LLMs lift e‑commerce revenue 10–30%

E-commerce teams want one simple answer here — do LLMs actually lift revenue, or is this just another AI slide-deck promise? The honest answer is yes, but not in the neat “LLMs add 10–30% revenue” way social posts imply. The strongest evidence is really about personalization more broadly, with LLMs acting as a newer layer on top. And the numbers split into two buckets: proven gains from better personalization, and much earlier signals from AI shopping channels. ### Where does the 10–30% claim come from? It mostly comes from older personalization research, not from a clean, recent LLM-only benchmark. McKinsey’s long-running work says personalization can lift revenue by 5–15% in most cases, with company-specific lifts reaching 5–25%, and improve marketing efficiency or ROI by 10–30%. That’s real — but it covers recommendation systems, triggered messaging, offers, and customer targeting broadly, not just large language models. (mckinsey.com) ### So what do LLMs add? Basically, LLMs make personalization less rigid. Old systems were good at “people who bought X also bought Y.” LLMs are better at reading messy language — reviews, search queries, support chats, product descriptions — and turning that into intent. That helps with harder retail moments, like matching vague needs to products, generating tailored copy fast, or answering pre-purchase questions without forcing shoppers through filters and menus. McKinsey’s retail work frames the upside as part of a much bigger gen-AI value pool — $240 billion to $390 billion across retail, equal to roughly 1.2 to 1.9 margin points industrywide. (mckinsey.com) ### Is there proof retailers are scaling this? Some proof — but not a universal rollout. In McKinsey’s April 2024 survey of 52 global Fortune 500 retail executives, 90% said they had started experimenting with gen AI, yet only two said they had successfully implemented it across their organizations. That gap matters. It means the technology is widely tested, but the operational rewiring — data quality, privacy, talent, systems integration — is still the bottleneck. (mckinsey.com) ### What about the “ChatGPT drives conversions” angle? That story is even earlier. OpenAI did make shopping much more concrete on September 29, 2025, when it launched Instant Checkout in ChatGPT with Stripe, starting with U.S. Etsy sellers and promising expansion to Shopify merchants. So ChatGPT is no longer just a discovery tool — it is becoming a purchase surface. But independent ecommerce data still says LLM referral traffic is tiny and converts worse than Google organic, email, and affiliates. In one 12-month study across 973 ecommerce sites, ChatGPT traffic was about 0.2% of sessions and lagged those channels on conversion rate and revenue per session. (mckinsey.com) ### Then why are people excited? Because the direction of travel is obvious. BCG says retail personalization leaders are growing revenue 10 percentage points faster than laggards, with $570 billion in incremental growth potentially accruing to those leaders by decade’s end. And generative AI gives retailers a way to push personalization deeper — more products, more queries, more content variants, more customer segments — without scaling headcount linearly. That is the real bull case. (openai.com) ### What’s the catch? Attribution. If a shopper discovers a product through ChatGPT, asks follow-up questions there, then later buys through Google, email, or direct traffic, the LLM may have influenced the sale without getting the credit. But the reverse is also true — a flashy demo can look like revenue lift when it is really just shifting where the same buyer clicks. Until merchants measure incrementality carefully, a lot of “LLM revenue” claims will stay fuzzy. (bcg.com) ### What should merchants believe right now? Believe the boring number first. Better personalization can move revenue, and 5–15% is the most defensible range. Believe the bigger 10–30% figure only when someone is clearly talking about marketing efficiency or a standout case, not average ecommerce sales lift. And treat LLM shopping channels as promising but immature — useful to test now, not solid enough yet to underwrite a plan on their own. (searchengineland.com) The bottom line is simple. LLMs are probably not rewriting ecommerce economics overnight. But they are making the old personalization playbook more conversational, more flexible, and eventually more scalable — which is exactly how a real revenue shift usually starts. (mckinsey.com)