Start for free

TOLSTOY RESEARCH

Online shopping, 2030. How much will AI run?

AI is already changing how brands create content, run ads and help shoppers. By 2030, it could be a better toolkit, a new operating model, or a different way to shop.

Build a scenario for US online retail. Then bring it back to your brand: explore Shop Bots, AI Photoshoots and Ads Engine workflows, with first steps for fashion and beauty.

Build your 2030 Read the method

New claims about AI and shopping arrive all the time. Chatbots will replace search. Agents will do the buying. Stores will run themselves. Some of these claims will come true, and some won’t. We study them at Tolstoy Research because Tolstoy builds AI for online stores: we have a stake in these questions, so we would rather test them in the open than guess.

AI Commerce 2030 is a small, open model of US online retail, with its assumptions laid out on this page. It starts from official sales data 1 and asks five questions. What can AI do in a store? How much will brands use it? How much will it do alone? How much faster will work get? And how often will shopping start with an AI agent? This is version 1.0, and we will keep improving it.

We built three scenarios. In the modest one, AI lifts US online sales 1.5% above where they would be in 2030 if AI stood still. In the substantial one, the lift is 6.3%. In the extreme one, where AI agents do much of the shopping, it is 15%, and the money brands spend on ads to get found starts flowing to the companies that run the agents.

Choose your path. Explore practical ways to put AI to work today, including fashion and beauty. Or follow the research: take one store job apart, explore three scenarios, then answer five questions to build your own 2030.

From the horizon to the everyday

2030 is a scenario.
Your next step can be practical.

You do not need to transform the whole business at once. Start with a useful job for a Shop Bot, connect the right tools and review the work. Here is where Tolstoy fits today.

These are present-day applications, separate from the US retail scenarios in this research. Availability depends on your connected tools and workflow setup.

  1. Creative Director

    Create the next photoshoot.

    Start with product photos and your brand direction. Create on-model, lifestyle and close-up options, then review the product details before saving a repeatable approach.

    Measure it Track time to an approved image, including review and revisions.

    Explore AI Photoshoots

  2. Ads Manager

    Put your Ads Engine to work.

    Turn approved assets into ad concepts, video and copy. Prepare a Meta campaign in Ads Studio, review its audience and budget, then use results to choose the next creative test.

    Measure it Compare a defined creative test against your campaign baseline.

    Explore Ads Studio

  3. Ecommerce Manager

    Connect the everyday store work.

    Bring catalog, customer and order context into the same workflow. Review orders and purchase history, find product-page gaps and prepare supported store updates for approval.

    Measure it Track completed work and review time, alongside corrections.

    See the Shopify connection

  4. Personal Shopper

    Help before and after the purchase.

    Answer product questions, suggest complementary items and help shoppers add to cart. With the required customer details and a compatible store connection, help them check an existing order’s status.

    Measure it Review answer accuracy, resolved questions and support handoffs.

    Meet the Shop Bots

A closer look

Different categories.
Useful first moves.

Fashion

Make the next drop
easier to shop.

Build a reviewed set of on-model views for a small product selection. Use the same direction for styling clips and ad variations, then bring approved video onto the product page.

First pilot
One collection, a small SKU set and a consistent shot list.
Review closely
Seams, construction, color, logos and drape. Try-on is a visualization, not a fit guarantee.
Look for progress
More products with approved assets; less time per accepted set. Measure conversion only in a defined live test.
Explore fashion workflows

Beauty

Show the texture.
Explain the routine.

Start with one product’s hero, texture and application visuals. Create a reusable set of campaign directions and put an approved, shoppable demo beside the product it explains.

First pilot
One hero product, one routine and a few creative angles.
Review closely
Packaging, shade, ingredients and product claims. Generated swatches and try-on do not guarantee an exact shade match.
Look for progress
A faster route to approved creative and clearer product answers. Compare the live placement with your baseline.
Explore beauty workflows

An incremental approach

Start small.
Keep what works.

A proposed pilot sequence for your team. Extra output, recovered time and actual cash savings are different results; measure each separately.

  1. Choose one job.

    Name the owner, the products and the quality standard. Record how the work gets done today.

  2. Run a small pilot.

    Create a small batch with the real inputs. Count approved outputs and the time spent reviewing them.

  3. Measure the change.

    Compare like-for-like work. For a live page or campaign, define the test and success metric before launch.

  4. Save, then expand.

    Keep the brief and review criteria that worked. Add more products or the next connected job.

Every online store is a to-do list

Behind every order, a pile of work

Before a sweater sells online, someone photographs it, describes it, prices it, advertises it and answers questions about it. Each square here is one of those tasks, somewhere in America’s online stores.

Meet Maya, an e-commerce manager

Maya is made up; her job isn’t. She runs online sales for a mid-size US clothing brand. Each square is one task her store needs, from the Labor Department’s O*NET list of what online merchants do 12.

Which tasks AI helps with or runs alone follows published ratings of how much AI can speed up each one 13. The exact split is an illustration, not a model result.

The list keeps changing

Jobs never sit still. Social media handed store teams new tasks, like answering comments. Automated email flows took routine sends off their plate. New tools keep adding tasks and taking others away. AI is the next shake-up.

Some tasks stay human

Some work barely changes. Someone still packs each order and restocks the boxes, mailers and tape. And people still decide which products the brand sells. That work runs on hands, taste and judgment.

Some tasks get a helper

For other tasks, AI does the first pass and Maya makes the call. It drafts a stack of ad variants; she keeps the few that sound like the brand. It predicts which sizes will sell out; she decides what to reorder.

Some leave her desk for good

Other tasks now run start to finish on their own. AI writes product pages, tags the catalog and answers “Does this run small?” before a person ever sees the question. Maya checks the results, not every step.

