Agent Experiences

November 2026 estimate

Revenue at risk

Businesses AI agents cannot read stand to lose about $52 billion a year of US consumer spending by 2030. Nobody had published this number, so we built it, and showed every input.

$52B

a year of US spending lost by 2030, base case

$3.9B–$285B

the range from our low case to our high case

27%

of well-known home pages an automated agent could not read at all

45%

of US consumers have used AI tools for local recommendations, up from 6% a year earlier

  • Spending through agents, 2030
  • Headed to businesses agents cannot read
  • Never won back

From agent spending to lost revenue

Each case starts with a published forecast of US spending through agents in 2030. Part of that spending would have gone to businesses an agent cannot read. Part of that part never comes back, because the shopper buys from whoever the agent could read.

In the base case, $385B goes through agents, $104B of it is headed to businesses agents cannot read, and $52B of that is never won back.

All three cases share one dollar scale. The low case looks small beside the high case because the forecasts it starts from are five times apart.

Change the inputs

The three cases are not the only combinations. Pick low, base, or high for each input. If you think agent spending will be large but most skipped shoppers come back anyway, set it that way.

Spending through agents

Morgan Stanley's bull case

Share agents cannot read

well-known home pages an agent could not read at all (27 of 100) (our State of Agent Readiness study)

Share not won back

our assumption

Lost each year by 2030

$52 billion

US consumer spending, our estimate

InputLowBaseHigh
Spending through agentsHow much US consumer spending goes through agents in 2030?$190BMorgan Stanley's base case · source$385BMorgan Stanley's bull case · source$1Tthe top of McKinsey's range · source
Share agents cannot readWhat share of businesses can an agent not read?8.3%top 10,000 sites blocking at least one AI crawler in robots.txt · source27%well-known home pages an agent could not read at all (27 of 100) · source38%those 27, plus the 11 readable pages that scored Poor · source
Share not won backWhen an agent skips a business, how much of that order never comes back another way?25%our assumption50%our assumption75%our assumption
Lost each year by 2030$3.9B$52B$285B

Which assumption matters most

Move one input from low to high and hold the other two at base. The longer the bar, the more the answer depends on that input.

Spending through agents matters most: it moves the answer from $26B to $135B. Even the narrowest, share not won back, spans $26B to $78B.

The share not won back is the one input nobody has measured. A study that followed shoppers after an agent skipped a business would replace our guess with a number.

  • Input at its low value
  • Input at its high value

What it means for one business

The same formula works for a single shop: your online sales, times the share that goes through agents, times the share of skipped orders you would not win back. It only applies if agents cannot read your business. If they can, this count is zero.

$
10%

Morgan Stanley forecasts 10% to 20% of US e-commerce.

50%

Nobody has measured this. Regulars who know your name come back; new customers mostly do not.

If agents cannot read you

$25,000

a year. If agents can read you, this count is $0, whatever your score.

  • Not readable
  • Readable, but Poor (under 50)
  • Readable, Fair or better

Where shoppers are most likely to be turned away

Our State of Agent Readiness study scanned ten well-known home pages in each of ten categories. Travel had the most that were unreadable or Poor: 8 of 10.

Ecommerce is where agent spending lands first, and 5 of its 10 home pages could not be read at all. Most were stopped by bot protection before the page loaded.

Ten sites per category describes those sites, not an industry. It shows where to look, not how much each industry loses.

Local businesses start further behind

People already ask AI where to buy locally. BrightLocal found 45% of US consumers had used ChatGPT or another AI tool for local recommendations in its 2026 survey, up from 6% the year before.

Few businesses get recommended. SOCi studied about 350,000 locations of 2,751 multi-location brands: ChatGPT recommended 1.2% of them, against 35.9% that appeared in Google’s local 3-pack.

Share of business locations each platform recommends or shows

SOCi, 12 August 2026.

One order

A personal agent noticed a birthday, found a local florist, showed its arrangements, and placed the order by text. The florist’s site scored 64 out of 100, below most of the national flower sites. It got the order because the agent could read it.

Agents that can buy arrived this autumn

Agent spending in 2030 is a forecast. The agents themselves are not. In ten weeks, four companies shipped personal agents that can find a business and pay it.

  1. Meta announces Muse

    A personal agent that also works inside WhatsApp. Meta says it has more than 10 million users. CNBC

  2. Instinct raises $1 billion; Shop Pay checkout inside it

    The texting agent raised at a $10 billion valuation. Shopify announced Shop Pay checkout in Instinct the same day. PYMNTS

  3. OpenAI launches dots

    Always-on agents for Pro and Business Premium users. TechCrunch

  4. Anthropic previews Claude Duet

    A proactive agent in research preview. It asks before any payment or message goes out. TechCrunch

  5. Shop Pay pays stores outside Shopify

    Shopify says Shop Pay issues a one-time card number for agent purchases off Shopify, so the agent never sees your card. Retail Brew

  6. Muse phones US businesses

    For paying subscribers, Muse can call a business when it cannot finish the job online. TechCrunch

  7. ChatGPT Wallet

    OpenAI lets dots complete purchases for US Pro users. CNBC

693%

growth in traffic from generative AI tools to US retail sites, holiday 2025 (Adobe)

90%

higher conversion for AI-referred visits than other traffic, September 2026 (Adobe)

83%

more revenue per visit from AI-referred traffic, September 2026 (Adobe)

Most of this money moves. It does not disappear.

Shoppers still buy the flowers, the dinner, and the running shoes. The money goes to the business the agent could read. For the economy, that is a transfer. For the business that was skipped, it is a sale it never sees and cannot measure: there is no abandoned cart, no bounce, no missed call.

Payment is no longer the hard part. Shopify says Shop Pay can now pay stores outside Shopify on an agent’s behalf, with a one-time card number. What an agent still needs from the business is a site it can read: the florist’s arrangements, the restaurant’s menu as text rather than a PDF, the plumber’s prices and booking.

How this was built, and what it cannot tell you

  • No one has published this figure. We looked in reports from McKinsey, Morgan Stanley, Gartner, Forrester, Bain, Salesforce, and Adobe. They size the spending that flows through agents, not what misses the businesses agents cannot read.
  • The share not won back is our assumption. It is the input with no measurement behind it.
  • The readiness shares are for well-known sites. We have not found a measurement of small business sites. If they do worse, the base case is low.
  • Buying inside an agent has not always worked. Walmart said purchases inside ChatGPT converted at about a third of the rate of shoppers who clicked through to its site (Search Engine Land, March 2026). Some agent spending will arrive as referrals rather than agent checkouts. An agent that cannot read you sends neither.
  • Goods only. McKinsey’s range excludes services. Salons, plumbers, and restaurants’ bookings are not in these numbers.
  • A 2030 figure. Agent spending today is a small fraction of it. The adoption curve is the reason to act early, not this year’s sales.

Sources