Every prior platform shift has minted its own payments layer. Paypal was a product of the early web and Stripe grew out of the mobile and cloud era, which resulted in a proliferation of new companies. Incumbents, investors, and founders recognize this opportunity for the AI era, and they are aggressively jockeying to become the dominant force for a future of agentic payments.
So far, the center of gravity in agentic payments has been on the consumer side. Visa and Mastercard have launched agent programs, Stripe has a plethora of products, and Google has created a commerce protocol of its own. There is no shortage of posturing, partnering, and maneuvering happening in the halls of power within payments and commerce, and given the size of the consumer payments market (more than $20 trillion of US consumer spending a year) and the implications for incumbents if they lose, it's understandable that this is the battleground du jour.
But there may be an even bigger opportunity in agentic payments in business payments.
The market is massive, fragmented, and composed of buyers that are happy to pay for better services. Over $100 trillion a year in business payments is moved globally, and the players who own it today are slow to innovate in the best of times. There is no love for business payments systems among its users (but try prying a rewards credit card out of a consumer's hands). Businesses, unlike consumers, will actually pay for efficiency, and the earliest real usage of payment-capable agents is already happening inside companies. Finally (and perhaps most important), the money movement in business payments has never been the product. The coding, approving, reconciling, and controlling around each payment is a labor pool worth hundreds of billions of dollars a year.
Payments volume demonstrates just how large this market is. In 2025 the ACH network moved 8.08 billion business-to-business payments worth $63.11 trillion in the United States. Fedwire alone clears more than $1.1 quadrillion in wires a year, of which about one-fifth is for commercial transactions. Together, these rails carry the overwhelming majority of business money movement in the US.
The real economic opportunity sits above the core transactional rails. Software-enabled invoicing is a $24 billion market growing ~20 percent a year. Accounts receivable and payments software are each ~$5 billion markets growing in the low teens. And this speaks nothing of the legacy, human-centric accounting industry, roughly $150 billion a year in the US alone and more than $650 billion globally.
At the same time, the legacy technology layer in business has been fragmented, stale, and underperforming. Bill.com peaked near $35 billion in enterprise value in late 2021 and trades around $4.5 billion today. AvidXchange, which spent 25 years building AP automation for the middle market, went public in 2021 at $25 a share and was taken private last October at $10. Fragmentation runs rampant. The largest pure-play in the space, Bill, processes roughly $370 billion a year, well under 1 percent of US B2B payment flows, and no business payments company on the NASDAQ holds anything approaching dominant share of even its own layer of the stack, with AP automation alone split across Bill, AvidXchange, Tipalti, Coupa, Stampli, and a number of embedded modules.
Most tellingly, technology still has not fully permeated this market. Modern Treasury finds 51 percent of companies still perform up to half of their payment operations manually. None of the incumbents has shipped anything close to a payment-capable agent. This lack of innovation will prove the death knell for these businesses with the AI platform shift underway. Enterprises are deploying agents across their operations, in engineering, support, procurement, and finance, and they will need to consume a new payments interface. Agents will get their new payments platform, and because the earliest real agent deployments are happening inside businesses rather than in consumers' pockets, that layer will be built for business payments first.
The consumer internet taught us one lesson about efficiency: consumers do not pay for it. They will trade data for it, they will trade attention for it, and they will accept lock-in for it, but they will not pay. Businesses are the opposite, because inefficiency in a business has a line item. Flex (a Restive portfolio company) reached nine-figure annualized revenue within three years of launch, by giving business owners one platform for banking, spend, and cross-border payments. Bill, Tipalti, and Coupa are all businesses built on the fact that a CFO will pay a subscription to process an invoice faster.
That willingness to pay is why agentic payment adoption is a no-brainer in B2B, and the proof is in the nascent products already shipping. Ramp's Agents for AP do not merely recommend. For eligible invoices, the agent creates a virtual card for the exact invoice amount, enters the payment details into the vendor's portal on the due date, matches the transaction to the bill once it clears, and updates the accounting records, which is an agent executing a payment and closing the ledger entry. In its first months, Ramp says the agents flagged more than $1 million in fraudulent invoices, issued 90,000 approval recommendations with a 90 percent acceptance rate, and got 85 percent of accounting fields right on the first pass. Bill Pay, the product these agents live in, has tripled payment volume and doubled customers year over year. Mercury has also gotten into the game, allowing users to issue cards directly to agents.
The same pattern is emerging one layer down. Amazon shipped AgentCore Payments in April, built on x402 and Coinbase's wallet infrastructure, with spending limits enforced at the infrastructure layer. AgentCore now exposes more than 10,000 x402 endpoints for agents to discover and pay, real protocol adoption even though the transactions are mostly sub-cent API calls today. These are business purchases for developers in the AWS marketplace.
In our own portfolio, Natural is building the layer that lets businesses (AI native and otherwise) manage the complexity of payments that agents introduce. Success involves orchestration and reconciliation across tens of thousands of customer relationships and hundreds of millions of transactions.
Building in B2B
With the opportunity laid out, there are some important considerations for founders building in B2B agentic payments.
