Goldman Sachs: Agentic AI Reshapes E-Commerce Intent Entry Points — Winners Determined by Five Key Capabilities
nashnova research
Goldman Sachs argues AI agents are shifting e-commerce competition from who owns traffic to who captures and executes purchase intent, a migration that may take three to five years and ripple across advertising, payments, and cybersecurity.
What does agentic shopping actually change?
Today consumers search, compare, and check out themselves. Tomorrow an AI agent understands the need, filters products, and triggers payment. This means → the value-chain battleground moves from page traffic to purchase intent itself.
Goldman condenses the thesis to one line: value should increasingly follow consumer intent.
The shift will not happen overnight — its timeline resembles e-commerce's own evolution cycle, three to five years or more. Winners are decided by five things: distribution, consumer trust, merchant participation, transaction infrastructure, and the ability to capture commercial intent.
Which purchases get automated first?
Paper goods, cleaning supplies, packaged food — low-risk staples go first. In plain terms = anything where a wrong buy costs you almost nothing is easiest for an AI to handle, including swapping brands when stock runs out.
Goldman estimates roughly $2.6 trillion in US consumer spending sits in high-likelihood categories (low-risk, repeat-purchase types). Another key number: every 2% of agentic penetration in offline card spending could add roughly one percentage point to e-commerce growth.
Hardest to automate: shoes, jewelry, luxury goods, and travel bookings — strong personal preference, high cost of error, and a need for more mature authorization and confirmation rules.
How will the advertising and retail entry points migrate?
Digital advertising has never been about ad slots; it is about commercial intent. In the agentic era, ads may follow the "default recommendation." This means → traditional search pages, retail storefronts, and feed-based ads all face repricing.
Goldman names three companies with the clearest positioning: Meta combines a personal AI with an AI-native ad interface; Alphabet still owns the discovery gateway and reaches the "intent-to-transaction" infrastructure; Amazon sits closest to the transaction — its retail media already runs on purchase intent.
Large retailers face a two-sided impact. Upside: AI agents naturally optimize price and delivery, favoring scaled platforms. Downside: once users stop opening the retailer's own app, first-party data and retail-media revenue erode. Walmart's AI assistant Sparky, launched June 2025, doubled weekly active customers year-over-year with average order values 40% higher.
How will consumer brands diverge?
In low-engagement, commoditized categories, an AI agent can run the full purchase cycle and auto-swap brands on stock-outs. Goldman places Clorox (CLX) and Kraft Heinz (KHC) on the higher-risk side.
High-engagement categories are more resilient — in premium beauty and skincare, an agent can recommend, but the consumer still makes the final call.
This reflects a deeper shift: brands may no longer compete for "page-one ranking" but for "being understood and recommended by AI." Mondelēz's Oreo appears in only 10% of AI cookie recommendations; Procter & Gamble is pursuing generative-engine optimization (GEO). Put simply = product attributes must be machine-readable, or the AI simply cannot "see" you.
How must payments and fraud logic be rewritten?
Agentic commerce does not bypass payment infrastructure. Google, Stripe, Visa, and Mastercard are plugging "agent-to-merchant" authorization flows into existing card-network rails. Visa's card-not-present tokenization — replacing card numbers with one-time security tokens — rose from 4% in 2019 to 54% in 2025.
Goldman estimates that a 5% uplift in Visa/Mastercard network-security value-added-service growth would accelerate VAS growth by 90 bps and 200 bps respectively, and lift total net-revenue growth by 20 bps and 80 bps.
Traditional fraud signals break down in agentic commerce — requests come from shared cloud infrastructure with no keystroke cadence or mouse movement. This means → fraud logic shifts from "does this look like a real person?" to "is this agent registered, authorized, and tied to a verified consumer?" Equifax, TransUnion, and FICO each sit at different nodes along that chain.
Who guards the gateway of this migration?
Cloudflare sits at the merchant edge, distinguishing legitimate shopping agents from malicious bots and protecting catalogs, inventory, and checkout APIs.
Okta sits at the identity and authorization layer, resolving how agents bind to consumers, which accounts and data an agent may access, and which actions require additional approval.
Goldman's final verdict: the most valuable position may not be the model itself but the position of "being chosen by default." In plain terms = when consumers stop browsing and comparing on their own, whoever decides the first recommendation holds the ultimate competitive leverage.
市场有风险，内容仅供研究参考，不构成投资建议。