TraceGains recently launched its Intelligent Document Processing (IDP) offering, an AI-driven solution for material compliance and lot-level ingredient quality control. The objective of IDP is to streamline Certificate of Analysis (CoA) processing for the company’s network of 90,000 global supplier locations and over 600,000 ingredients and products.
IDP was built on OpenAI’s foundation model, and was trained and validated against global food and beverage CoAs. Paul Bradley, senior director of product marketing, TraceGains, said CoAs pose a challenge given they are critical for ensuring that the product reaching a manufacturer’s facility is safe, free of contaminants and in line with specs.
“Traditionally, however, CoAs can appear in a variety of non-standard and confusing formats, making it cumbersome and time-consuming for brands to ensure that received product is appropriate for use,” said Bradley.
“TraceGains currently has the market leading product for automated CoA data extraction in the food and beverage industry using older extraction technologies, and we have a very deep understanding of CoA processing and the needs of customers in this space (which goes far beyond data extraction). With the addition of our advanced AI capabilities, we can help brands extract non-standard data from CoA formats and automate the comparison to specifications, flagging potential issues and saving brands time and money while reducing risk.”
Bradley said there’s a strong use case for AI-driven IDP, which opens the door for many other data extraction and automation use cases in the future.
“Within the Dietary Supplement world in particular, IDP could be used to support elements of, for example, label verification processes, analyses of adherence to GMPs, and cross-checking supplier documentation against regulatory standards,” he said.
Accuracy
TraceGains’ IDP solution is highly accurate across a wide range of CoA formats and can handle the complexity of shipments with multiple lots or even multiple suppliers and items, said Bradley.
“In large scale internal tests, we’ve matured our out-of-the-box technology to achieve high accuracy right out of the gate. We’ve also introduced methodologies for customers to improve that score based on customer-specific configuration updates to the IDP system.
“We plan on creating a ‘best of both worlds’ approach with respect to human intervention and feedback. We believe the future of AI in enterprise should be configurable and enable customers to optimize AI solutions for their specific needs, whether it relates to IDP or, for example, compliance and formulation agents. In the IDP case, humans can ‘help’ the AI by providing feedback for specific data points that prove difficult for out-of-the-box approaches to extract,” he said.
Bradley added that the broader supplier compliance product in which IDP operates to extract CoA data still does require some human configuration. For example, they need to know what data points users want to collect from the CoAs they process.
“However, this is far less effort than traditional, zone-based Optical Character Recognition (OCR) solutions, speeding the configuration process up dramatically,” he noted.
Addressing fraudulent CoAs
One issue that continues to plague the dietary supplement industry is fake and misleading CoAs. Bradley said many brands struggle to simply find the time and resources to compare every inbound CoA to the appropriate spec.
“This is where automation is valuable, helping humans in the factory focus on exceptions and flagged issues that require close attention,” said Bradley.
“The good news is that TraceGains can help brands identify fake or misleading CoAs days prior to receiving the product, allowing bad shipments to be refused before they ever arrive and reducing the risk of costly factory disruptions. When the CoAs are received by TraceGains, the system looks for specific key terms on the document to recognize it, and then it gets run through the customer’s specifications. A fake CoA will immediately get flagged as out of spec, and notifications will be triggered for the appropriate people to take action.”
CoAs are just the beginning
With IDP’s capabilities of addressing specialized tasks effectively, the company is looking beyond CoAs.
Within a year, TraceGains plans to begin releasing sophisticated AI and agentic systems that deploy large language models (LLMs) in combination with other machine learning models and deterministic computational methods to support a range of quality/compliance and formulation tasks.
“This is an area of heavy investment for TraceGains that will be highly differentiated due to our network, partner ecosystem and data advantage,” Bradley said.