And new tasks appear

AI also creates work that didn’t exist before. Someone has to check AI-made product shots against the real fabric. Someone has to brief the store’s shopping assistant. And someone has to rewrite product data so AI shopping agents can read it.

Same title, different job

Maya keeps her title, but her week looks different. Less of it goes to making things and more to directing them: setting goals, checking AI’s work, deciding what the brand stands for. And she gets more done in every hour she works.

Now picture every store at once

Now zoom out to every online store in the US. Each task on Maya’s list gets done again and again, in store after store. Each column is one task; each small square stands for many of those repeats. Colored squares: done with AI.

From one store to a whole market

Add up all that work and you get a market. In 2025, US online retail sales reached $1.23 trillion 1. That official figure, from the Census Bureau, is where our model starts.

Where the market goes from here

First, imagine AI stood still at today’s level. Online sales would still grow, to $1.74 trillion a year by 2030, in 2025 dollars. That is our baseline. The real question is how much more AI adds on top.

Maya’s store, task by task

  • Plan the launch calendar — AI helps: AI drafts the calendar from past sales and stock; Maya sets the dates.
  • Write product pages — AI does it alone: AI writes and publishes each page from the spec sheet, in the brand’s voice.
  • Pick products to sell — Stays human: Deciding what the brand sells next is a bet on taste. People make it.
  • Shoot and edit photos — AI helps: AI turns one studio shoot into photos in every color; a person checks each one.
  • Tag the catalog — AI does it alone: AI fills in color, fabric, fit and occasion for every product, so search can find it.
  • Make ad variants — AI helps: AI makes many versions of each ad; Maya keeps the ones that fit the brand.
  • Pack and ship orders — Stays human: Folding, boxing and labeling each order is hands-on work.
  • Answer shopper questions — AI does it alone: A shopping assistant answers “Will this fit me?” from the size chart and reviews.
  • Build email flows — AI helps: AI drafts the welcome and abandoned-cart emails; Maya edits the tone and the offers.
  • Set prices and promos — AI helps: AI shows how a discount would hit margin and stock; Maya decides.
  • Send order updates — AI does it alone: Order, shipping and delivery emails go out on their own, with tracking details.
  • Forecast stock by size — AI helps: AI predicts which sizes sell out first; Maya decides what to reorder.
  • Restock shipping supplies — Stays human: Boxes, mailers and tape still get counted and stacked by hand.
  • Handle complaints — AI helps: AI pulls up the order and drafts a reply; a person decides how to make it right.
  • Update customer records — AI does it alone: Addresses, preferences and order history stay up to date without anyone typing.
  • Choose where to sell — AI helps: AI compares marketplaces, social shops and the brand’s own site; Maya picks where to list.
  • Study site traffic — AI helps: AI flags where shoppers drop off; Maya decides what to fix first.
  • Track sales and costs — AI does it alone: Sales, costs and margins add up on their own, with a plain-English summary every Monday.
  • Design the storefront — AI helps: AI mocks up new page layouts; Maya picks the one that feels like the brand.
  • Scout new suppliers — AI helps: AI shortlists factories and wholesalers; Maya visits and judges the samples.
  • Check AI product shots — New tasks created by AI: Making sure AI-made images show the real fabric, fit and color, and look like the brand.
  • Brief the shopping assistant — New tasks created by AI: Teaching the store’s AI what to say, what never to promise, and when to call a person.
  • Write for shopping agents — New tasks created by AI: Making product data clear enough for AI shopping agents to read and compare.
US online retail sales, in 2025 dollars Faint lines: illustrative paths that vary the timing and size of AI’s effects. “Today” is our 2026 estimate from Census data.

Three versions of 2030

No one knows how fast AI will change online retail, so we built three scenarios instead of one forecast. Each is a different set of answers to the five questions you will see below. The faint lines are illustrative paths that vary the timing and size of AI’s effects.

  • Modest
  • Substantial
  • Extreme
  • If AI stood still

Modest

A better toolkit

AI makes stores faster, not different. Brands adopt AI roughly as fast as shoppers once adopted smartphones: it writes product pages, retouches photos and clears routine tickets. But shopping agents stay a side door: only 12 in 100 online purchases start with one. Online sales end 1.5% above the baseline.

2030 $1.77 trillion +1.5% above the baseline

Substantial

AI can do 70% of store tasks

Brands hand AI 70% of the work it can do, and about half of that runs with no person checking each step. AI shopping agents become a real sales channel: 25 in 100 online purchases start with one. Teams spend more of their time directing AI and less making content. Online sales end 6.3% above the baseline.

2030 $1.85 trillion +6.3% above the baseline

Extreme

Agents do much of the shopping

Shoppers hand everyday buying to AI agents that compare, choose and pay; 50 in 100 online purchases start with one. AI does 52% of store work on its own. Stores become catalogs written for machines, and the money brands spend to get found moves from ad platforms to the agents. Online sales end 15% above the baseline: about as far above trend as they were in 2023, three years after the pandemic jump, by our count 3. Unlike then, the gap lasts, and agents spread about twice as fast as mobile shopping did at its fastest 38.

2030 $2.01 trillion +15% above the baseline

Which 2030 do you expect?

Answer five quick questions about AI and online stores. We will play your answers forward to 2030 and show how your future compares with our three scenarios.

Build your 2030 Skip to the findings

Your 2030 in five questions

Enable JavaScript to build your own scenario. You can read the three scenarios, practical workflows, findings and sources without it.

Capabilities 1/5

By 2030, what share of the tasks it takes to run an online store could AI do?

Think product photos, pages, ads, emails, replies, pricing and stock planning. Count a task if AI could do it in half the time at the same quality, alone or with a person checking, whether or not brands use it.

Each “Today” mark is our estimate of where things stand now. Click it to see how we got it, and the source.

Your answers, played forward to 2030

These are our three scenarios, played forward to 2030. Answer the five questions to add your own future to each chart.

