B2B Agent Commerce Is Already Bigger Than B2C. What DTC Operators Should Steal From Procurement

Agent Commerce · B2B Mirror

Read the trade press on agent commerce and you would think DTC is where this is being invented. ChatGPT recommending shoes. Claude buying skincare. Every operator demo features a consumer flow.

The DTC examples are the loudest. They are not the largest.

In 2026, the dollar volume flowing through agent-mediated buying is overwhelmingly B2B. Gartner forecasts that 90% of B2B purchases will be mediated by AI agents by 2028, representing over $15 trillion in spending. The B2B procurement stack (SAP Ariba, Coupa, Mercury, Ramp, Brex, Bill.com) has been rebuilt for agent buyers two years ahead of any DTC platform.

That is a strategic gift for DTC operators willing to look. The hard problems in agent commerce (intent matching, structured negotiation, trust verification, payment hand-off, post-purchase memory) all get solved in B2B first, because the dollar stakes force the solution. The B2B Mirror is the framework for copying those solutions into DTC before your category hits the same inflection. This post is five specific primitives DTC operators can steal from B2B procurement, with the source platform named for each, and the 2028 prediction the cluster ends on.

The biggest agent commerce category in 2026 is not DTC. It is the one DTC operators are not looking at.

Why B2B is two years ahead

Three structural reasons put B2B ahead. First, the dollar stakes per transaction force the solution: a single enterprise purchase order can exceed ten thousand DTC orders, so the buyer has every incentive to automate it well, and vendors have every incentive to support that automation. Second, procurement systems were already structured-data heavy long before agents arrived: catalogs, supplier records, and contracts have lived as queryable data for two decades, which is exactly the substrate an agent needs. Third, B2B buying was already non-human in the consideration stage: procurement officers and multi-stakeholder approval workflows behave far more like agents than a browsing consumer does, so the leap to an actual agent was short.

Put those together and the scale is decisive: Gartner projects B2B agent-mediated purchasing will represent over $15 trillion in spending by 2028. The hard problems get solved against that number first. DTC is still learning the basic sequence, the seven stages laid out in the 7-stage agent checkout that DTC is still learning, while procurement has been running an industrialized version of the same flow for years. That gap is the opportunity.

Primitive 1: structured supplier identity (steal from SAP Ariba)

In procurement, every supplier on a network like SAP Ariba carries a verifiable identity record an agent can query before transacting: legal entity, certifications, tax status, performance history. The buying agent does not have to guess whether a supplier is real; the network already verified it. SAP Ariba built structured supplier identity at scale, and it is the cleanest template for the DTC equivalent.

What DTC should steal is the same identity-as-data discipline, applied to brand verification: heavy Schema.org Organization markup, consistent external profiles, the verifiable identity layer that lets a consumer agent treat your brand the way a procurement agent treats a verified supplier. The reason this matters now is the same reason it mattered in procurement, just arriving later in consumer: Gartner's $15 trillion B2B figure is what a fully verified-identity market looks like at maturity, and the consumer market is moving toward the same structure.

Primitive 2: contract-as-data (steal from Coupa)

Coupa's core move is that contracts and terms live as structured data the agent can read directly: pricing, payment terms, SLAs, and renewal conditions are fields, not free-text buried in a PDF. A procurement agent never has to parse prose to know the terms, because the terms are machine-readable by construction. That is why B2B agents can negotiate and transact without a human re-reading every contract.

The DTC translation is direct: shipping terms, return terms, and warranty terms should be structured data, not footer pages. This is the same discipline as the 12-Field Agent SKU and the structured return-policy field: a consumer agent reading a hasMerchantReturnPolicy entity is doing exactly what a procurement agent does reading a Coupa contract field. The consumer market is arriving at this fast: ChatGPT alone reached 900 million weekly active users by February 2026, and every one of those buyers is routed by an agent that prefers terms it can read as data.

