Brazil built one of the most ambitious Open Finance frameworks in the world, and for its first years almost nobody outside a bank could feel it. The regulation let a person authorize one institution to see their accounts at another and, later, to move money between them. What the market did with that permission was narrow. Companies asked users to connect their accounts, read their income and spending, and used the data to decide whether to offer them credit and at what rate.
Open Finance became a background process. The person who granted the consent got a loan offer, or nothing, and never saw the connection again. Magie was the first company to treat it as a product experience: something a person opens, uses and comes back to, inside a conversation on WhatsApp.
Context
Magie is a conversational bank on WhatsApp. AI agents handle Pix, boletos, balances and payments through a chat, for Magie's own users and, through its B2B platform, for the customers of partner organizations. Most of Magie's users already hold accounts at one or more traditional banks and were never going to move everything to a new one. That fact shaped the whole idea: instead of asking people to bring their money to Magie, let them bring their banks to the conversation. I am the only designer on the team and part of the founding group, responsible for the conversational experience across Magie's journeys.
The problem
Two problems sat on top of each other. The first was the market's. Open Finance had been implemented as data extraction: connect, read, score, offer credit. Nobody had built a daily-use product on top of it, so users had no reason to connect an account and no reference for what a good connection experience looked like. The second was ours.
The Open Finance journey has hard constraints that fight a conversational interface. Consent legally has to happen inside the user's own bank, which means leaving WhatsApp mid-flow and coming back. Balances at other banks are sensitive in a way a chat notification on a lock screen makes worse. Payments, authentication and the parameters of a connection need structured, verifiable input that a free-text message cannot provide, and the Central Bank's rules on consent and confirmation leave no room for improvisation.
And the vocabulary is regulatory: "consent", "initiation", "holding institution", words that mean nothing to a person paying a phone bill.
The risk was building something technically correct that nobody used, which is what the rest of the market had done.
My role and the decisions that shaped it
I designed the conversational experience of Open Finance at Magie: how a person connects a bank, what they can do once it is connected, how the agent talks about money that lives somewhere else, and how the mandatory friction of the regulation is handled inside a chat.
Eight decisions carried the work:
Lead with the outcome, in the user's words. The pitch is never about the mechanism. It is "pay your bill with the balance from any of your accounts, without opening the bank app," and the agent teaches by showing the literal sentence a person can type or say, like "pay my bill with my Itaú balance." Showing the command is the shortest way to explain a conversational product.
Turn the consent redirect into a security signal. The user has to leave WhatsApp and authorize the connection inside their bank. Hiding that would be dishonest and impossible. So the agent says it plainly: the authorization happens in your bank's app because it is your bank confirming with you, and I will tell you here when it is done. The friction the regulation imposes becomes the reason to trust the product.
Talk about sufficiency, never about numbers, when the agent speaks first. A proactive message can say the connected balance already covers the bill. It never prints the balance of another bank, because that message may sit on a lock screen. The statement belongs to the bank. The convenience belongs to the channel.
Use WhatsApp Flows where the regulation demands structure, and keep everything else conversational. A chat is a bad place to choose an institution from a list, set a daily limit, define how long a connection lasts, or type a password. It is also a place the Central Bank's rules were never written for: consent has to be explicit and unambiguous, authentication has to happen on a protected surface, and a payment confirmation has to show the person exactly what they are approving. WhatsApp Flows, the native forms that open on top of the conversation, solve all of this. They gave us screens with real fields, lists and secure inputs, inside the same window, without sending the person to a browser or another app.
Use WhatsApp Flows where the regulation demands structure, and keep everything else conversational. A chat is a bad place to choose an institution from a list, set a daily limit, define how long a connection lasts, or type a password. It is also a place the Central Bank's rules were never written for: consent has to be explicit and unambiguous, authentication has to happen on a protected surface, and a payment confirmation has to show the person exactly what they are approving. WhatsApp Flows, the native forms that open on top of the conversation, solve all of this. They gave us screens with real fields, lists and secure inputs, inside the same window, without sending the person to a browser or another app.
The trade-off is real and I accepted it knowingly: those moments become more linear and less conversational than the rest of the product. What they buy is compliance by construction, since a Flow cannot be misread or half-answered the way a message can, and they cost the user almost nothing, because the form opens and closes within the chat and the agent picks the thread back up as soon as it is done. The rule we settled on is simple. Anything the regulator needs to be able to audit happens in a Flow. Anything a person just needs to say happens in the conversation.
Put every payment behind a full summary and a secure surface. Before any money moves, the agent states amount, source account and date in one line in the chat, and the confirmation happens through a password inside a Flow, never in the chat keyboard. The person confirms what they are seeing, on a screen built for it.
Give control back explicitly. Users set a daily limit, choose how long the connection lasts, connect more than one bank and disconnect at any time at no cost. This is said out loud in the flow because it is what separates automatic debit from a surprise on a statement.
Offer automatic debit at the moment of highest trust. Right after a payment has been scheduled and confirmed, the agent offers to put the next bills on automatic, always with a warning before each debit. The product's real goal lands when the person has just seen it work.
One rule tied these together and ended up as a house rule for the whole platform: in a financial chat, each message carries at most one benefit and one question. If it needs a second paragraph, the benefit was chosen badly or the message is actually two.
Engineering and I worked through the constraints that could not be solved in copy, especially how identity verification travels between app and WhatsApp and how the consent redirect returns the person to the conversation. Product owned scope, and scope was kept deliberately narrow: a connection has to be immediately useful for one thing before it earns the right to do everything.
What the product does
A person connects one or more bank accounts to Magie in a few messages and one authorization inside their bank. From then on, inside the same WhatsApp conversation, they can pay boletos, pay bills and send Pix using the balance of a connected account, see the balance of all their accounts in one place, schedule payments, and leave recurring bills on automatic debit from the bank of their choice.
The agent reminds them when a bill is due, tells them whether a connected account covers it, asks which account to use, confirms with a summary in the chat and a password in a Flow, and reports back when the payment is done. Connecting a bank, setting limits and authorizing a payment feel like short forms that appear and disappear inside the conversation. Everything else feels like talking.
Results and impact
Open Finance in Brazil had been a pipe for pulling data out of people. Magie showed it could be a product people reach for, and did it in the most intimate channel Brazilians have.
Open Finance Consent
Biggest Open Finance Consent in any B2C product.
Transactioned
Transactioned between banks using Magie's and Iniciador infrastructure
Magie shipped the first Open Finance experience in Brazil designed around daily use rather than credit scoring. The same journey became a building block of the B2B platform: partner organizations can offer their own customers the ability to pay that partner's bills with the balance from any bank, in the partner's own voice, without building any of it themselves.
The larger result is the reframing. Open Finance in Brazil had been a pipe for pulling data out of people. Magie showed it could be a product people reach for, and did it in the most intimate channel Brazilians have.