How a commercial marine purchasing agent went from manually deciphering unstructured RFQs to an AI-powered matching system that turns messy, multi-format requests into clean, action-ready data, almost instantly.
The Buying Network is a purchasing agent and distributor serving the commercial marine industry: fishing vessels, seafood processors, and ships operating in remote locations like the Bering Sea. They source everything from galley supplies and rain gear to marine chemicals and engine room filters. When a vessel or processor needs supplies, they send an RFQ.
The problem was how those RFQs arrived. A vessel captain preparing for a season does not send a neatly formatted spreadsheet with part numbers. Requests came in as PDFs, Excel files, scanned documents, or long emails, filled with inconsistent descriptions, missing part numbers, shorthand, and industry jargon.
Before this engagement, processing a single RFQ meant their team had to read every line, interpret what was actually being asked for, search inventory and vendor catalogs to find the right match, and manually enter each item into their quoting system. Some RFQs consumed up to six hours before a quote could even be drafted.
The pattern is familiar to any mid-market operation that sits between multiple parties and unstructured inputs: the work is not the work. The work is getting the work into a state where it can be worked on.
PDFs, spreadsheets, and emails with inconsistent descriptions, missing identifiers, and industry shorthand. No two RFQs looked the same.
All of that time was spent just getting the data into a usable state, before any actual quoting work could begin.
The process depended on individual expertise that was nearly impossible to scale or delegate to newer team members.
The unstructured, unpredictable nature of field requests made every existing solution inadequate.
Atelion built a custom AI-assisted tool designed specifically for how The Buying Network processes incoming work. The system takes the messy, unstructured reality of real-world requests and converts them into clean, structured, action-ready data with minimal manual effort.
The system reads incoming requests regardless of format (PDFs, Excel files, emails) and extracts each line item automatically. No more manually copying information row by row. The AI interprets inconsistent naming, shorthand, and partial descriptions to understand what the customer is actually requesting.
For each extracted line, the system searches across multiple data sources in priority order: the customer’s past purchase history first, then the internal master database, then any loaded supplier catalogs. Every suggested match is labeled with its source, so the team knows exactly why a particular recommendation was made.
Every match comes with a confidence score. High-confidence matches can be approved quickly; low-confidence or unmatched lines are flagged for human review. The interface places each request line alongside its suggested match for side-by-side comparison, turning what used to be a research task into a simple review-and-approve workflow.
When the team approves a match, the system remembers it. The next time a similar line appears in a future request, whether from the same customer or a different one, the tool draws on that confirmed history to suggest the same match automatically. Over time, the system gets faster and more accurate, codifying the team’s institutional knowledge into the software itself.
The impact was dramatic and immediate. Requests that previously consumed up to six hours of manual processing now move through the system in under five minutes. With ~90% first-pass match accuracy, the team’s role shifted from manual data entry and product research to reviewing and confirming AI-suggested matches. A fundamentally different and faster workflow.
The time savings were only part of the story. The reduction in manual processing freed enough staff capacity to redirect roughly $80,000 a year in labor toward higher-value work. By codifying the team’s institutional knowledge into the matching engine, The Buying Network removed a critical dependency on individual expertise. Accurate matching no longer required decades of industry-specific experience to be consistent and fast.
For a 25-year-old business operating in a demanding, relationship-driven industry, this project did not just save time. It turned the intake bottleneck into a competitive advantage.
“I was brought in to help The Buying Network find someone who could pull off an AI solution that felt like a real stretch. I really did not know if it was possible, but the Atelion team absolutely delivered. They were on time, on budget, and super clear. Most importantly, they stuck with us through the real-world challenges to get it adopted by the people who needed to use it. If you’re looking for a team that’s solid and follows through, I’d highly recommend them.”
Any operation that sits at the intersection of unstructured inputs, high transaction volume, and expert judgment has this same bottleneck. The work is not the core value-add. It’s all the interpretation, extraction, and preparation that has to happen before the real work can start.
That’s the pattern Atelion is built to solve. We do not replace your team’s expertise. We capture it, scale it, and free your people to spend their time on the decisions and relationships that actually move the business forward.
Tell us where the work piles up before the actual work can start. We’ll tell you whether automation and AI can help, and whether we’re the right team to build it.
Talk with us