14:30
Amazon seller strategy
The headline story is that Amazon has turned its shopping AI (the age of Amazon, rebranded from Rufus) into a licensable product sold through AWS, letting any retailer stand up a tailored AI shopping assistant in as little as 60 days — the same build-then-sell playbook Amazon used with cloud, cashierless checkout, and logistics — and the episode frames this as reshaping how sellers must optimize for AI agents, alongside a run of time-sensitive, actionable items: recoverable IEEPA tariff refunds via CBP, a new auto-buy feature competing with Subscribe & Save, and fresh data on TikTok Shop Europe and YouTube title performance.
Amazon is licensing the technology behind Alexa for shopping (formerly Rufus) through AWS so any retailer can launch its own AI shopping assistant tailored to its catalog and branding in as little as 60 days; Kate Spade (under Tapestry) is the first customer, using it to build a gifting assistant.
The host frames this as the AWS playbook repeating: build internally to solve Amazon's own problem, then sell the solution to the industry, competitors included, as Amazon previously did with cloud, cashierless checkout, and its logistics network.
The product is deliberately sold through AWS rather than Amazon's retail arm, because retailers who wouldn't share catalog and customer data with the Amazon storefront may already trust the cloud unit they pay monthly.
The host notes a tension: Amazon argues brands know their products and customers better than any general-purpose AI and shouldn't hand off their shopping experience to an intermediary, while Amazon itself is that intermediary for its third-party sellers and its 'buy for me' feature reaches onto rival sites to complete purchases.
Andrew Erickson (founder of Inventory Hero) got $25,242 wired back from CBP as a tariff refund plus interest, after the Supreme Court struck down the IEEPA reciprocal and 'final' tariffs in February; CBP is repaying that money via a tool called CAKE inside the ACE portal.
Erickson's six-step refund playbook: (1) confirm you're owed money by scanning 2025 CBP form 7501 for HTS codes 9903.01 or 9903.02; (2) confirm you're the importer of record (your EIN on line 22 of the 7501) — if you ship DDP and your supplier paid duties, they were the importer of record and CBP refunds them, not you; AGL users remain the importer of record even though Amazon brokers the shipment, and Amazon Customs and Trade will email a pre-built CSV and offer to file the claim; (3) check timing — Phase 1 covers entries unliquidated or liquidated within the last 80 days, Phase 2 has no timeline yet, and a protest under 19 USC 1514 has a hard 180-day window from the liquidation date; (4) open an ACE portal account and enroll a US bank account specifically for ACH refunds (ACH set up for paying duties doesn't count); (5) file the CAKE claim in the CAKE tab in ACE or authorize a broker; (6) get paid — refunds typically hit ACH 3 to 5 weeks after liquidation.
The refund is a cost reduction, not income — it reduces COGS in the year it relates to, not the year it lands, and sellers are advised not to keep reordering at the old, tariff-inflated landed cost.
New TikTok Shop Europe data: the UK leads with about 52.5 million euros (Jan/Feb, top 50 shops), with 82% coming through creator affiliates; Germany is second (~12M euros), France third (~7.5M euros), then Italy and Spain; Spain has the highest creator dependency (87% affiliate) while Italy is the outlier with about 34% of sales from self-operated stores.
The Amazon Orders Intelligence Dashboard tool breaks orders down by state, city, zip code, hour of day, average order value, SKU, ASIN, business orders, and Subscribe & Save; the host cites a case where a regional sales dip traced to delayed shipping lead times in one area, which tanked conversion, then ranking, then sales — invisible from total revenue alone.
Amazon rolled out an 'auto buy' feature nationwide: a shopper sets a maximum price and a start/stop watching window on a product page, and Amazon purchases automatically the moment the listing hits that price — one unit, one-time, FBA only, no coupons or promos; Vanessa Hong was among the first to spot and share it.
Auto buy and Subscribe & Save now compete for the same reorder: since auto buy has no promo exceptions, any discount a seller runs clears the auto-buy queue at the deepest price instead of converting that shopper into a subscriber, and auto-buy shoppers stop actively comparison-shopping, which makes demand-signal data look quiet even though those buyers are still in-market.
The suggested countermeasure is to make Subscribe & Save the easier choice before a shopper ever sets an auto-buy trigger: a subscription discount that beats waiting, a stable list price so the 90-day-low trigger stays out of reach, and reliable FBA stock.
