Tackling EPR reporting with AI and experts
Smarter Packaging is a collaboration between Washburn Consulting and Smarter Sorting
As packaging extended producer responsibility programs come on-line in the U.S., a familiar name in the consulting realm and a software company specializing in AI classification have teamed up to develop an AI model to help with the data side of EPR compliance.
Smarter Packaging is a collaboration between Washburn Consulting and Smarter Sorting. Michael Washburn, founder of Washburn Consulting, said it was the next logical step after spending several years focused on the non-data aspects of extended producer responsibility (EPR) compliance.
“We were leaving the data collection, analysis and reporting to others, because we did not have a depth of expertise in data management,” he said, but over the course of the past year, it became “painfully obvious to me that this made no sense. Why would we not be supporting our clients in a holistic way?”
Washburn reached out to Charlie Vallely, who is a cofounder and CEO of Smarter Sorting, with a decade of experience in classification, because “Charlie’s team has so much capacity and muscle memory around classifying products, which is such a close cousin” to EPR compliance.
Vallely said he had been keeping an eye on the ways his current work was running parallel to EPR packaging data compliance needs, and he thought “we can do something cool here.”
Together, they built an AI-powered model that can take data from clients – however incomplete – and work to fill in the gaps, classify each item, specify the reporting details by state regulation, and generate a report.
Oftentimes, producers do not have all the data required in each report. Washburn said Smarter Packaging can handle “anything resembling a spreadsheet.”
“Maybe they have a description. Maybe they have a UPC code,” he said. “Smarter Packaging can go out and hunt for all the available data on the web and it says ‘hey, is it this?’ And it can give us an image and we can say ‘yeah, it’s that,’ or ‘no, it’s not, show me another,’ and find the data that are missing, validate that against other data that we already have, and fill in those blanks.”
It can go a step further, Washburn said, and break down an item – say a bottle – into the bottle body, the label, the cap, and the safety seal, then match those components to each state’s covered material list. That’s “a terribly long, painful process” to do manually, he added.
For companies with thousands of packaging items, that task is even more daunting, Vallely said, but the model has access to over 500 million UPCs he’s built up over time, and can crawl the web to grow product coverage continuously.
Another vital aspect of the AI model is the built-in validation, both automatic and human.
“You need a way to validate or figure out the product truth, so as we get data from the internet, it’s really important to us to validate for this exact product,” Vallely said. “So we’ve built these validation layers to confirm that.”
In addition, as each item is classified by the AI model, it also gets a written reasoning of how the model arrived at the decision. That can help producers who may need to explain the choices made to regulators, and may also help regulators, Vallely suggested, because “they’re trying to figure this out, too.”
Washburn said as the technology and reporting evolves, having a methodology that clarifies which parts of the data were generated by an AI model and what it was validated against will be vital. And nothing is submitted without being validated by both the consulting team and the producer.
“Expert tools trained by experts – this isn’t just ChatGPT going on here,” he said, adding that getting the scope correct and data as precise as possible is “the cornerstone of compliance” and key to “putting together a credible, defensible, valid, verifiable report.”
Washburn had previously noted his concerns with producers turning to generic AI chatbots for help, because of the complexity – and stakes – of EPR programs.
“If you can get an AI chatbot to tell you weights for a particular item, it may or may not be accurate,” he said. “There’s no way to validate. And what that chatbot isn’t going to do for you is tell you is that in scope for you in Oregon versus Colorado? And if it’s food serviceware, are you the first seller, is it branded, is it not branded, is it exempt or excluded?”
“If the scope is wrong, if the tool is not informed about the context, you’re running into some risk there which translates directly into fees or noncompliance if you miss something,” he added.
To ensure that the context is correct, Vallely said the AI model will ask a series of questions if the data is not at a certain standard.
“If the inputs were lacking, it’s going to ask you contextual follow-up questions,” he said. “Can you confirm the plastic weight? There’s an exception for small format, we estimated this, can you confirm that? Are all sides less than two inches? Then you submit that and reclassify on the fly.”
The model can also help create source reduction strategies by identifying high portfolio concentrations of certain items and suggesting material alternatives, as well as flagging unnecessary items.
And as reporting requirements and regulations change, the model can be rapidly adjusted.
Overall, Washburn said he hopes tools like Smarter Packaging help soothe producer anxiety and reduce “blank page syndrome,” where starting the task feels impossible.
“We have to make this accessible at a human level, and that’s where the combined expertise of these two companies create a differentiation from other organizations that are so focused on the data exercise that they’re not supporting people in the very human journey of understanding this in the first place, helping with it in the second and thirdly being able to move forward in a productive and efficient way,” Washburn said.
Vallely said he’s looking forward to the future, when the model has been trained on even more items and starts to function more like an operating system, “where you just add your packaging and it’s defacto compliant when you add it.”
“I hope as part of that, that it should be easier to make regulations that work for all parties because you have a common language,” he said, so as new laws come on-line, the lift is lessened.
While there is a feeling of competition in the EPR data reporting and compliance space right now, Washburn said “there’s certainly more room to grow because the demands are just extraordinary.”
“There are so many producers who are not engaged, and are at risk of being out of compliance, that there’s plenty of work for everybody,” he said. “We could have 10 more firms doing this and there would be market share to go around.”