You skipped the questions, so the findings below show our three scenarios only. Answer the five questions to add your own 2030.

Build your 2030

Finding 1: Online sales

AI lifts online sales in all three scenarios

All three scenarios start from $1.23 trillion of online sales in 2025 and about $1.36 trillion in 2026, our estimate. If AI stood still, sales would reach $1.74 trillion by 2030. In our model, AI adds to that through cheaper stores, better stores and shopping agents. Agents count most: in the substantial scenario, they add 4.5% on their own.

Sales end 1.5% above the baseline in the modest scenario, 6.3% in the substantial one and 15% in the extreme one. The tiles show the mix of store work behind each bar.

US online retail sales in 2030, by scenario

Even the modest scenario ends above the baseline. In our model, the last of the five questions moves the result most: how many purchases start with an AI agent.

Finding 2: Where shopping starts

The front door of shopping moves

Today, about 7 in 100 US shoppers start their product search with an AI assistant, two 2026 surveys found 2829. We use that as our estimate for purchases; by the same estimate, most of the rest start on a marketplace or a search engine.

By 2030, agents start 12 in 100 purchases in the modest scenario, 25 in the substantial one and 50 in the extreme one. In the model, every other front door shrinks in proportion.

Where online purchases start, today vs. 2030

Today’s split comes from a 2026 shopper survey 28. “Social media and other” is the share it leaves unassigned, so it is a residual.

A shopper who asks an AI may never see a brand’s homepage, its banners or the ads it paid for. The more purchases start that way, the more a store’s catalog has to speak to machines as well as people.

Finding 3: The work of store teams

Store teams spend more time directing

Our model splits store work using the official list of online-merchant tasks 12. By our rough count, about 39% of team time goes to directing today (deciding, briefing, checking), the rest to producing. As AI takes on production, directing grows in every scenario, to 49% in the substantial one. Only in the extreme one does it overtake producing, reaching 72%.

Output per person rises too. Counting only AI’s effect (2025 = 100), it reaches 103 in the modest scenario, 134 in the substantial one and 243 in the extreme one.

Store teams: time spent and output per person

If the model is right, what stays human is mostly judgment: knowing what good looks like, and catching AI when it gets things wrong.

Finding 4: Who gets paid

Ad money moves to AI agents

In our simplified split of each online dollar, most stays with the brand: the product, shipping, staff and profit. Some pays for the ads that found the shopper; the rest goes in fees for payments, marketplaces and software.

In the model, a purchase that starts with an AI agent skips the ad, though the agent may charge a fee. In the extreme scenario, agents get 4.5% of each dollar and ads drop from 9% to 4.8%. The brand’s share barely moves while agents charge less than marketplaces plus ads.

Where each online dollar goes, today vs. 2030

In our model, the big change for brands is not how much of each dollar they keep. It is whom they pay to be found: less to ad platforms, more to the AI agents that answer shoppers’ questions.

Brands that sell direct usually spend well above this average on ads. In our model ads shrink in proportion, so for a brand spending two or three times the average, two or three times as much of each dollar moves to the agents’ side.

If agents end up taking what marketplaces take today, a 15% fee plus ads, brands would keep about 75¢ of each online dollar in the extreme scenario instead of about 82¢, and agents about 12¢, by our estimate (see the method) 43.

In the extreme scenario, the best shelf space is no longer a search ad or a marketplace listing. It is the answer an AI agent gives.

2030 is still being decided

None of these scenarios is a forecast. Each one shows what follows from a few big assumptions, so you can see which assumptions matter most.

What actually happens depends on choices being made now: what brands hand to AI, how platforms let agents shop and get paid, and which answers shoppers learn to trust.

We will update the model as new data comes in. In the meantime, the choices brands make about content, advertising and shopper assistance are a practical place to start.

How the model works

We start from official data. In 2025, Americans spent $1.23 trillion in online stores, the Census Bureau reports 1: about one retail dollar in six (16.4%). Sales in the first half of 2026 were 11% above a year earlier 2, so “today” on our charts is our 2026 estimate: $1.36 trillion in 2025 dollars.

Then we draw a baseline where AI stands still. From 2022 to 2025, online sales grew 7.3% a year in current dollars 3, while prices of goods sold online rose about 0.8% a year 56. We extend that trend from 2026 to 2030, where it reaches $1.74 trillion in 2025 dollars. On this line, AI stays at its 2026 level: every AI effect on this page is measured from there.

A store is a bundle of tasks. We use the Labor Department’s O*NET list of what online merchants do: 34 tasks, from writing product pages to setting prices 12. In 2023 ratings, a chatbot alone could halve the time of about a third of them 13.

Your five answers set what AI does in 2030. Capabilities times adoption gives the share of store work done with AI. Autonomy splits it into work AI does alone and work where it helps a person. Productivity sets how fast that assisted work goes. AI also creates new work for people: briefing, checking and directing it.

In the model, AI lifts online sales in three ways.

Cheaper stores. When store work costs less, part of the savings reaches shoppers as lower prices. Physical stores gain less from AI 13, so online prices fall a little compared with theirs. Shoppers respond: when Amazon began charging sales tax, households cut purchases there about 1.3% per 1% price rise 34.

Better stores. Better pages, photos, search and answers turn more visitors into buyers. On a large cross-border online retail platform, AI product descriptions raised sales 2% in field experiments, and a pre-sale chatbot 16% for shoppers who used it, versus no chat 14. A store using AI for all its work might sell about 5% more 14, but much of that comes from other stores, so we count only the part new to online: one sale in four, our assumption.

Shopping agents. We estimate about 7 in 100 online purchases start with an AI assistant today, based on two 2026 shopper surveys 2829. We assume most purchases that move to agents would have happened online anyway; one in four is new.

Change builds up over time. Each answer moves from today’s level to your 2030 answer along an S-curve: slow at first, fastest around 2028.