Jake Thomas analyzed over 800,000 YouTube titles to rank the ten most powerful title words by performance lift: never (+3%), habit (+4%), stop (+4%), only (+9%), change (+9%), forever (+12%), regret (+13%), every (+14%), any (+16%), and the current year 2026 (+30%, #1).
Additional brief items ('hot picks'): Shark Ninja mitigated tariffs through supplier diversification; Amazon launched a 'know before you buy' feature on product pages; Walmart credits its AI assistant Sparky with lifting sales and average order value; the EU fined Temu 200 million euros over illegal product risk failures.
The 'Stump Bezos' trivia answer: revenue per seller in the US Amazon marketplace is about $200,000 higher than in the next-largest marketplace.
Routing the new AI-shopping-assistant product through AWS rather than Amazon's retail division is framed as a trust-engineering move — retailers who'd refuse to hand catalog and customer data to a retail competitor may already trust the cloud vendor they pay every month.
The host highlights a structural irony: Amazon tells other retailers not to cede their shopping experience to an intermediary AI, while Amazon itself is exactly that intermediary for its own third-party sellers via its buy-for-me feature.
Auto buy quietly distorts demand analytics — a shopper who sets a trigger and walks away stops comparison-shopping, so seller search-and-demand data goes quiet on customers who are still very much in-market rather than genuinely churned; reading that quiet as churn leads to the wrong call.
Because auto buy has no promo exception, every discount a seller runs now serves double duty as an accidental one-time-buyer magnet, actively working against converting deal-seekers into Subscribe & Save subscribers.
DDP shipping terms create a specific trap in the tariff-refund process: if a supplier is the importer of record, CBP pays the refund to the supplier, not the seller, turning recovery into a private negotiation rather than a straightforward CBP claim — while AGL sellers are explicitly exempt from this trap since they remain importer of record even with Amazon brokering.
The 180-day protest window under 19 USC 1514 runs from each entry's liquidation date rather than from the tariff ruling or Phase 2's (still undefined) rollout, meaning sellers past the 80-day Phase 1 cutoff need to track their own liquidation dates instead of waiting on Amazon or CBP timelines.
In the YouTube title-word data, simply inserting the current year (+30%) outperformed every other psychological hook tested — scarcity ('only'), universality ('any'), completeness ('every'), loss-aversion ('regret'), and permanence ('forever') — suggesting recency framing was the single strongest lever in that dataset.
The regional-order-breakdown case study shows a causal chain invisible in aggregate revenue: a shipping-delay problem in one region tanked conversion, which tanked search ranking, which tanked sales — a chain that only became visible once orders were segmented geographically.
«The big story today is Amazon just turned its shopping AI into a product and is now licensing it to other retailers through AWS. And that changes the game for everybody.»
— 00:30
«Amazon is handing over the architecture, starter code, and learnings, and says that a brand can launch something tailored to its own catalog and branding in as little as 60 days.»
— 06:23
«Build internally, solve your own problem, then sell the solution to the industry, competitors included.»
— 06:36
«Amazon is telling every other retailer not to hand their shopping experience to an intermediary, arguing brands know their products and customers better than any general-purpose AI. Meanwhile, Amazon is the intermediary for its own third-party sellers, and its buy for me feature reaches onto rival sites to complete purchases.»
— 07:08
«this may be the biggest one-time cash event your brand will see this decade.»
— 02:37
«If you ship DDP and your supplier included duties to your warehouse, they were the importer of record. CBP refunds them, not you, and getting it back becomes a private fight with your supplier, not a CBP claim.»
— 03:26
«They're not gone, they're in line at a price you haven't hit yet.»
— 10:41
«Just putting 2026 in your title was by far the strongest thing you could do.»
— 12:55
«Vulnerability is the birthplace of innovation, creativity, and change.»
— 13:38
«The answer is $200,000. Revenue per seller in the US is 200K above the next largest marketplace.»
— 14:03
Reception
No comments are available to assess audience reception.
The episode functions as a fast-moving trade-press digest that pairs a genuinely actionable, step-by-step IEEPA tariff-refund walkthrough and an early warning on the Auto Buy vs. Subscribe & Save conflict with lighter-touch relay of AWS's AI-shopping licensing move and third-party statistics (TikTok Shop Europe, the YouTube title-word study); its value is concentrated in the two segments built around a named process rather than in original reporting.
14:30