Designing the Trust Layer for Autonomous Commerce.
With 73% consumers now use AI in their shopping journey. AI agents are starting to shop on behalf of users. They browse, compare, and purchase across merchants. But checkout is currently is breaking context and drop off occurs when consumers are redirected.
Role: Design lead
I led agentic checkout, owned checkout strategy, Defined trust framework, Identified return member gaps, codified presentment principles, cross functional alignment, collaborated with 8+ partners
Team: Product manager | Content designer | Researcher | Engineering team | Partnership teams
Strategy & Roadmap | Cross-functional alignment with partnerships | Product Design | Visual Design
For PayPal
As partners like Google and OpenAI build their own shopping protocols, PayPal risks becoming invisible. I identified this gap and defined how PayPal shows up in agentic commerce, not just as a payment button, but as a trust layer between users and agents.
Key Metrics
Must wins:
Branded TPV from agentic channels
Number of Branded Checkout transactions
Transaction margin (Member vs. Guest mix)
Leading:
Branded Member presentment rate
Branded Member selection rate
Member vs. Guest mix
Problem: Checkout should not be the most tedious step.
Hypothesis
By keeping users in-context, we will help reduce context break and make the experience more seamless, increase conversion, and reduce drop-off.
Conversation recommendations might feel untrustworthy. “How did they know my information?”
Inline Checkout
I started with the known user pain: in an agentic flow checkout is being redirected. My initial instinct was inline checkout, embed PayPal directly inside the agent's interface. No redirect. Fewer steps, less drop-off, higher conversion.
Different exploration of how something similar to the existing PayPal paysheet can appear in context.
Elegant solution failed
Two blockers killed it.
Technical: PayPal's auth requires a login redirect that couldn't be embedded in third-party agent UIs.
Psychological: users didn't want to enter credentials inside an unfamiliar interface.
“It's a bit scary. How does the AI know my payment information? I also just noticed there's buy now pay later options. I think it's nice to have everything in conversation, but it feels fake and maybe I have to be even more aware of fraud.” -Annia, 35 · Online shopper · Frequent AI user
So I pivoted to redirecting to standard PayPal checkout. The redirect to PayPal wasn't friction it was a trust signal. Removing would suppress selection rate.
A Trust System
I started with the familiar paysheet from traditional checkout. Instead of building new components, I took existing elements that were already earning trust.
I reframed the redirect as a feature, not a bug, and designed a trust system around the standard checkout flow. Instead of eliminating steps, I made every step build confidence.
Inconsistent Presentment
Trade-off between partner design systems vs PayPal visibility
Inconsistent Presentment
Trade-off between partner design systems vs PayPal visibility
Consistent Presentment
Regardless of the entry, checkout experience mirrors the paysheet users already know. Same layout, same interactions, same sense of control.
In an unfamiliar agentic context, familiarity is a deliberate trust anchor.
Presentment Principles
Trade-off: Partners don’t want additional branding vs PayPal trust signals
Subtle design choices that’s still meaningful: One of the presentment principal was 'Powered by PayPal', a trust signal for guests. Even if they didn’t use PayPal to checkout, seeing a familiar logo is still better than trusting a unfamiliar agent. Terms of service goes to user agreement page.
Upstream messaging, payment selection, button consistency, standard paysheet
The result we got from Perplexity was 99% conversion overall, but 70% of add-to-cart users never clicked the PayPal button.
Return members
Based on the PerPlexity test, return members were droping off to due to inconsistency. Typically, they see a ‘vaulted’ page where their payment is already store. From there, it’s a one-click checkout. This validated the hypothesis for consistency.
Returning members have been key levers for PayPal. They converted at significantly better rates. They still had pain points such as not being able to edit their card or use Pay Later.
I created a design that improved vaulting before the checkout button.
Similarly, this applied to return buttons. Vaulted buttons remain a existing checkout issue even outside of agentic checkout.
Add Link PayPal to <Platform>
Create an additional step for more control
Return ‘vaulted’ button
Return users previously had issues with a one-click experience. They can’t see their card they previously saved, they can’t change or edit.
“It’s too fast (checkout). Sometimes I just want to pause and reviews the card. I don’t always remember what’s on file and I can’t even change my card.” -Karl, 40 · Frequent online shopper · Member for 8 years”
Business conversion isn't user confidence
Insight and learnings: Trust over speed.
Post purchase confidence
Dispute and return rate
Repeated usage
Agent to manual fallback
‘Instant Buy’ Vision
The foundational framework was trust over speed, every design decision prioritized user confidence over fewer steps. But the test data revealed a second opportunity: moving checkout upstream. Payment execution wasn’t the issue. Commitment was.
Instead of waiting for users to reach a payment step, we surface PayPal at the moment they're already considering a purchase. This isn't about skipping steps, it's about meeting shoppers where they already are, giving them the information and confidence to commit earlier. The checkout comes to the user, not the other way around.
Learnings + Vision
I still believe inline is the future to solve for in-context checkout. Once we have more trust from users, figure out the technical constraints, we can provide a truly, in-context experience.
Additionally, since PayPal already has merchant catalog information, we can also expand PayPal’s agentic checkout capabilities by being a embedded shopping assistant to smaller 1st party merchants.