Banking Exchange x Paywhere Webinar
On October 29, Nick Lazo, Chief Revenue Officer at Paywhere, spent an hour with bank and credit union leaders on the Banking Exchange webinar series. We asked the audience three questions. Their answers describe where the industry stands better than any analyst note.
Three polls, one gap
Where is your institution planning, or actually budgeting for customer-facing AI? One in five said: already funded, piloting or live. Just over half said they are discussing it with no approved budget. The rest have no current plans or are not sure. Nobody chose the fourth option, "We're waiting for AI to submit the business case."
When do you expect customers to want to access your institution through their own AI assistant? Half said they already do, or will within 18 months. Just under 30 percent said more than 18 months from now. About one in five said they do not expect the demand at all.
If a customer asked today to connect an AI agent to your institution, how would you respond? Half said they have no defined approach or are not sure. One in five would review the request case by case. One in seven has an approved connection path or an active pilot. And alarmingly, but not surprising, one in seven chose "Please hold while I transfer you to the future."
Half the room expects the demand inside 18 months. Half has no answer for it today. The host called it an honest group. Nick's reply: not surprising, and the reason for the session. Waiting is a decision too. Right now it is an expensive one.
Four doors, and the one your customers are shown
An AI agent can reach a financial institution four ways today. Shared credentials: the customer gives the agent their login, and your institution cannot prove what was authorized. Browser automation: the agent drives a browser, and your fraud tools cannot tell it from a human. An aggregator: the agent connects through a provider like Plaid, and your institution becomes the data pipe behind someone else's interface. Or a governed front door: your institution sets the rules, decides what an agent may do, and keeps both the relationship and the insight.
Nick showed the room which door is the default. He opened a new AI assistant and asked, "Can you help me do my banking?" It said yes and offered a Plaid connection. Unless your institution offers a better path, that is the one your customers take.
The deposit math
Agents are starting to get wallets. Nick walked through one: the customer tops it up from a checking account, sets a budget, and tells the agent to order lunch every weekday at noon, buy concert tickets the moment a presale opens, and pay the apartment bills when they are due. It is early. The funding still comes from a bank account.
The pattern is familiar. Roughly $3 trillion has moved out of legacy institutions and into fintech deposit and investment accounts over the past five years, before AI was part of the picture. People vote with their dollars for convenience.
The counterweight is real. Your brand still means something to your customers. You hold the trust, the compliance muscle, the community presence and most of the deposits. In Nick's words, you are not out of the fight.
What the Q&A asked
What does a bank risk by waiting another year? Habits. People are learning to use AI now and building financial workflows into it. If your institution is not reachable when that happens, they will find one that is, and the habit will be hard to break. Then do the math on time: think about the last big initiative you shipped, how long it took, and how much AI changed in that window.
Does supporting customer AI agents mean replacing the core, hiring engineers or taking on a major technology project? No. The connection sits on top of existing infrastructure. Technically it is a matter of days. In practice, plan on a four-to-six-week project so everyone is comfortable, with help on how to introduce it to customers and employees.
Many banks are banning AI agent access. How do the security, fraud and privacy issues get resolved? Time, and a better door. Blocking is not anyone's long-term plan; the industry said the same about blockchain and then built its own. And since most systems cannot tell a human from an agent, a block does not hold. The durable version is a governed channel: rules, approvals, an audit trail, access you can revoke, and a human in the loop for any payment.
What about fully autonomous agentic payments? Not yet. If money is going to move, a human reviews and approves it, whether the request came from a live chat, a scheduled task or a persistent agent. Your institution decides how much diligence to add on top of that, such as multi-factor authentication.
One assignment
Nick left the room with homework. Pretend you are not a banker. Connect your accounts to an AI assistant the way a customer would, and ask it the questions a customer asks: what is going on with my spending, how do I save an extra $500 a month. Look at what comes back. That is the financial conversation your customers are already having. Whether your institution is in the chat is your decision.
His closing words were not a pitch. Paywhere sits on the same side of the table as the institution, and the first conversation is about what is happening inside your institution and what a governed front door would look like in your channels. No hard sell.
Watch the recording here.
Slides are available on request.
