PicPay (PICS): AI Platform Lifts Efficiency Ratio 210bp, Freezes Hiring

PicPay (PICS) told investors on its August 24, 2026 earnings call that its in-house AI platform now reaches most of the company and that its adjusted efficiency ratio improved 210 basis points sequentially [1]. Management attached no specific amount to AI itself.

What PicPay does

PicPay is a Brazilian digital bank that grew out of mobile payments and transfers and now earns most of its money from credit, lending to consumers and small merchants and capturing the spread between loan yields and funding costs, with payment, account and cross-sell fees on top.

It is a people-heavy technology bank. Headcount was 4,156 at the end of 2023, 4,202 at the end of 2024 and 4,629 at the end of 2025, with 2,712 of those employees in technology and product development [3]. Staff is the company's largest variable cost, so whether AI can substitute for hiring lands directly in the expense line.

The AI platform inside the company

PicPay built its own AI stack at the start of 2026 and connected agents to daily work in coding, design and credit product development. Management says roughly 90% of employees use the platform, most of them daily, and about 30% now ship AI-produced work into production.

To make that affordable, the company engineered down its inference costs: token spend is 70% lower year to date, which lets it keep using frontier models without scaling total token cost. There is a customer-facing side as well. Second-generation WhatsApp and in-app agents execute more than 70 multi-step transactional tools, and new marketing agents for small merchants drew 10,000 opt-ins and 1,500 campaigns in their first week, though nothing on that side is monetized yet [1].

The headcount claim

Management stated the cost result plainly: AI's impact is already measurable, headcount has been flat since October 2025, the 10% increase planned for 2026 will not happen, and productivity gains are converting into margin instead of hiring [1].

The half of that claim an outsider can check holds up. Headcount actually grew 10.2% in 2025, so a 10% hiring plan for 2026 was a real planning baseline rather than a convenient reference [3]. The other half cannot be checked. The last hard headcount figure is end-2025, no 2026 disclosure carries a new number, and the 4,202 figure in public databases is the older count. Flat headcount is also within this company's own history: it grew just 1.1% in 2024, with no AI attribution at the time [3].

The efficiency ratio, and what moved it

The improvement itself is real. The adjusted efficiency ratio was 44.8% in the second quarter against 46.9% in the first, the CFO restated it in Q&A, and management guided the year-end level into the low 40s or high 30s [1].

The ratio measures expenses against revenue, so it can improve because costs fall or because revenue accelerates. This quarter it was the latter. Holding the definition constant, total costs still grew about 11.8% sequentially and adjusted operating expenses rose to 955 million reais. Nothing shrank; management's own phrasing is that spending grew at a fraction of the pace of revenue [2]. The quarter also carried roughly 30 million reais of World Cup marketing pulled forward, which the CFO expects to reverse next quarter [2].

What the 210 basis points do and do not show

The more accurate reading is that a bank with rapidly scaling revenue held expense growth well below it, with AI named as one contributor. Management's own term is a key driver of the dynamic, not the driver. No amount and no basis-point figure is attributed to AI, the flat-headcount statement is a counterfactual against an internal plan that no outside party can test, and credit performance remains the larger swing factor in this bank's earnings [2].

What is confirmed and what is not

The AI platform is genuinely running at scale inside the company, with disclosed coverage, deployment rates and inference-cost figures, and this is the first time management has framed it in income-statement terms while calling the efficiency trend a floor rather than a ceiling. What remains unquantified is how much expense AI actually saved, and no 2026 headcount disclosure exists to anchor the claim that hiring has stopped.

Use case: cost reduction. Adoption stage: scaled. Value released: moderate. Confidence: medium.

Sources

[1] Drillr · PicPay (PICS) · 2026-08-24 · earnings call

"A key driver of this dynamic is AI. Its impact on our operations is already tangible and measurable. Our headcount has been flat since October 2025, and the projected 10% increase we had originally anticipated for 2026 will not materialize. Productivity gains are translating directly into margin expansion rather than incremental hiring."

[2] Drillr · PicPay (PICS) · profit and expense verification

[3] Drillr · PicPay (PICS) · headcount and efficiency verification