The findings use the same numbers. Agents take the share of purchases you chose; other front doors shrink in proportion. Store teams: we count the hours left for producing and directing. Each dollar: card fees and marketplace commissions stay 4443, ads shrink with the purchases that skip them, and agents’ fees rise from 4% today toward a marketplace-style 15% 4743.

Three scenarios, anchored on the past. In the modest one, brands adopt AI roughly as fast as shoppers adopted smartphones (an analogy), but agents stay a side door 3723. In the extreme one, online sales end about as far above trend as in 2023, three years after the pandemic jump (by our count), but for good 3. The substantial one sits in between.

What moves the result most. In our tests: the share of purchases that start with an agent, and how many AI-driven sales are new to online shopping rather than moved between stores. No study measures that yet, so it is our main assumption.

Everything is published. Each value below says where it comes from, and a separate script checks every result.

What we assumed

Starting values25 values, each with its status and source
Starting values
Starting valuesValueYearStatusSources
US online retail sales, 2025Total US retail e-commerce sales, 2025, not seasonally adjusted (Census, Quarterly Retail E-Commerce Sales, 4th Quarter 2025, released 10 Mar 2026).$1,233.7bn2025Established1
Online share of US retail sales, 2025E-commerce sales accounted for 16.4% of total 2025 retail sales.16.4% of total retail sales2025Established1
US retail sales, 2025Derived: 1,233.7 / 0.164 (rounding of the 16.4% share gives about +/-0.3%).$7,522.6bn2025Established1
Growth of US online sales, 2025E-commerce sales +5.4% (+/-1.2%) in 2025 vs 2024.5.4% per year, nominal2025Established1
Growth of US online sales, 2022–2025Computed from Census quarterly e-commerce (not seasonally adjusted): $997.5bn in 2022 -> $1,233.7bn in 2025.7.34% per year, nominal2025Established3
Consumer prices, 2025CPI-U all items, +2.7%.2.7% Dec 2024 → Dec 20252025Established4
Consumer prices, 2022–2025CPI-U all items +3.4% (2023), +2.9% (2024), +2.7% (2025): 3.0% a year. Alternative deflator, not used: it is mostly services, which are not sold online.3% per year, Dec 2022 → Dec 20252025Established4
Growth of US online sales, first half of 2026Census e-commerce, not seasonally adjusted: Q1 2026 $301.2bn (revised) + Q2 2026 $329.5bn (preliminary) = $630.8bn vs $568.9bn in H1 2025 (+10.9%). Used as the 2026 nowcast.10.88% H1 2026 vs H1 2025, nominal2026Established2
Growth of US retail sales, first half of 2026Total retail, not seasonally adjusted: +4.0% y/y in Q1 2026 and +6.6% in Q2 2026; average 5.3%. Share sanity check only.5.3% H1 2026 vs H1 2025, nominal2026Estimate2
US online sales, last four quartersSum of the last four published quarters (not seasonally adjusted): cross-check for the 2026 nowcast.$1,295.6bn nominal, Q3 2025-Q2 20262026Established3
Online share of US retail sales, spring 2026Latest quarter at the time of writing: Q2 2026, $340.2bn s.a., +12.2% year on year.17.1% of total retail sales, s.a.2026Established2
Jump in the online share, spring 2020E-commerce share of retail: 11.9% (Q1 2020) -> 16.3% (Q2 2020), s.a. Upper-bound precedent for a channel shock.4.4 pts in one quarter2020Established3
Online share of British retail (context only)Great Britain, Dec 2025, retail excluding vehicles. Context only: US shares here use Census total retail, which includes car dealers and fuel, so they are not comparable.28.3% of retail sales2025Established10
Phones’ share of online orders, holidays 2025Smartphones, Nov-Dec 2025 (Adobe). Precedent for channel diffusion: 22% of Black Friday sales in 2015.56.4% of online transactions2025Estimate23
Growth of AI-referred visits, holidays 2025Traffic to US retail sites from generative-AI tools +693.4% year on year, holiday 2025, from a small base.+693.4% year on year2025Estimate23
Extra conversion of AI-referred visitsAI-referred retail visits converted 54% better than other visits in May 2026 (42% in Mar 2026, ~50% worse a year earlier).+54% vs non-AI traffic2026Estimate24
Online merchants’ tasks AI can speed upO*NET 13-1199.06 Online Merchants, GPT-4 ratings, core tasks weighted 2x (reproduced from the published task file).62.93% of tasks (E1 + 0.5 E2)2023Established13
Store salespeople’s tasks AI can speed upO*NET 41-2031.00 Retail Salespersons, same measure: store work is about half as exposed as online store work.34.21% of tasks (E1 + 0.5 E2)2023Established13
Amazon ad revenue, 2025Advertising services, FY2025.$68.6bn worldwide2025Established39
Amazon gross merchandise sales, 2025Marketplace Pulse estimate; US ~ $440bn (35.7% of US e-commerce); third-party sellers 69% of GMV.$830bn worldwide2025Estimate40
Wayfair ad spendAdvertising $1,425m / net revenue $12,457m.11.4% of net revenue2025Established45
Wayfair operating costsSOTG&A 14.3% + customer service and merchant fees 3.8%.18.1% of net revenue2025Established45
Amazon referral feeMost common referral fee; most categories 8%-15%.15% of item price2026Established43
Card processing feeStripe standard US online card rate.2.9% of order + $0.302026Established44
ChatGPT checkout feeOpenAI fee on ChatGPT Instant Checkout sales by Shopify merchants (Jan 2026); Gemini and Copilot charged nothing then.4% of order2026Established47
Model parameters25 values, each with its status and source
Model parameters
Model parametersValueYearStatusSources
Online sales trend, before inflationCensus e-commerce CAGR 2022-2025 ($997.5bn -> $1,233.7bn). Applied from 2027; 2026 is the nowcast from the first half of 2026. Not extrapolated: the 2026 acceleration.7.34% per year, nominal (range 5.4% to 9%)Established3
Price rise at online retailersSame window, 2022-25: implicit price deflator of online retailers' sales (NAICS 4541), which BLS deflates product line by product line with matching CPIs: +2.0% (2023), -0.2% (2024), +0.7% (2025). Upper bound: CPI-U all items, 3.0% a year (mostly services not sold online). The deflator only converts the same nominal path into 2025 dollars; lifts in % do not depend on it.0.84% per year (range 0% to 3%)Estimate5
US retail trend, before inflationTotal retail +3.5% in 2025, applied from 2027 (2026: first-half growth). Share sanity check only.3.5% per year, nominal (range 2.5% to 5%)Estimate1
Cost of the store work AI can touchSpend on the store work AI can touch (content, merchandising, marketing ops, service, analytics, site). Wayfair spends 14.3% on SOTG&A and 3.8% on service + merchant fees; we exclude tech infrastructure and card fees.12% of online sales (range 8% to 18%)Estimate45
