encompasses supply chain as a service, logistics as a service, and procurement as a service. The Q-powered SCM Assistant helps supply chain practitioners with: - Calculating optimal landed costs for customers considering distribution center locations - Identifying supply chain risks from external events (earthquakes, typhoons, hurricanes) - Finding alternative suppliers within geographic proximity when disruptions occur - Providing market intelligence to support sales conversations A demonstrated use case showed querying comprehensive information about a customer ("Method Electronics"), pulling together internal customer data with external information to support supply chain service sales discussions. The business impact focuses on reducing sales cycle time by providing convincing, data-backed insights to potential customers, ultimately increasing revenue and margins for Jabil's supply chain services business. ## Organizational and Governance Considerations Beyond the technical implementation, the presentation offers valuable insights into the organizational structure supporting LLMOps at scale: The Data and AI Council represents senior leadership from each function and serves as the decision-making body for data and AI initiatives. This council collectively designed the AI and data policy, continues to refine governance as the technology evolves, and ensures that data ownership remains with functional units rather than IT. A GenAI sub-council (referred to as "baby tiger cubs from the big AI council") focuses specifically on generative AI initiatives. Members must present use cases with clear ROI justification, and the council collectively decides which initiatives receive resources. Once approved, teams are empowered to execute following established guidelines. The creation of a dedicated "value tracker" role—a financial controller pulled from existing teams—specifically to measure and report ROI on transformation initiatives demonstrates the seriousness with which Jabil approaches proving value. ## Key Decisions and Lessons Learned Several strategic decisions stand out from the presentation: **Vendor vs. build decision**: Jabil made "a very hard decision" to stop all internal chatbot development and focus on vendor solutions like Amazon Q. This was driven by recognition that partners like AWS invest heavily in R&D, security, and data protection—investments that would be difficult to replicate internally. **AI literacy investment**: Using their Workday platform, Jabil rolled out AI literacy training to all employees, ensuring a common vocabulary and understanding when discussing data, AI, and GenAI. May conducted education sessions with management and even the board of directors, leveraging the ChatGPT hype as an opportunity to build awareness and support. **Scaling mindset from the start**: May explicitly stated, "I'm not interested in experimenting... I'm interested in actually productizing it and making sure that the solution that we make, it needs to replicate throughout all my 100 locations." This production-first mindset shaped how solutions were designed and deployed. ## Integration with Amazon QuickSight The presentation mentioned that Jabil uses both Amazon Q and Amazon Q QuickSight, and May expressed enthusiasm about announcements (likely from the AWS keynote the same day) regarding tighter integration between Q and QuickSight. This suggests future directions for combining conversational AI with business intelligence dashboards for more comprehensive analytics capabilities. ## Assessment and Considerations While the presentation provides compelling use cases, it's worth noting some limitations in the information provided: Specific quantitative metrics for results were not shared beyond general statements about "reduced downtime" and "increased efficiency." The solutions appear to be relatively early in deployment, with May acknowledging they are still in "baby steps" and continuing to evolve. The presentation comes from a vendor conference (AWS) with Jabil as a customer case study, which naturally emphasizes success. Challenges, implementation difficulties, or solutions that didn't work as expected weren't discussed. The emphasis on governance and organizational structure, however, suggests a mature and thoughtful approach that increases confidence in the sustainability of these initiatives. The explicit rejection of "POC culture" and insistence on value-driven, production-grade deployments indicates lessons learned from earlier, less successful AI initiatives. Overall, this case study provides valuable insights into how a large, complex manufacturing enterprise approaches GenAI deployment with proper governance, clear business objectives, and a focus on scaling solutions across a global organization.