Don’t just be the tollbooth. As you are building, focus on the services that surround the transaction as opposed to the money movement in isolation. For example, the average invoice costs around $9 to process and takes more than a week to clear, not because moving the money is hard but because everything around it is manual. Agents convert that work, multiplied across every payment a business makes, into software revenue. This explains the difference between Bill.com’s $4.5 billion market cap and the fact that Ramp is currently raising a round valued at $60 billion: the bulk of Bill’s revenue is a fee on the transaction as opposed to the future of agentic operations for finance.
Controls, compliance, and security are core to the product. An agent with a wallet is, from the CFO's chair, a control failure waiting to happen. The products that are gaining B2B adoption today have all converged on the same pattern: the human sets the bounds, the agent operates inside them, and the agent cannot expand beyond that perimeter. Mercury's rule that agents cannot raise their own limits or unfreeze their own cards is the clearest statement of it. AgentCore's time-boxed budgets, along the lines of one dollar that expires in five minutes, are another, and Ramp's agents block money from leaving before the bill is even created. Layered on top are problems nobody has fully solved and ambiguous liability that will take time to resolve. Failing to move beyond strict perimeters or human-in-the-loop will shortchange the potential, but not getting the controls right will slow adoption. Expect a hard but steady improvement in capabilities, but founders should be continually pushing the boundaries here.
The majority of money sits with the slowest adopters. Natural's early customers are AI-native companies, AWS is selling AgentCore Payments to developers, and Mercury's agent cards are being picked up by founders running Claude Code against their bank account. That is the right place to start and the wrong place to stop, as these are the early adopters that will drive meaningful purchase volumes but do not represent the bulk of the $100 trillion-plus that moves between businesses each year. Find your wedge so you can get early feedback, but a multi-billion dollar company will need to find a way into the more established market.
Pricing is unsolved. Outcome-based pricing sounds right for agents but has still largely been resisted by CFOs unless it acts as a commission for revenue. Seat-based pricing is at odds with taking over the labor budget, and leaning on interchange monetizes the flow but earns you Bill's multiple rather than software's. The likely answer in the short run is a blend of a platform fee for predictability, usage to ensure token costs don’t get out of control, and a share of the payment economics. In the long run I expect we’ll land on a single pricing metric which will resemble outcome based pricing strategies. This space is evolving rapidly, so continue to experiment.
You have to decide how to sit with existing systems. There are three models for how to approach selling into this market: rip and replace existing systems, coexist alongside these systems, or find customers where this is their first technology buy. Rip and replace is typically lucrative, sticky, and a very slow sales cycle. Coexisting means the agent works on top of what is already there, reading the inbox, logging into vendor portals, and writing back to the ledger, which is how Ramp's agents operate today. Leapfrogging means going after the businesses that never bought software at all or they are new companies looking for the most modern solutions. Each path implies a different product, a different sales motion, a different ICP, and a different time to the money.
Platform vs. embed. Ramp and Mercury have been successful because they've had a captive platform of users executing agentic actions within that environment. This offers a lot of control on the part of Ramp and Mercury, but it doesn't embed within the workflows of an organization or adhere to the specifics of that company. The next wave of innovation will likely take a much more personalized, embedded model, and be much better positioned to take over labor budgets. But the ramp up will be slower because it requires customized workflows, deeper integrations, and there is no captive audience from Day 1.
We’ve rewritten much more about the systems needed to scale agentic commerce outside of the B2B context.
The consumer war will keep the headlines, and it will most likely be won by someone who already owns a rail or an agent surface. The B2B version will mint a new platform, because the flows are unclaimed, the buyers pay for efficiency, and the services stack has barely been built. The prize is here; the attention is not.
If you are building in B2B agentic payments, we want to hear from you!
1 bea.gov
2 https://www.fortunebusinessinsights.com/b2b-payments-market-108853
3 https://www.nacha.org/content/business-business-b2b
4 https://www.frbservices.org/resources/financial-services/wires/volume-value-stats/annual-stats.html
5 https://www.occ.treas.gov/publications-and-resources/publications/comptrollers-handbook/files/payment-sys-funds-transfer-activities/pub-ch-payments-previous.pdf. Note that this is the most recent data we could find, although presumably directionally correct
6 https://www.thebusinessresearchcompany.com/report/e-invoicing-global-market-report
7 https://www.grandviewresearch.com/industry-analysis/accounts-receivable-automation-market-report
8 https://www.custommarketinsights.com/report/accounts-payable-automation-market/
9 https://www.emergenresearch.com/industry-report/us-accounting-services-market
10 https://www.avidxchange.com/press-releases/avidxchange-agrees-to-be-acquired-by-tpg-in-partnership-with-corpay-for-2-2-billion/
11 https://investor.bill.com/news/news-details/2026/BILL-Reports-Fourth-Quarter-and-Fiscal-Year-2026-Financial-Results/default.aspx
12 https://www.moderntreasury.com/newsroom/press-releases/9-in-10-companies-struggle-with-payment-operations
13 https://ramp.com/blog/ramp-ap-agents-announcement
14 https://aws.amazon.com/about-aws/whats-new/2026/04/amazon-bedrock-agentcore-payments-preview/
15 https://www.wexinc.com/resources/blog/what-does-it-cost-to-process-an-invoice-ap-benchmarks-every-finance-team-should-know/
© 2026 Restive®, Inc.
