AI Referrals Are Splitting Your Customer Cohorts in Two

AI referral traffic is not a single cohort. Treat it like one and your return rate math collapses. Shoppers arriving from AI-generated recommendations — ChatGPT product suggestions, Perplexity shopping answers, Gemini comparisons — split into two distinct behavioral groups on first visit. One group returns. Repeatedly. The other disappears and never opens another email. The median buyer in the middle barely exists. That bifurcation changes how you build retention spend, how you assign LTV in your models, and which SKUs you push into AI-indexed content.

The Acquisition Cost You're Not Calculating

Pull your last 90 days of acquisition data by traffic source. Separate AI referral from paid social, paid search, and direct. Now look at 30-day repurchase rate by cohort. If you haven't done this yet, stop reading and go run it. AI-referred cohorts with low initial average order value churn at a rate that makes their CAC look catastrophic in arrears. AI-referred cohorts with mid-to-high AOV on the first transaction? Their 60-day retention can outperform your best email-reactivated segment. Two groups. Same traffic source label. Completely different NetPPM implications. Your attribution model is probably averaging them together and giving you a number that describes neither.

Platform Creep Is Eating Your B2B Channel

Home Depot and Lowe's are not building better stores. They are building software platforms for contractors — project management tools, materials ordering integrations, job-site delivery scheduling. That move reframes what it means to be a preferred vendor in the home improvement supply chain. If your brand sells into pro or contractor segments and you are not inside the workflow these platforms are building, your SKUs become comparison items. Easy to switch. No stickiness. The operators who will win that channel in 2027 are the ones who push SKU data, spec sheets, and inventory velocity signals into those platform APIs now — before the data moat gets dug.

Target's 3,000-SKU Signal

Target added approximately 3,000 food and beverage items in Q1 2026. That is not a category refresh. That is a deliberate traffic-frequency play. Grocery trips happen 2.3 times per week. Apparel and general merchandise trips happen far less often. Target is buying visit velocity with margin compression on food. For every brand competing for Target shelf space outside of grocery — home goods, apparel, beauty, wellness — that move tightens your promotional calendar competition and concentrates buyer attention on categories that drive footfall. Your sell-through windows in non-food categories just got shorter. Plan your Q3 reorder cycles accordingly.

Vans, ABG, and the Turnaround Operator's Dilemma

Vans posted a sliver of US growth. Not a recovery. A sliver. Authentic Brands Group just appointed a new CEO while founder Jamie Salter remains involved in strategic direction. Both situations share an operational subtext: brand turnarounds that depend on wholesale channel health are structurally slower than DTC-led recoveries. When your revenue is routed through retail partners, your velocity data arrives late, your markdown exposure is shared but your brand damage is not, and your cycle count on cultural relevance runs at wholesale cadence instead of consumer cadence. If you are operating a brand with more than 40% of revenue in wholesale right now, that lag is your biggest operational risk — not the macro environment.

Three Questions to Pressure-Test

First: Have you segmented AI referral traffic into distinct cohorts by AOV band — and are you running separate retention sequences for each, or one generic flow? Second: If Home Depot or Lowe's launched a contractor procurement platform tomorrow that your category touches, would your SKU data be clean enough to integrate within 30 days? Third: With Target compressing non-food sell-through windows, does your Q3 promotional calendar account for reduced buyer bandwidth at your top three retail accounts — or is it built on last year's access assumptions? Run the AI cohort split first. Everything else depends on what you find there.