Picture this: A customer clicks “return” on a pair of sneakers that didn’t fit. That single click triggers a cascade of decisions—where does the product go, who inspects it, can it be resold, should it be liquidated? For years, humans have been making these calls manually, one return at a time. Two Boxes, a Denver-based startup, just raised $3.2 million to let AI agents handle this entire workflow instead.
Assembly Ventures led the funding round, betting that returns processing is ripe for automation. And from a bot-building perspective, this makes total sense. Returns aren’t creative work—they’re decision trees at scale. Every returned item follows a similar pattern: assess condition, determine destination, update inventory, process refund. That’s exactly the kind of repetitive, rules-based workflow that AI excels at.
Why Returns Processing Matters for Bot Builders
If you’re building bots for e-commerce or logistics, returns are where things get messy. The forward flow—customer orders, warehouse picks, shipping—has been optimized for decades. Returns flow backward through the same infrastructure, but with way more variables. Is the item damaged? Can it go back to stock? Does it need refurbishment? Should it be sent to a liquidator?
Two Boxes is targeting 3PLs (third-party logistics providers) and retailers with their platform. These are organizations processing thousands of returns daily, each requiring human judgment calls. That’s a perfect use case for AI agents that can learn patterns, make consistent decisions, and scale without adding headcount.
What This Means for AI Architecture
From an architecture standpoint, a returns processing bot needs several components working together. You need computer vision to assess product condition from photos. You need decision logic that factors in the item’s value, condition, and market demand. You need integration with inventory systems, refund processors, and shipping APIs. And you need all of this to happen fast enough that returns don’t pile up in warehouses.
The interesting challenge here is handling edge cases. Most returns are straightforward—unopened box, full refund, back to stock. But what about the sweater with a missing button? The electronics with cosmetic scratches? The shoes that smell like they’ve been worn for a month? Training AI to make nuanced calls on these scenarios requires solid data and careful model design.
The 3PL Angle
Two Boxes plans to engage more aggressively with 3PLs using this funding. That’s smart positioning. 3PLs handle returns for multiple retailers, giving them massive data advantages. If you’re building a returns bot, you want to train it on diverse product categories, return reasons, and condition assessments. A 3PL processing returns for fashion brands, electronics retailers, and home goods companies sees all of that in one facility.
For bot builders, this highlights an important principle: your AI is only as good as your training data. Two Boxes isn’t just selling software—they’re building a system that gets smarter with every return processed. That’s the real moat in AI products.
Building Bots That Make Money Decisions
Returns processing bots need to make financial decisions, not just operational ones. Should this returned jacket be resold at full price, marked down 30%, or sent to a liquidator? That decision has real P&L impact. Get it wrong consistently, and you’re leaving money on the table or flooding your store with unsellable inventory.
This is where AI shines compared to rigid rule-based systems. A traditional system might say “any item with visible wear goes to liquidation.” An AI system can learn that certain brands hold value even with minor wear, that some product categories have strong secondary markets, and that timing matters—liquidate winter coats in March, hold them for next season in June.
What’s Next
Two Boxes says they’ll use the funding to advance their product roadmap and expand enterprise reach. For those of us building bots, this is a space worth watching. Returns processing sits at the intersection of computer vision, decision automation, and logistics optimization. It’s messy, high-volume, and expensive when done manually.
The broader lesson here: look for workflows where humans are making repetitive decisions based on observable criteria. That’s where AI agents can deliver immediate value. Returns processing is one example. Customer service triage is another. Inventory allocation, pricing decisions, quality control—these are all domains where bots can take over the grunt work and let humans focus on exceptions.
Two Boxes is betting $3.2 million that retailers and 3PLs are ready to hand returns processing over to AI. Based on the fundamentals, that’s a solid bet.
🕒 Published: