About Order.co
Order.co is a procurement technology innovator headquartered in New York City with 160 employees across the U.S. Their unified platform streamlines vendor relationships, order placement, and payments for thousands of suppliers, serving industries from healthcare to hospitality to retail. Their core capabilities include vendor catalog integration, automated purchasing workflows, spend analytics, and consolidated invoicing, removing friction from procurement and securing customers the best possible pricing.
Global brands like Hugo Boss, Miniso, Lark Hotels, and I Love Vacations trust Order.co to simplify purchasing across complex vendor ecosystems. With demand surging toward $1B in purchasing volume annually, the company is reshaping enterprise procurement in one of the fastest-growing B2B markets.
The Customer's Challenge
Order.co, a trusted leader in procurement, operated with highly manual, browser-based ordering workflows that strained operational efficiency. Each transaction relied on traditional RPA-style automation that required a substantial amount of engineering cost to maintain, with human agents as fallbacks.
These processes led to escalating labor costs, frequent errors, and slow cycle times.
As Order.co’s client base and SKU volume increased, so did operational headcount and the risk of exceptions. Despite these obstacles, including shifting UIs, CAPTCHAs, and complex payment flows that defeated traditional automation, Order.co pushed forward to find a scalable solution for its customers.
phData's Solution
phData delivered an Agentic AI-powered automation solution in just six weeks, leveraging a powerful combination of Amazon Nova-Act for intelligent action generation and PlaywrightMCP for robust browser automation. This dual-tool approach provided the foundation to successfully navigate the end-to-end procurement processes for all seven vendors.
The core of the system is Nova-Act that acts as the intelligent brain for the operation. To navigate the vendors’ sites, the system mapped each process into golden paths, which are stored as Vendor Agent Plans. These plans provide the AI agent with proven strategies for executing the correct login, cart management, checkout, and payment steps.
To ensure accuracy and prevent errors, every instruction generated by the LLM is first checked by an Action Validation module before execution. If an action is invalid, an immediate feedback loop prompts the LLM to self-correct and generate a new action. For complex scenarios with low confidence, like a CAPTCHA, the system intelligently triggers a Handoff to Human, ensuring process continuity.
The solution is engineered to enhance its performance over time. A nightly review of Golden Paths analyzes successful runs to refine the agent’s strategies, while performance data is collected to continuously fine-tune the core LLM itself. The entire application is containerized in an ECS Pod for scalability, and comprehensive Action Logging provides full visibility for auditing and analysis.
The Full Story
Order.co is a fast-growing procurement organization serving a diverse vendor network. Prior to phData’s involvement, Order.co’s teams spent hundreds (if not thousands) of hours manually executing procurement orders on various vendor sites.
Each vendor required a unique workflow, and frequent site updates created expensive exceptions and delays. Order entry and onboarding for new vendors was slow, complex, and demanded increasing headcount for each additional client. These bottlenecks limited Order.co’s ability to scale and led to a high exception rate in order fulfillment.
Recognizing the need for specialized guidance, Order.co brought in phData, a trusted provider of data and AI services with deep AWS expertise. AWS was chosen as the preferred platform for its advanced AI/ML services and the ability to provide secure, high-performance, and globally scalable infrastructure.
phData partnered with Order.co to automate the vendor workflows using a modular agentic architecture across two specific AI paradigms: LLM Tool Calling and MCP.
Tool Calling
- Amazon Nova-Act was selected for its native natural language API and ease of integration, enabling rapid prototyping.
- Stagehand and Playwright were also integrated as callable tools by LLMs, performing well across procurement steps.
MCP
- Anthropic Sonnet 4, available through Amazon Bedrock, powered agent reasoning and step-by-step browser action planning.
- Using Playwright MCP, the agent executed actions in real browser sessions, reliably managing diverse site layouts, logins, carts, checkouts, and secure payment flows.
- Each vendor process was mapped as a “golden path” and annotated for agent guidance.
- Validation layers ensured only accurate orders were submitted; mismatches between planned and actual order details triggered halts and review.
- Automated audit logging and session screenshots guaranteed traceability and supported further model refinement.
The final demo successfully placed real orders across seven diverse vendor platforms.
Results
By automating procurement workflows with Agentic AI, Order.co achieved:
- Achieved a 100% success rate in end-to-end order automation for all seven vendors during the initial proof of concept, establishing a strong foundation for scaling the solution across the entire vendor base.
- Orders that previously took hours are now processed in minutes.
- Streamlined vendor onboarding, automation playbooks, and golden path workflows enable rapid onboarding and retraining for new vendors.
- Comprehensive auditability and traceability with automated action logging and session screenshots.
- A scalable foundation, ready to expand automation coverage across Order.co’s vendor network without increasing operational headcount.
Success Rate
End-to-end order automation for
7 out of 7 vendors in the initial PoC
Why phData?
Order.co selected phData for its proven expertise in AI-powered automation tailored to procurement complexity.
I was really impressed that phData brought in highly talented, high-quality people into our very first calls. Those same engineers are helping us build the ability for us to place orders entirely automatically on any vendor’s website, regardless of what that website looks like. I would highly recommend phData to anyone in the industry.
Take the next step
with phData.
Learn how phData can help solve your most challenging data analytics and machine learning problems.
