The Watson Weekly is sponsored by Avalara — the agentic AI platform automating global tax and compliance for leading eCommerce brands with simple plans, transparent pricing, and complete coverage across Shopify, Stripe, WooCommerce, BigCommerce, and more.

I sat with Kyle Bertin's story about the footwear CFO for a while after we stopped recording. A 35% return rate, and the man said he did not care. He was not being lazy about it. He gave four reasons, and every one of them held at the time: sub-20% COGS out of Vietnam, duty-free entry under Section 321, inventory in transit financed at effectively zero on a public balance sheet, and investors who paid for growth rather than profit.

That list has nothing in it about warehouses or process. The CFO had done the arithmetic and concluded that returns recovery was not worth his attention, which was the correct conclusion given free capital and cheap goods. All four of those conditions have since gone away, so what looks today like a decade of operational neglect was mostly a rational response to conditions that stopped holding somewhere around 2023. The conclusion is what nobody updated.

Kyle's math on a $100M apparel brand is uncomfortable in a useful way. A 25% return rate puts $25M of GMV back through the receiving door. Restock half of that, which is roughly where unoptimized operations sit, and $12.5M never returns to full price. Push restock to 80% and you have freed $7.5M of inventory, which after sell-through and gross margin gets you $3 to $5M of net income.

Compare that against what it would take to add $3M to the bottom line from the revenue side. At typical apparel margins you need something in the range of $25 to $35M of incremental sales, and nobody is finding that in a quarter. I have watched a lot of brands spend a lot of money chasing growth that was already sitting in a corner of their own fulfillment center, waiting for someone to inspect it properly..

I asked Kyle the AI question expecting the usual answer and got a claim about onboarding speed instead. Two Boxes used to need his cofounder Evan sixteen hours to hand-code the decision tree for one merchant's inspection rules. That work now takes seconds, and they onboarded 280 merchants in twelve months with a team of one.

Kyle was direct, this is not automation replacing warehouse labor, and he does not see that arriving on any timeline he can name. What changed is the implementation cost that kept returns technology out of the mid-market. If a deployment costs sixteen engineer-hours per merchant, you can only afford to sell to merchants large enough to justify it. Take that cost to near zero and the addressable market is a different market.

The piece I have not resolved is Kyle's figure that only about 5% of returned items are genuinely unsalvageable, with 85% of single-defect items remediable. That is data from merchants who bought software specifically to improve that number. Whether it survives contact with the rest of the industry is an open question, and I have not seen anyone answer it.