DruckFin

PicPay Beats Guidance Across the Board as AI Freezes Headcount and Payroll Loan Risk-Taking Widens NPL Gap

Second quarter 2026 earnings call, August 24, 2026

PicPay delivered a second consecutive quarter of results that outpaced its own guidance on every major line item, but the call's real substance came from two structural stories: an AI deployment that has frozen headcount growth entirely, and a deliberate strategy of pushing further into higher-risk payroll loan segments that is now visibly showing up in delinquency metrics. Management was explicit that both dynamics are intentional rather than signs of weakening execution, though investors will need to keep testing that claim as the payroll book ages through 2026 and into 2027.

AI Is Now a Margin Driver, Not a Buzzword

The most concrete new data point on the call was headcount. CEO Eduardo Chedid said the company had originally planned for a 10% increase in staff this year; instead, headcount has been flat since October 2025 and will stay that way. "Productivity gains are translating directly into margin expansion rather than incremental hiring," he said, calling the current efficiency trajectory "not a ceiling but a floor." The adjusted efficiency ratio fell to 44.8% from 46.9% in the first quarter, a 210 basis point sequential improvement, and management is guiding to the low 40s by year-end.

Danilo Caffaro, PicPay's Vice President of Consumer Banking who also oversees AI initiatives, gave the most granular detail. The company built a proprietary AI harness with model routing and caching that cut token costs by 70% since the start of the year, allowing continued access to frontier models without scaling total spend. PicPay says it is the first Brazilian bank with an official plug-in in both the Claude and OpenAI ecosystems, and its consumer-facing agents now execute more than 70 tools, including bill payments, PIX transfers, and debt renegotiations, across app and WhatsApp channels. Internally, roughly 90% of employees are actively using the company's AI platform, and Caffaro said the number of employees contributing to real product deployments has doubled since January, adding that "most of them, such as myself, wouldn't be able to contribute without AI." A newly launched marketing AI agent for small and medium business clients generated 1,500 campaigns and reached 1.7 million individuals in its first week. A proprietary foundation model for personal loan underwriting is also rolling out, with an expected 15% to 20% improvement over the prior model.

Payroll Loan Risk-Taking Is Deliberate, But the Market Will Keep Pressing

PicPay's private payroll loan book grew to 7.2 billion reais with more than 3.6 million contracts, and the company resumed originations in growth-oriented risk clusters after resolving earlier operational issues. But the strategy is now visibly widening delinquency ratios. NPL over 90 days rose to 9.8% and Stage 3 reached 12.9%, and CFO Andre Cazotto guided that NPL over 90 days could reach "low teens" by year-end, converging toward current Stage 3 levels, purely as a function of portfolio aging and intentional risk expansion, not underlying deterioration. Citi's Gustavo Schroden pushed back directly, noting that Central Bank data shows delinquency in some payroll cohorts approaching levels seen in unsecured personal loans. Chedid's response was that PicPay has not seen deterioration within the same risk profile, only intentional expansion into riskier segments, and that the company is redeploying underwriting model gains into holding origination steady rather than accelerating growth further "so that asset quality remains in control." Cazotto added that the risk framework itself is unchanged, with the loss absorption ratio still targeted between 40% and 60% and ROE above 20%. On the specific question of how much stress the book can absorb, Chedid offered a hard number: private payroll loan vintages can withstand up to a 70% increase in delinquency rates before hitting breakeven, and current consensus unemployment forecasts, rising from 5.4% toward 6% through 2027, sit well within that buffer even in pessimistic scenarios peaking near 7.2%.

Desenrola Program Is Flattering Near-Term Credit Metrics

A meaningful portion of this quarter's asset quality improvement came from Brazil's Desenrola debt renegotiation program, which Bank of America's Mario Pierry pressed management to quantify. Cazotto confirmed the program contributed a 117 basis point improvement to NPL over 90 days and roughly 59 million reais, or about 5% of total cost of credit, as a positive impact on the cost of risk line. Stage 3 formation fell to 3.65% from 3.9%, but excluding Desenrola, the underlying figure would have been closer to 4%, broadly in line with prior quarters. Stage 3 coverage also declined to 74.1% from 77%, a mechanical effect of the FGO guarantee attached to renegotiated loans rather than a change in provisioning philosophy, and management expects coverage to drift back into the high 70% range as the effect normalizes. Cazotto said Desenrola will provide "some additional positive impact" in the third quarter but at a more limited scale than in the second.

Revenue Mix Is Shifting Toward Lower-Risk Streams

Total revenue grew 67% year over year to 4.1 billion reais, but the more important structural point was the mix shift: 71% of revenue now comes from no- or lower-credit-risk sources, float income, hedge accounting, fees, commissions, and secured or partially secured credit, up from 63% a year ago. Secured credit revenue nearly tripled year over year to 1 billion reais on the back of the payroll loan franchise, while noncredit revenue, spanning fees, insurance, and acquiring, rose 57% to 1.9 billion reais. ARPAC climbed to 92 reais per active client, up 52% year over year, driven primarily by cross-selling credit and insurance products into the existing base rather than new client acquisition, a dynamic management expects to persist as the primary growth lever going forward.

Cover Acquisition Closes, Rebrands as CAV

PicPay closed its acquisition of insurance platform Cover on August 3 after clearing insurance regulator, antitrust, and Central Bank approvals, and the business has been rebranded as CAV. Management guided to an 80 million to 100 million real net income contribution from the deal between August and December, with roughly 70% of Cover's existing business run through high-quality third-party distribution partners that PicPay expects to keep scaling. The deal will consume approximately 150 basis points of capital in the third quarter, though management still expects to close the year with a total capital ratio near 14% and a common equity Tier 1 ratio between 12% and 12.5%, comfortably above internal thresholds. Management did not yet have a revenue contribution figure to share when asked directly by RBC's Dan Perlin.

Guidance Points to a Net Income Dip That Isn't What It Looks Like

Third quarter guidance calls for adjusted net income of approximately 265 million reais, down 6% sequentially, even as pretax earnings are guided up 34% to around 360 million reais. Cazotto was direct about the disconnect: the second quarter benefited from a concentrated tax reduction tied to Brazil's Lei do Bem R&D incentive program, and the effective tax rate normalizes in the third quarter back toward first-quarter levels. Investors modeling straight-line net income growth off the second-quarter print should adjust for this one-time tax benefit rather than reading the sequential decline as an operating slowdown.

Disclaimer: This article is for informational purposes only and does not constitute investment advice or a recommendation to buy, sell, or hold any security. Our analysts provide detailed coverage of corporate events but can make mistakes, always conduct your own due diligence. The views and opinions expressed do not necessarily reflect those of DruckFin. We have not independently verified all information used herein, and it may contain errors or omissions. Before making any investment decision, consult a qualified financial advisor. DruckFin and its affiliates disclaim any liability for any losses arising from reliance on this content. For full terms, see our Terms of Use.