Savings passed on to pricesOnline retail is competitive; most but not all cost savings reach shoppers within five years.70% of cost savings passed to prices (range 50% to 100%)Assumption—
Shoppers’ response to online pricesEstimate, not a measured channel elasticity. Baugh, Ben-David & Park (J. Finance 2018) find 1.2-1.4 for Amazon purchases against ALL competitors, online ones included, which overstates online-vs-store substitution. Einav et al. (AER 2014) find residents buy ~2% more online per 1-point rise in local sales tax, which points higher. We keep 1.3 between the two.1.3 online sales response to online-vs-store price (range 1 to 2)Estimate34
Savings matched by physical storesStores use AI too. We assume they capture savings in proportion to exposure: Retail Salespersons 0.342 vs Online Merchants 0.629 (= 0.54).54.36% of the online cost advantage matched by stores (range 30% to 80%)Estimate13
Sales lift from better stores (one store)Field experiments on a large online retail platform: AI product descriptions +2.05% sales, AI search-query refinement +2.93%, pre-sale chatbot +16.3% on exposed sessions, ads no detectable effect (Fang et al. 2025). A store using AI across all its work: ~5%, measured against its own control group.5% one store's sales lift if all its work were AI-done at x2 (range 2% to 10%)Estimate14
Better-store sales that are new to onlineSame logic as for agent-started purchases. A store's gain over its own control includes sales taken from other online stores; when every store adopts AI, that part cancels out (composition fallacy). We assume 1 in 4 extra purchases is new to online (from stores or not made at all).25% of a store's content gain that is new to the online channel (range 10% to 50%)Assumption—
Agent-started sales that are new to onlineMost agent-started purchases would have happened online anyway (via search or a marketplace); about 1 in 4 is new to the online channel. Bound: COVID put e-commerce 28% above its 2015-19 trend in 2021, fading to 0% by 2025 (Census, our calculation).25% of agent-started purchases that would not have been online (range 10% to 50%)Assumption—
Cost left when AI runs a taskCompute, software and exception handling for tasks AI runs alone (oversight time is counted in new tasks).10% cost of an automated task vs a human one (range 5% to 30%)Assumption—
New human work per AI-done taskDirecting, checking and feeding AI (briefs, QA, catalog data for agents). Acemoglu & Restrepo (JEP 2019): new tasks offset part of automation's displacement.0.2 new human task instances per AI-done instance (range 0.1 to 0.4)Assumption—
Producing share of store tasks, 2025Online Merchants: 20 of 34 task statements are production/routine (36 of 58 weighted points, core x2); 14 are direction/decision.62.07% of store task statements (range 50% to 75%)Estimate12
Share of AI’s work that is producingMass x exposure: production tasks score 0.681 (E1 + 0.5 E2), direction tasks 0.545.67.12% of AI-done work that is production (range 60% to 90%)Estimate13
Midpoint of the AI S-curveAI effects start from today (2026) and half of the 2026->2030 change is reached in 2028 (symmetric S-curve). Affects the path, not 2030.3 years after 2025 (range 2 years to 4 years)Assumption—
Steepness of the AI S-curveS-curve steepness: 18% of the change by 2027, 50% by 2028, 82% by 2029.1.2 per year logistic slope (range 0.8 per year to 2 per year)Assumption—
PaymentsCard processing: 2.9% + $0.30 list price; large merchants pay less.3% of each online dollar (range 2% to 3.5%)Estimate44
Online sales made on marketplacesAmazon US GMV ~$440bn x 69% third-party = ~$304bn (24.6% of US e-commerce), plus eBay, Etsy, Walmart Marketplace, Temu, TikTok Shop.30% of US e-commerce sold by third parties on marketplaces (range 25% to 40%)Estimate40
Marketplace commissionAmazon referral fees: 8%-15% in most categories, 15% the most common.15% referral fee (range 8% to 15%)Established43
Store softwareStore platforms, apps and plugins.1% of each online dollar (range 0.5% to 2%)Assumption—
Paid ads, 2025Amazon ad revenue $68.6bn / Amazon GMV ~$830bn = 8.3%; Wayfair 11.4% of revenue; Amazon sellers up to 15% (Marketplace Pulse).9% of each online dollar (range 6% to 13%)Estimate39
Agent fee todayOpenAI charges Shopify merchants 4% on ChatGPT Instant Checkout sales (Jan 2026).4% fee on monetised agent purchases (range 0% to 5%)Established47
Agent fee at full scaleAs agents control more demand their take rises toward a marketplace referral fee (15%).15% fee if all purchases started with agents (range 8% to 20%)Assumption43
Agent-started sales that pay the agent todayMost AI-started journeys click out to Amazon or a brand site for free (PYMNTS: 59.7% of AI-assisted purchases completed on Amazon).10% of agent-started purchases that pay the agent (range 5% to 25%)Assumption—
Agent share at which every agent sale paysCheckout, affiliate fees or paid placement on all agent purchases once agents start half of purchases.50% agent share at which every agent purchase pays (range 30% to 80%)Assumption—
Scenario inputsToday and the three scenarios, question by question
Scenario inputs
Scenario inputsToday (est.)ModestSubstantialExtreme
CapabilitiesIn a 2023 study, raters judged that a chatbot alone, used by a person, could do about a third of an online merchant’s tasks in half the time at the same quality 13. Counting tasks that need AI tools built on top at half weight gives about 63%, so our estimate of 35% is a floor.35 (30–60) Estimate457090
AdoptionIn late 2024, AI helped with only a few percent of all US work hours 17, and in spring 2026 only about one retail firm in seven used it 20. Store work suits AI better than most jobs 13, so our estimate is that brands already hand AI about 20% of the work it can do.20 (10–35) Estimate357090
AutonomyAI is starting to act alone: about half of Claude.ai conversations in August 2025 handed it a whole task 21, and Klarna says its assistant handled two-thirds of its service chats in its first month 22. But in most stores a person still approves AI’s copy, images and replies before they go out, so our estimate for today is “some”.Some (Some–About half) EstimateSomeAbout halfMost
ProductivityIn studies of AI at work, support agents resolved 14% more issues per hour 15, workers got 33% more done in each hour they used AI 17, and professionals took 40% less time on writing tasks 16. The middle of these results is closest to ×1.5, our estimate for today.×1.5 (×1–×1.5) Estimate×1.5×2×4
AI shopping agentsTwo 2026 surveys found that about 7 in 100 US shoppers start their product search with an AI assistant 2829. Surveys count people, not purchases, so this is a rough estimate: anywhere from 2 to 12 in 100 is plausible.7 (2–12) Estimate122550