Primitives 3 and 4: programmable spend and verified buyer identity (steal from Mercury, Ramp, Brex)

Mercury, Ramp, and Brex rebuilt the corporate card around programmable spend controls, verified buyer identity per transaction, and real-time approval routing. The agent transacting on a Ramp card is operating inside guardrails that pre-authorize what it can spend, on what, and with whom, and the merchant on the other side receives a transaction whose buyer identity is already verified. That combination is why B2B agent payments clear with low fraud and little friction: the trust work happened before the transaction, not during it.

The DTC translation is the buyer-side mirror of the 8-Signal Trust Stack on the brand side: the cleaner and more verifiable the identity and authorization signals on both ends, the faster checkout completes and the lower the fraud loss. Brands that present the merchant-side trust signals an agent can verify get the same low-friction completion that programmable-spend platforms give the buyer side. The consumer urgency is real and accelerating: Adobe reported AI-driven traffic to U.S. retail sites grew 393% year over year in Q1 2026, and that traffic completes best where both ends of the trust handshake are machine-verifiable. It is also why this is where paid amplification compounds, through OpenAI Ads against an already-trusted brand.

Primitive 5: feedback loops at the line-item level (steal from Bill.com and NetSuite)

In mature procurement systems like Bill.com and NetSuite, post-purchase signals attach to the supplier record as data, not anecdote: delivery accuracy, dispute rate, payment latency, and fulfillment exceptions are all structured fields that update the supplier's standing automatically. The next buying decision reads that history as data. A supplier that ships late twice does not rely on someone remembering; the record reflects it, and the agent routes accordingly.

This is the highest-leverage and least-built primitive for DTC, and it is the structured-feedback version of the post-purchase memory stage. The DTC translation is to capture delivery accuracy, return reasons, and resolution outcomes as structured signals that attach to the brand and feed the next recommendation, rather than as support tickets that evaporate. The payoff shows up in conversion quality: Adobe found AI referrals converted 31% better than non-AI traffic during the 2025 holiday season, because a clean, data-backed history is what makes an agent confident enough to recommend again.

The 2028 prediction: when B2C catches up

Here is the dated bet, stated as a prediction rather than a fact. B2B agent procurement crossed roughly a quarter of enterprise purchase-order volume during 2025. Consumer DTC agent commerce, by a modeled estimate, sits at low single digits of GMV in 2026. The B2B-to-DTC lag has run roughly 24 months across the primitives that have already crossed over, so the prediction is that consumer DTC agent commerce crosses the same 25% inflection around Q4 2027. That is a prediction, not a certainty, and the exact quarter will be wrong; the direction and the rough lag are the durable part. What is not a prediction is the destination: Bain projects agentic AI will account for 25% of U.S. ecommerce sales by 2030, so the inflection arrives well inside this decade regardless of the precise quarter.

The operator consequence is the whole point of the mirror. A DTC brand that ships the five primitives in 2026, while its category is still at low single-digit agent penetration, is roughly 18 months ahead of its market at the 2027 inflection. The primitives are not speculative; they are copied from systems that already process the largest agent-commerce volume in the world. All of it feeds the Recommendation Loop the entire stack feeds into: identity, contract-as-data, trust signals, and structured feedback are the inputs that make an agent recommend, remember, and re-recommend a brand.

The contrarian read is the useful one: agent commerce is not being invented in DTC demos; it is being industrialized in procurement, two years ahead, against the largest dollar volume in the market. The five primitives (structured supplier identity, contract-as-data, programmable-spend trust signals, verified buyer identity, and line-item feedback) are already proven at enterprise scale, which means a DTC operator copying them in 2026 is not betting on an unproven idea but importing a solved one ahead of their own category. The brands that do it are the ones positioned to compound on OpenAI Ads when consumer agent commerce hits its inflection, predicted here around 2027 and certain well before 2030. Cresva implements the five primitives on the DTC side: brand identity, contract-as-data, trust signals, and structured feedback. Request early access to build the mirror before your category needs it.