McKinsey expects AI agents to orchestrate $3 trillion to $5 trillion of commerce globally by 2030. Juniper Research forecasts $1.5 trillion of agentic commerce spend in the same year. Nobody here is wrong. McKinsey is counting commerce that agents influence, Juniper is counting money that agents move, and those are different quantities that happen to share a name.
Read the verb before you read the number. Here's what each house counts, and the question none of them answers.
How big is the agentic commerce market?
Published forecasts range from $300 billion to more than $15 trillion, and they don't measure the same thing. McKinsey expects agents to orchestrate $3 trillion to $5 trillion of global commerce by 2030. Juniper Research forecasts $1.5 trillion of agentic commerce spend that year. Bain puts US agent-touched transactions at $300 billion to $500 billion. Gartner's $15 trillion covers B2B spending in 2028.
| House | Figure | Year | Scope | What the number counts |
|---|---|---|---|---|
| McKinsey QuantumBlack | $3 trillion to $5 trillion | 2030 | global, consumer and retail | commerce agents orchestrate, by influencing discovery, decision-making and transactions |
| Juniper Research | $1.5 trillion | 2030 | global | agentic commerce spend, meaning transaction value |
| Bain & Company | $300 billion to $500 billion | 2030 | United States | transactions agents "initiate, influence, or complete", put at 15% to 25% of US online retail sales |
| Gartner | more than $15 trillion | 2028 | global, B2B | spending intermediated by AI agents |
The cleanest comparison there is McKinsey against Bain, because both published a US number for 2030. McKinsey's US sub-figure is up to $1 trillion of orchestrated retail revenue. Bain's is $300 billion to $500 billion. Same country, same year, and a gap of two to three times produced entirely by where each house drew the boundary.
Why do the forecasts differ by that much?
They measure different quantities. McKinsey's estimate covers commerce that agents influence, which includes purchases a person still clicks through themselves. Juniper's covers spend that passes through agent-led transactions. A forecast of influence will always run larger than a forecast of money moved, and the space between them is every purchase where an agent did the shopping and a human did the paying.
McKinsey's own sentence is precise about it: "Our research estimates that by 2030, agentic commerce could orchestrate $3 trillion to $5 trillion globally, as AI agents increasingly influence discovery, decision-making, and transactions across categories." That's from The agentic commerce opportunity, published by McKinsey QuantumBlack in October 2025 and reaffirmed unchanged in March 2026. Orchestrate and influence are the operative words.
Juniper's is the closest public analogue to a payments number. Its Agentic Commerce Market 2026-2031, published 7 April 2026 by analyst Nick Maynard, forecasts agentic commerce spend rising from roughly $8 billion in 2026 to $1.5 trillion in 2030. That's transaction value, and the 2026 base is the part most people skip.
Where does Gartner's $15 trillion fit?
Gartner's figure sits in a different segment and a different year. Daryl Plummer told the Gartner IT Symposium/Xpo in 2025 that AI agents will intermediate more than $15 trillion in B2B spending by 2028. The other three forecasts cover consumer and retail commerce in 2030. The verb is intermediate, and the money is business spending rather than consumer checkout.
Your own agents buying data and compute sit inside Gartner's figure rather than McKinsey's consumer estimate.
What do none of these forecasts tell you?
None of them says who holds the card. Every forecast here describes volume, and each of those transactions is a payment somebody has to authorize, with a credential that belongs to somebody and a limit that somebody set. No report in this comparison takes a position on how that credential gets scoped or on what stops a purchase nobody intended.
That's Shatale's argument. None of the four houses makes it.
An agent carrying a card can buy from any merchant that accepts it, including merchants nobody put on a list. Controls attached to approved endpoints cover the endpoints you predicted. A card scoped to one agent covers everything that agent reaches.
Shatale gives each agent its own scoped virtual card, enforces your policy at the authorization moment, routes purchases above your thresholds to a person, and keeps an immutable per-agent record of each decision. A purchase outside policy is blocked or escalated for approval while the merchant is still waiting, rather than found on next month's statement. More on control at authorization versus reconciliation afterwards.
Which number should you plan against?
Plan against the one whose verb matches your decision. If you're sizing agent-led buying inside your own company, Gartner's B2B figure is the nearer reference. If you're sizing revenue arriving through somebody else's agent, Juniper's spend number fits a payments question and McKinsey's orchestration number fits a marketing one.
Where this goes wrong is on a slide. "$3 trillion to $5 trillion agentic commerce market" placed next to a payments product implies agents will be pushing that much through payment rails by 2030, and no published forecast says that. The agentic commerce map sorts which companies sit on which side of that line, and card versus crypto rails covers why the payments figure splits again by rail.
What to ask
- Which forecast is this vendor's market size drawn from, and what verb does it use?
- Does the number describe payments an agent initiates, or commerce an agent influences?
- Is the figure consumer retail or B2B spending, and does that match how we'd use this?
- Where is our spending policy stored, and can we change a limit without shipping code?
- If an agent attempts a purchase outside policy, is it blocked, escalated to a person, or logged for us to find later?
FAQ
How big is the agentic commerce market?
Published forecasts range from $300 billion to more than $15 trillion because they measure different things. McKinsey expects agents to orchestrate $3 trillion to $5 trillion of commerce globally by 2030. Juniper Research forecasts $1.5 trillion of agentic commerce spend that year. Bain puts US agent-touched transactions at $300 billion to $500 billion, and Gartner's $15 trillion covers B2B spending in 2028.
What does McKinsey's $3 trillion to $5 trillion figure measure?
Commerce that agents orchestrate, globally, by 2030. McKinsey's wording is that agentic commerce could orchestrate that amount "as AI agents increasingly influence discovery, decision-making, and transactions across categories." It counts purchases agents influence, including ones a person completes, so it covers more than agent-initiated payment volume. McKinsey also published a US sub-figure of up to $1 trillion by 2030.
Why is Juniper's forecast lower than McKinsey's?
Juniper measures agentic commerce spend, meaning transaction value moving through agent-led purchases. McKinsey measures commerce agents orchestrate, which includes purchases an agent shaped and a human paid for. Juniper's Agentic Commerce Market 2026-2031, published 7 April 2026, forecasts $1.5 trillion in 2030 against roughly $8 billion in 2026.
Is Gartner's $15 trillion the same market?
No. Gartner's number covers B2B spending intermediated by AI agents by 2028, presented by Daryl Plummer at the Gartner IT Symposium/Xpo in 2025. The McKinsey, Juniper and Bain figures cover consumer and retail commerce in 2030. The segment and the year both differ, and so does the quantity counted, so the numbers can't be added or ranked against each other.
The Agentic Payments Alliance is where the card networks and the stablecoin issuers are arguing the same question at the standards layer. Early access is free for publishers.
Shatale is the control layer for AI-agent payments. Its authorization architecture is the subject of European patent application EP26194994.5 (filed; priority 28 July 2026).