What this model leaves out

  • It is simple on purpose. A handful of levers stand in for a market shaped by prices, competition, supply chains and rules.
  • It covers US retail e-commerce as the Census Bureau measures it 1. Online services such as travel bookings are not included.
  • “Today” is a nowcast from the first half of 2026: it assumes the second half grows as fast, and part of that growth may already come from AI. We will update it with the Census third-quarter release on November 19, 2026 2.
  • We assume prices at online retailers keep rising as slowly as in 2022–25 (the government index covers online-only sellers) 5. Faster inflation, from tariffs for example, would shrink the figures in 2025 dollars but not the lifts.
  • Our main assumptions are unmeasured: how many of the sales that agents and better stores bring are truly new to online shopping, rather than moved between stores. We assume one in four for both. No study measures this yet, and it moves the results more than anything else: with one in ten instead, the substantial scenario’s lift of 6.3% roughly halves; with one in two, it nearly doubles.
  • When online prices fall, how many shoppers move from stores to online is our estimate, based on two studies of related effects 3435.
  • Where purchases start today comes from small online surveys of shoppers, not records of purchases 2829. “Social media and other” is the share the main survey leaves unassigned. In 2030, every other starting point loses share to agents in proportion.
  • Who gets each dollar is a simplified split of the average online sale. Platform fees are held constant, and shipping and AI tool costs sit inside the brand’s share. Ad and store-work costs are anchored on one large online retailer and on Amazon 4539; small brands spend more on ads. If every purchase that starts with an agent paid it what a marketplace takes today, a 15% fee plus about 9% in ads, agents would take about 12¢ of each dollar in the extreme scenario and brands would keep about 75¢ (Finding 4) 43.
  • The AI ratings of store tasks date from 2023 13, so today’s capabilities mark is a floor.
  • It does not model jobs or pay. Changes in how teams spend their time are not forecasts of hiring or layoffs.
  • It looks at the market as a whole. It does not say which brands win or lose.
  • It leaves out recessions, tariffs, the prices of goods, returns and regulation. The faint lines vary only the timing and size of AI’s effects, not shocks like these.
  • Every “Today” mark is an estimate built on early, fast-moving data.

