Google's Agent Payments Protocol (AP2) launched in September 2025 with more than 60 partners, including Mastercard, PayPal, Salesforce, and PwC. It lets AI agents complete a purchase on a customer's behalf. That's forcing a question banks have been avoiding: once a bank's role in a transaction becomes invisible, what is a bank actually still providing?
That question sits underneath two terms banks use almost interchangeably: embedded finance and banking-as-a-service. They aren't the same decision. Confusing them is why so many banks can't answer a smaller, more useful question first: which one should we actually be doing?
What's the difference between embedded finance and banking-as-a-service?
Banking-as-a-service is the infrastructure layer. A licensed bank connects to a non-bank company through APIs, and that company builds a financial product on top of it under its own brand. The bank holds the license and the risk. The non-bank owns the customer.
Embedded finance is the broader outcome. It's what happens when a financial service, a loan or a payment, gets built directly into a workflow the customer already uses, at the moment they need it. A point-of-sale loan is embedded finance. Whether that loan runs on a BaaS partnership or a direct bank integration is a separate question.
BaaS is one way to deliver embedded finance. It isn't the only way, and for a bank deciding its own strategy, that distinction changes the whole decision. For the infrastructure-level detail behind the BaaS side of that choice, see 7 banking as a service platforms compared for 2026.
Why is this suddenly urgent for banks?
AP2 is the clearest signal that this decision can't wait. Google built it with Mastercard, PayPal, Salesforce, Adyen, Coinbase, and dozens of others. It lets an AI agent prove a real customer authorized a purchase, without a human clicking "buy" at all. PwC is named directly in Google's own announcement as a launch partner.
If an agent can already initiate and authorize a purchase, a bank's job changes. It's no longer only approving a transaction a person made. It has to decide in advance what an agent may do on a customer's behalf, and through which financial products. That's live infrastructure with named partners attached to it today, not a future scenario.
Every agent-initiated purchase still needs a Platform, a Realizer, and someone holding the license behind it. AP2 doesn't remove those roles. It just decides who fills them, faster than most banks are used to moving.
How big is the opportunity?
Estimates for this market vary widely, mostly because "embedded finance" gets defined differently by whoever's counting. PYMNTS Intelligence, working with Green Dot, put the global market at more than $7.2 trillion by 2030, and found something more useful than the number itself: trust in a provider, not speed of deployment, has become the deciding factor in how companies choose a partner.
That shift matters more than the total addressable market. Embedded finance stopped being a race to launch fast. It's a decision about who a business, or a bank, is willing to depend on long term.
Which of the four roles should your bank actually play?
A bank doesn't need to answer "embedded finance or not." It needs to answer a sharper question: which position does it want to hold once financial services stop living inside one app?
Four roles recur across the embedded finance value chain, each with a distinct job. MichaΕ Kurowski, Director at Strategy& Poland (PwC), laid out this framework on Banking Reinvented.
- Platform: owns the direct customer relationship and aggregates services from multiple providers underneath it.
- Realizer: provides the actual regulated financial product, the loan, the account, the card, that a Platform sells under its own brand.
- Enabler: supplies supporting infrastructure, like compliance tooling, that makes the other roles work.
- License Holder: carries the regulatory and financial scrutiny that lets any of the above operate legally.
Most banks default into Realizer without deciding to. A bank that wants to compete as a Platform, aggregating outside products under its own brand, needs a different roadmap than one content to stay the regulated engine behind someone else's app.
Deciding which role fits is the strategy question. Backbase's own conversation with Datos Insights covered three practical paths banks take to get there, partnering with a fintech, acquiring one, or building the capability in-house, in From concept to customer value: scaling embedded finance.
What's the best embedded finance platform for a bank?
There's no single best embedded finance platform, because "best" depends entirely on which role a bank just chose.
A bank moving toward Platform or Realizer, one that wants to own the stack and coordinate its own embedded finance products, is evaluating different vendors than a bank that wants to rent capability and get out of the way. Embedded finance platform comparison breaks down the infrastructure-level options for banks building that stack directly.
A bank that decides renting is the faster path should start with 7 banking as a service platforms compared, which compares providers built specifically for that model.
The platform comparison only makes sense after the role decision. Skipping straight to a vendor list is how banks end up locked into infrastructure built for the wrong role.
Where does agentic commerce fit into this?
OpenAI's own agentic checkout, built on the rival ACP protocol, launched inside ChatGPT in February 2026 and was already scaled back to an app-based model within weeks. That's not a sign agentic commerce is slowing down. It's a sign the standards underneath it are still being contested, and AP2, backed by the broader card network and processor coalition, looks like the more durable bet right now.
For a bank, the practical takeaway isn't which protocol wins. It's that a bank without a clear position on Platform, Realizer, Enabler, or License Holder will find that position assigned to it by whichever protocol and partner moves first.
Embedded finance and banking-as-a-service will keep getting used as if they're the same strategy. They aren't, and the gap between them is exactly where a bank either designs its own role or has one handed to it by default. Book a strategy call to work through which role fits your bank's existing infrastructure.
Frequently asked questions
What is embedded finance for business banking?
It's the integration of financial services, payments and lending most commonly, directly into the software a business already uses to run its operations, ERP systems or accounting platforms, so the business doesn't have to leave that workflow to bank.
What is banking-as-a-service for SMB banks?
It lets an SMB-focused bank become the regulated infrastructure behind someone else's product. Instead of serving business customers only directly, the bank connects its core systems through APIs so fintechs and software platforms can offer accounts and cards under their own brand, while the bank holds the license and manages the underlying risk.
How does embedded finance threaten commercial banks?
It threatens banks that default into a passive Realizer role, supplying the regulated product behind someone else's brand with no direct customer relationship. That bank has limited pricing power and no access to the data or loyalty that comes with owning the relationship. Choosing a role deliberately protects against that outcome. Drifting into one by default does not.
What's the best embedded finance platform for a bank?
It depends on the role a bank is choosing. Banks building their own Platform or Realizer stack should evaluate infrastructure providers on API strength and compliance coverage. Banks choosing to rent capability instead should evaluate providers built specifically for the banking-as-a-service model.
