AWS Builder Center

Builder Center

Transforming an Order to Cash workflow with Karini AI and Amazon Bedrock

Swagelok Southeast Texas Success Story

Neel Mitra

Swagelok Sou theast Texas (SS T)  is one of the top five most significant and independent Swagelok sales and service centers worldwide. Serving customers across the upstream, midstream, and downstream oil and gas markets from Houston to Louisiana, their team has built a reputation for excellence in providing industrial fittings and value-added services. Their impressive client list includes energy companies and organizations like NASA and research facilities like the University of Texas’s MD Anderson Cancer Center.
At SST, the customer service representatives (CSRs) received sales inquiries via email or phone calls and manually searched through various systems and documents to gather the necessary information before responding. This process was not only slow but also prone to errors and inconsistencies. Recognizing the need for improvement, the company implemented a more automated solution using AWS Services. While this represented significant progress, it created technical debt, a complex solution requiring specialized AI development knowledge.
Thus, SST and Karini AI partnered to develop a prototype for an order-to-cash system using Karini's no-code agentic platform that surpassed the performance of legacy systems' purchase order generation by 70% on the sample set. After a rapid prototyping phase of two days, the SST leadership adopted Karini's platform. Instead of hiring specialized AI engineers, they recruited a recent MBA graduate passionate about artificial intelligence to lead the project. Within three weeks, they deployed a fully functioning intelligent order-to-cash workflow.
"With our AI foundation, I tell my CEO all the time that we can solve any problem now that they propose to us," remarks Peter Hory, CFO and CIO at Swagelok Southeast Texas. "That power and that culture that pervades is going to build upon the future for how we attack problems."

Beyond Simple Knowledge Management: The Unified Assistant using Multi-Agentic system

After the success of order-2-cash intelligent automation, the SST team wanted to tackle the knowledge management problem across multiple systems. Employees had to navigate through ERP, Salesforce, Zendesk, and Data Warehouse to understand customer priorities and critical business metrics. For example, if management wanted to understand potential problems with the top three opportunities, it required navigating to Salesforce and then researching potential tickets in Zendesk for issues or engagement. Instead of creating isolated systems for different functions, the SST team decided to develop a "unified assistant," an AI system that serves as a central hub for knowledge and actions throughout the organization.
This unified assistant not only provides information but also drives actions. It accesses information from various sources, including documents stored in an Amazon S3 bucket dating back to 2012, purchase orders, invoices, and credit memos. Moreover, it integrates with multiple enterprise systems, including SAP, Data Warehouse, Standard Operating Procedures, Salesforce, and Zendesk.
The architecture utilizes a sophisticated "domain router" that directs queries to the appropriate specialized agents based on the nature of the request. For instance, when a user asks about sales performance, the system recognizes this as a Salesforce-related query. It routes it to the appropriate agent, which then constructs the proper query and returns relevant information.
The system serves different stakeholders throughout the organization:
● Management can quickly access performance metrics, business forecasts, customer trends, and team performance without navigating multiple systems or waiting for reports.
● Customer service representatives benefit from instant access to customer histories, previous orders, and support tickets through their existing Zendesk interface.
The unified assistant seamlessly integrates with existing workflows. The end user application implementation utilizes Microsoft Teams, enabling employees to interact with the AI assistant through a platform they already use daily. It integrates directly with Zendesk for order processing, providing AI-generated communications and context without requiring users to learn a new interface.

Karini AI Deployment on AWS

Instead of requiring the team to assemble dozens of different components, the Karini AI platform offers a unified foundation that serves as a system of record. The Karini AI platform runs on Amazon EKS, a scalable compute service, and leverages Amazon Bedrock for large language model (LLM) inference, knowledge bases, and Guardrails for AI safety.
The platform is deployed in the customer's Virtual Private Cloud (VPC), ensuring data security and compliance. The Karini AI platform enables rapid proof-of-concept development within a few hours, transitioning to a minimum viable product (MVP) in about a week, and seamless scaling to production. The system employs autonomous AI agents for complex workflows that can perform specific tasks or answer particular types of questions. These agents communicate with each other, creating multi-agentic workflows. For example, when processing an order, one agent might extract information from an email, another might check inventory in SAP, and a third might update the customer record in Salesforce.

Transformative Results

The results of this implementation have been remarkable. The Swagelok SST experienced immense value in a short period.
  • 1544% expected ROI
  • 90% time savings per quarter
  • $1 million in savings compared to manual processes
  • 20-day projected payback period
Here are a few key learnings from Swagelok Southeast Texas for Business executives embarking on a similar generative AI Journey.
  • Experiment rapidly and fail fast. The ability to switch platforms in just one month allowed rapid testing and iteration without getting bogged down in lengthy implementation cycles.
  • Business knowledge drives successful AI projects. The Karini AI platform enabled business experts to become "citizen AI engineers" without specialized technical training, putting AI development in the hands of those who best understood the business problems.
  • Intellectual property protection matters. With many vendors seeking access to customer data and know-how, organizations must carefully guard the elements that make their company unique, such as prompts and agent workflows.
  • Seamless integration drives adoption. By integrating with existing platforms like Microsoft Teams and Zendesk, the company ensured employees didn't need to change their workflows to benefit from AI.
Perhaps most importantly, the implementation laid the groundwork for continuous innovation and problem-solving throughout the organization, transforming the company's approach to business challenges.
Please watch the full webinar  jointly presented by AWS, Swagelok Southeast Texas, and Karini AI here.

Summary

The Swagelok Southeast Texas implementation offers a compelling blueprint for manufacturing and industrial organizations facing similar challenges with operational inefficiencies. The approach demonstrates that with the right platform and methodology, enterprises can improve operational efficiencies and make enterprise knowledge more accessible and actionable than ever before.
Organizations looking to implement similar solutions should:
● Start with a clear business problem and measurable ROI metrics
● Choose platforms enabling rapid experimentation without extensive development
● Empower business experts with no-code tools to build AI solutions
● Focus on seamless integration with existing workflows
● Build a scalable foundation that can address multiple use cases
As industries worldwide grapple with operational inefficiencies and the need to preserve institutional knowledge, the Swagelok Southeast Texas story offers both inspiration and practical guidance.

Contributors:

To learn more about implementing similar solutions, contact Karini AI at www.karini.ai or reach out to your AWS account manager.
 
# gen-ai-assistant# idp# knowledge-management# nocode# agentic-workflows
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article