Sources

  1. Quarterly Retail E-Commerce Sales, 4th Quarter 2025 — U.S. Census Bureau (2026). Our starting point: US online retail sales were $1.23 trillion in 2025, up 5.4% on 2024 and 16.4% of all retail sales.
  2. Quarterly Retail E-Commerce Sales, 2nd Quarter 2026 (CB26-133) — U.S. Census Bureau (2026). Our “today”: online sales in the first half of 2026 ran 10.9% above a year earlier, the basis of our 2026 estimate, which the November release will update.
  3. E-Commerce Retail Sales (ECOMNSA) and E-Commerce Retail Sales as a Percent of Total Sales (ECOMPCTSA) — U.S. Census Bureau, via FRED (Federal Reserve Bank of St. Louis) (2026). Our baseline trend: online sales grew about 7.3% a year from 2022 to 2025 in current dollars; the pandemic’s gap above trend is our extreme-case precedent.
  4. Consumer Price Index: 2025 in review (The Economics Daily) — U.S. Bureau of Labor Statistics (2026). General inflation was 2.7% in 2025; over 2022–25 it averaged 3% a year, our upper bound for price growth, not the index we use.
  5. Sectoral Output and Real Sectoral Output for Retail Trade: Electronic Shopping and Mail-Order Houses (NAICS 4541) — U.S. Bureau of Labor Statistics (Industry Productivity), via FRED (2026). The price index we use for 2025 dollars: prices of goods sold by online retailers rose about 0.84% a year in 2022–25, by our calculation from BLS data.
  6. Sales versus margins: alternative measures of output and productivity for retail trade — Kim, Miller, Myers, Price, Robic & Rudd, Monthly Labor Review (U.S. Bureau of Labor Statistics), March 2025 (2025). Why that index fits: BLS turns online retailers’ sales into volumes by deflating each product line with a matching consumer price index.
  7. Consumer Price Index: 2024 in review (The Economics Daily) — U.S. Bureau of Labor Statistics (2025). General inflation was 2.9% in 2024; with 2023 and 2025, it sets the upper bound for price growth in our baseline.
  8. Consumer Price Index: 2023 in review (The Economics Daily) — U.S. Bureau of Labor Statistics (2024). General inflation was 3.4% in 2023; with 2024 and 2025, it sets the upper bound for price growth in our baseline.
  9. Productivity and Costs by Industry: Wholesale Trade and Retail Trade (USDL-26-0772) — U.S. Bureau of Labor Statistics (2026). Output per hour at online and mail-order retailers grew 10.3% in 2025 and 10.6% a year on average since 1987; our team chart adds only AI’s part.
  10. Retail sales, Great Britain: December 2025 — Office for National Statistics (UK) (2026). For context, 28.3% of British retail sales excluding vehicles were made online in December 2025, but our US shares include car dealers and fuel, so they don’t compare.
  11. US Retail In 2030: E-Commerce Expands, Stores Still Matter — Forrester (2026). A cross-check: Forrester estimates US online sales of $1.8 trillion in 2030, in current dollars, in line with our baseline before converting to 2025 dollars.
  12. O*NET OnLine summary report: 13-1199.06 Online Merchants — O*NET Resource Center (U.S. Department of Labor) (2026). The official US list of what online merchants do, 34 tasks from writing product pages to setting prices; Maya’s store and our task split build on it.
  13. GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (paper and task-level data) — Eloundou, Manning, Mishkin & Rock (OpenAI / University of Pennsylvania), arXiv:2303.10130 (2023). In 2023 ratings, a chatbot alone could halve the time of about a third of an online merchant’s tasks: our floor for today’s capabilities.
  14. Generative AI and Sales Productivity: Field Experiments in Online Retail — Fang, Yuan, Zhang, Donati & Sarvary, arXiv:2510.12049 (CESifo WP 12201) (2025). Field tests on a large cross-border online retail platform: AI product descriptions lifted sales 2%, a shopping chatbot 16% for its users; the basis of our better-stores effect.
  15. Generative AI at Work — Brynjolfsson, Li & Raymond, Quarterly Journal of Economics 140(2): 889-942 (NBER w31161) (2025). An AI assistant helped customer-support agents resolve 14% more issues per hour, one of the studies behind our estimate of today’s productivity gain.
  16. Experimental evidence on the productivity effects of generative artificial intelligence — Noy & Zhang, Science (July 2023) (2023). Professionals given ChatGPT finished writing tasks in 40% less time, with better quality: the high end of our estimate of today’s productivity gain.
  17. The Impact of Generative AI on Work Productivity (On the Economy blog) — Federal Reserve Bank of St. Louis (2025). Feeds two Today marks: AI assisted 1.3% to 5.4% of US work hours in late 2024, and workers did 33% more in each hour using it.
  18. The Rapid Adoption of Generative AI — Bick, Blandin & Deming (Federal Reserve Bank of St. Louis research; NBER w32966) (2024). Generative AI spread faster than the PC or the internet: 39% of US adults under 65 had used it by August 2024, a pace our faster scenarios lean on.
  19. Monitoring AI Adoption in the U.S. Economy (FEDS Notes) — Board of Governors of the Federal Reserve System (2026). Work use of generative AI kept growing: 40.7% of US workers used it by November 2025, up 31% in a year, which raises our estimate of today’s adoption.
  20. Large Firms With at Least 20 Employees Biggest AI Users — U.S. Census Bureau (Business Trends and Outlook Survey) (2026). A brake on our adoption estimate: only about 14% of US retail firms used AI in any part of their business in spring 2026.
  21. Anthropic Economic Index report: Uneven geographic and enterprise AI adoption — Anthropic (arXiv:2511.15080) (2025). How often people hand AI a whole task: 49% of Claude.ai chats and 77% of business use in August 2025; stores sit lower, as people approve most output.
  22. Klarna AI assistant handles two-thirds of customer service chats in its first month — Klarna (press release) (2024). Klarna says its AI assistant handled two-thirds of its service chats in its first month: a company estimate, and our example of AI working alone.
  23. Adobe: Holiday Shopping Season Drove a Record $257.8 Billion Online with Consumers Embracing Generative AI Tools — Adobe (Adobe Analytics) (2026). Adobe estimates AI traffic to US retail sites grew 693% over the 2025 holidays, from a small base, and that phones made 56.4% of online purchases.
  24. Q3 2026 AI-sourced traffic insights (Adobe Digital Insights) — Adobe (2026). Adobe estimates that visits from AI tools converted 54% better than other retail visits in May 2026, likely because those shoppers arrive closer to buying.
  25. Adobe AI referral traffic data, July 2026 — Digital Commerce 360 (2026). By Adobe’s estimate, AI referrals to US retail sites grew 62% in the year to July 2026, slower than before as the base grows.
