TL;DR
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Splitit CEO Nandan Sheth told PYMNTS that AI agents should weigh a consumer’s liquidity, existing obligations and borrowing costs before recommending pay later. He said a Splitit test found higher conversion when shoppers received added repayment information, but the report does not provide test methods or sample size. Sheth also acknowledged privacy concerns and expects consumers to retain decision-making authority in many cases.
Splitit CEO Nandan Sheth told PYMNTS that AI agents will need more than product prices and monthly-payment amounts to recommend pay-later financing responsibly: they should also account for a consumer’s available cash, existing obligations and total borrowing cost. His comments frame a developing question for AI shopping tools: whether they can help people compare financing choices without taking control away from the consumer or exposing sensitive financial data.
Sheth made the case in an interview with PYMNTS CEO Karen Webster, describing AI as a potential adviser that can compare ways to pay, rather than merely surface a financing offer at checkout. He said the starting point is that the agent works for the consumer, and expects people initially to review recommendations and make the final choice themselves. A joint study by Splitit and PYMNTS Intelligence, cited by Webster, found that 61% of consumers would accept an AI recommendation for credit or pay later, while 2% would let the agent decide on its own. The report does not provide the survey’s sample size or field dates.
Sheth said the recommendation should put an installment amount in context. An $80 monthly payment, for example, may look different when an agent also considers liquidity, other payment plans and the total financing cost. He described a Splitit test comparing one presentation that showed a monthly payment, APR and total cost with another that also explained there was no prepayment penalty and that a customer could pay off the balance early. Sheth said the latter version had a conversion rate about two to 2.5 times as high. The account does not specify the test’s sample, dates, or whether the result has been independently verified.
The discussion also covered decisions after checkout. Webster raised the possibility that a consumer could choose lower payments at first, then repay the balance once more money becomes available. She described an ongoing exchange informed by the timing of income and expenses. Sheth said an agent could recommend paying immediately for one purchase and using pay later for another, depending on the transaction’s economics.
Why Full-Context Payment Advice Matters
Pay-later offers are often presented as a payment amount or a choice among checkout methods. Sheth’s proposal would make an AI agent compare that offer with a consumer’s cash position, other debts and likely repayment costs. That could help shoppers see when installments fit their circumstances, rather than treating a lower monthly bill as sufficient evidence that borrowing is a good choice.
The stakes extend beyond the financing decision. If consumers rely on agents to choose payment methods, the agent’s ranking could influence which lenders, cards or merchants receive a transaction. Sheth linked this to the industry’s focus on being a consumer’s preferred or “top-of-wallet” payment method: an agent could assess a purchase using financing costs, liquidity and other benefits, rather than simply repeating a shopper’s usual preference. That possibility also gives data access and recommendation practices a direct role in household financial choices.
The reported conversion result is relevant to merchants, but it is not evidence that added context improves a consumer’s financial outcome. A higher rate of completed purchases does not, on its own, show whether customers borrowed less, paid less interest or made choices that suited their budgets. The test details supplied in the report are limited, so its result should be read as a company executive’s account, not a general measure of AI’s effect on borrowing.
As an affiliate, we earn on qualifying purchases.
From Checkout Offers to AI Advice
Sheth illustrated AI shopping recommendations with a personal example: while looking for a watch for his wife, an AI tool returned several prices for the requested product and suggested an unfamiliar boutique brand based on information it had about him and his family. Sheth said he bought the alternative and that it was roughly 10% to 15% cheaper. The example concerned product discovery, not a pay-later recommendation, but it shows the kind of personal information that could shape an agent’s suggestions.
Applying that approach to financing requires different inputs. Sheth named bank balances, credit-card statements, revolving balances and brokerage statements as possible sources of information, while saying an agent would not necessarily need access to all of them. Even limited, securely shared data, he argued, could help with choices involving spending, saving, investing or borrowing. The report does not describe a live Splitit product that gathers those records and makes such recommendations.
For transactions where consumers may be more willing to delegate, Sheth pointed to routine, lower-value purchases. He used groceries as an example: a person might set a $300 monthly budget, describe typical and discretionary purchases, then ask an agent to find suitable deals. He said that a defined limit could make it easier to imagine an agent completing payment. These were examples of potential uses, not a timetable or announcement of an operating service.
“a dynamic relationship between agent and consumer”
— Karen Webster, PYMNTS CEO
As an affiliate, we earn on qualifying purchases.
Privacy, Testing and Consumer Control
The report does not identify a launched Splitit service that connects to bank, card or brokerage records to choose between immediate payment and financing. It also gives no technical details about how a consumer would grant, limit or revoke an agent’s access, how recommendations would be audited, or who would be responsible if an agent’s advice caused harm. Sheth suggested that processing information locally on a personalized language model could limit how much sensitive data leaves a consumer’s environment; the report does not establish that this approach is in use or explain how it would work in practice.
The conversion comparison also lacks key details, including the test population, sample size, period, and the exact financing terms shown. It is not clear whether consumers in the test made better-informed choices or whether the higher conversion rate translated into better outcomes for them. Nor is there a confirmed timeline for consumer adoption of agents that complete purchases or accept credit on someone’s behalf. Sheth said willingness to delegate would likely vary with the purchase, merchant and transaction size.
As an affiliate, we earn on qualifying purchases.
How AI Payment Choices May Develop
For now, the development described is a set of proposals and reported findings from an executive interview, rather than a product launch or regulatory decision. The next evidence readers would need includes details of any payment agent Splitit introduces, the financial information it requests, the protections around that data and whether users can approve each recommendation before a transaction is made.
Further information about the cited consumer survey and conversion test would help establish how broadly the figures apply. In particular, test design, participant numbers and financing terms would clarify what the reported conversion difference measures. More detail on data localization and consumer consent would also show whether Sheth’s proposed safeguards can address the concerns he described.
As shopping agents develop, the practical test will be whether they explain trade-offs clearly and preserve consumer choice. A useful pay-later recommendation would need to show not only the installment amount, but also the cost of borrowing, repayment flexibility and how the choice fits a person’s other obligations. Until specific systems and evidence are available, those remain proposed capabilities, not confirmed outcomes.
As an affiliate, we earn on qualifying purchases.
Key Questions
What does Splitit’s CEO say AI agents need to recommend pay later?
Sheth says they need context beyond the monthly payment, including a consumer’s available liquidity, existing obligations and total financing cost. He also identified financial records such as bank balances and credit-card statements as possible inputs, though he said an agent would not necessarily need all of them.
What did the Splitit test reportedly find?
Sheth said a version that added information about early repayment and the absence of a prepayment penalty produced a conversion rate about two to 2.5 times as high as a version showing the monthly payment, APR and total cost. The report does not supply the sample size, test period or other methods.
Are consumers ready to let AI choose credit or pay later?
A joint Splitit and PYMNTS Intelligence study cited in the interview found 61% would accept an AI recommendation, while 2% would let an agent decide independently. The report does not state the survey’s sample size or field dates, and Sheth expects recommendations to precede full delegation.
What privacy concerns did Sheth raise?
Sheth said that giving an agent access to sensitive financial information raises fears of a data breach or misuse. He proposed local processing as one possible way to limit data sent outside a consumer’s environment, but the interview did not describe a deployed system or specific protections.
Has Splitit announced an AI agent that makes payment decisions?
The source report describes Sheth’s views, examples and a company test; it does not announce a launched agent that connects to personal financial accounts and selects pay-later financing. A product, availability date and operating safeguards were not specified.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