  26. Lessons learned from Adobe's 2026 Q2 AI traffic report — Search Engine Journal (2026). A second report of Adobe’s estimates: AI-referred visits to US retailers grew 393% in the year to the first quarter of 2026.
  27. As consumers pull back, AI reshapes the entire shopping journey — eMarketer (citing Similarweb) (2025). Our floor for today’s agent share: Similarweb estimates ChatGPT sent about 1% of Walmart.com’s visits in August 2025, though such data miss many AI journeys.
  28. 59% of shoppers plan to use AI this Black Friday, but just 7% will start their search there — Omni Calculator (2026). Our estimate for today: 7% of US adults said they would start their product search with an AI chatbot; its split of other starting points shapes Finding 2.
  29. Google still leads shopping searches as AI falls short at 7% — PPC Land (reporting a Smarty Marketing survey) (2026). A second survey estimate that matches the first: 7.3% of US consumers named ChatGPT as their first choice to start a product search, in May 2026.
  30. From Prompt to Purchase: Who Wins the AI Shopper? — PYMNTS Intelligence (2026). Survey estimate: 23% of US consumers say AI is their main starting point for shopping research (our upper bound), and 59.7% finished their last AI-assisted purchase on Amazon.
  31. The State of Ecommerce 2025 — Constructor and Shopify (2025). Our check on social media as a starting point: 25% of US, UK and German shoppers start searches on TikTok, a survey estimate with several answers allowed.
  32. Here Come the Shopping Bots — Morgan Stanley (2025). Morgan Stanley estimates AI agents will handle 10% to 20% of US online spending by 2030; our modest and substantial scenarios sit near that range, on a slightly broader measure.
  33. The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants — McKinsey & Company (QuantumBlack) (2025). McKinsey estimates AI agents could orchestrate $900 billion to $1 trillion of US retail sales by 2030, an upper reference for our extreme scenario.
  34. Can Taxes Shape an Industry? Evidence from the Implementation of the "Amazon Tax" — Baugh, Ben-David & Park, Journal of Finance 73(4): 1819-1855 (NBER w20052) (2018). When Amazon began charging sales tax, households cut Amazon purchases about 1.3% per 1% price rise; our estimate of how shoppers switch between online and stores builds on it.
  35. Sales Taxes and Internet Commerce — Einav, Knoepfle, Levin & Sundaresan, American Economic Review 104(1): 1-26 (NBER w18018) (2014). eBay data show shoppers buy almost 2% more online for each 1-point rise in their state’s sales tax; it sets the high end of our estimate of price response.
  36. Growth, Adoption, and Use of Mobile E-Commerce — Einav, Levin, Popov & Sundaresan, American Economic Review 104(5): 489-494 (2014). When eBay shoppers adopted the mobile app, they bought more in total instead of just moving purchases: evidence that a new way to shop can add sales.
  37. Record 22% of US Black Friday sales via smartphone - Adobe — Telecoms.com (reporting Adobe Digital Index) (2015). The pace of the last big channel shift: Adobe estimates smartphones carried 22% of US Black Friday online sales in 2015, up 70% on 2014.
  38. Mobile accounts for nearly 35% of 2017 e-commerce sales — Digital Commerce 360 (citing eMarketer) (2018). The same shift two years on: an estimated 34.5% of US online sales were made on mobile devices in 2017.
  39. Amazon.com Announces Fourth Quarter Results (FY2025) — Amazon.com, Inc. (Business Wire) (2026). Amazon earned $68.6 billion from advertising in 2025, about 8% of the goods sold on its platform by our calculation: our anchor for ads’ share of each dollar.
  40. Amazon GMV Surpassed $800 Billion in 2025 — Marketplace Pulse (2026). Marketplace Pulse estimates $830 billion of goods were sold on Amazon in 2025, 69% by outside sellers, which sizes marketplaces’ cut of US online sales.
  41. Amazon and Shopify Are Now Half of U.S. E-Commerce — Marketplace Pulse (2026). Context on concentration: Marketplace Pulse estimates Amazon made 35.7% of US online sales in 2025, and Amazon plus Shopify stores about half.
  42. Amazon Takes a 50% Cut of Sellers' Revenue — Marketplace Pulse (2023). Marketplace Pulse estimates Amazon takes over half of a typical seller’s revenue in fees and ads; we leave out its warehouse and shipping fees, which buy a service.
  43. Amazon selling fees and pricing — Amazon (sell.amazon.com) (2026). Amazon’s most common referral fee is 15% of the item price; we use it for marketplace fees and as the most agents may one day charge.
  44. Stripe pricing — Stripe (2026). Card processing costs about 2.9% plus 30 cents per online payment, so we count about 3% of each online dollar as payment fees.
  45. Wayfair Announces Fourth Quarter and Full Year 2025 Results (Exhibit 99.1) — Wayfair Inc. (2026). One large online retailer’s 2025 accounts: advertising took 11.4% of Wayfair’s revenue, and they anchor our estimates of store-work and ad costs.
  46. Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol — OpenAI (2025). Since September 2025, US ChatGPT users can buy without leaving the chat, and merchants pay a small fee: the start of agents getting paid.
  47. Shopify merchants have to pay an additional 4% fee if customers buy via ChatGPT — Retail Brew (2026). Shopify merchants pay OpenAI a 4% fee on ChatGPT checkout sales, while Gemini and Copilot charge nothing for now: our starting fee for agents.
  48. The Simple Macroeconomics of AI — Acemoglu, Economic Policy 40(121): 13-58 (NBER w32487) (2025). Our cheaper-stores effect uses this method: the gain equals the share of tasks AI touches times the cost it saves on each one.
  49. Automation and New Tasks: How Technology Displaces and Reinstates Labor — Acemoglu & Restrepo, Journal of Economic Perspectives 33(2): 3-30 (2019). Our task framework: AI can take tasks away from people, but it also creates new tasks in which people have the edge.
  50. Economic scenarios explorer (Anthropic Institute) — Anthropic (2026). The explorer that inspired this page, for the whole US economy: its modest GDP lift (1.6%) is close to ours, and its larger ones are higher than ours.

Version 1.0, November 2026

This is the first public version of AI Commerce 2030. It simplifies a complex market on purpose, so anyone can see how the pieces fit and argue with them.

The scenarios are illustrations built on stated assumptions. They show how the model responds to your choices, not what will happen.

Each update to the model will get a new version number.

Research, writing, design and engineering by Tolstoy Research.

The format of this explorer is inspired by Anthropic’s economic scenarios explorer.

AI Commerce 2030 · model v1.0