Business

Know the Business

Figures converted from KZT at historical period-end FX rates — see data/company.json.fx_rates for the rate table. Ratios, margins, and multiples are unitless and unchanged.

Kaspi.kz is a Kazakhstan bank wrapped in a daily-use super app — three profit pools (Payments, Marketplace, Fintech) sitting on one biometric identity, one deposit book, and one in-house underwriting engine. It is not a tech multiple business and it is not a regional bank, which is why the market keeps mispricing it: the bear sees a $21.9 billion balance-sheet lender exposed to a single emerging market; the bull sees software-grade margins on regulator-protected distribution. The thing actually worth underwriting is whether the three engines, taken together, can keep compounding at 50%+ ROE while management ports the playbook to Türkiye — that, not the consolidated 7x P/E, is the question.

1. How This Business Actually Works

Kaspi runs one app on top of one bank that monetises the same customer three different ways in the same session. A consumer opens the app to pay a utility bill, browses Marketplace for a fridge, takes a 12-month BNPL loan to finance it, and parks salary in a Kaspi Deposit on the way out. Each step is a different P and L line and a different margin profile, but the user only consents once — the moat lives in the consent, not the code.

Where the marginal profit actually comes from. The structural answer is deposit funding. Kaspi Bank holds $14.9B of customer deposits (93% in KZT), most of it in the form of low- or zero-interest current and savings accounts that exist only because customers use the app for payments and salary deposit. Those deposits fund the $14.2B loan book at a cost of funds materially below any bond-funded fintech, and the spread between that funding cost and the 24% Fintech yield is where the lion's share of net income is earned. The Payments business is the user-acquisition channel and the Marketplace business is the engagement maximiser; the Fintech business is the monetisation engine that converts engagement into cash.

Bargaining power. Merchants take the short end of every stick because the platform owns the demand and the only realistic alternative — Visa/Mastercard acceptance — costs more than Kaspi QR. Consumers have no realistic substitute either, given Halyk Bank is a slower legacy product and cross-border apps (Wildberries, Ozon) don't carry payments or BNPL. The real counter-power sits with the National Bank of Kazakhstan: one rule on consumer-loan reserves or one base-rate move can shift Fintech NIM by hundreds of basis points, which is exactly what happened in FY2025 (15.25% → 18.0%) and is the reason consolidated net income grew only 1% on a 60% revenue increase.

2. The Playing Field

There is no Kaspi competitor inside Kazakhstan that runs the same three-platform model, so the relevant peer set is the global cohort of emerging-market consumer-financial platforms — the companies investors actually substitute Kaspi against in their portfolios. None of them compete with Kaspi at home; all of them clarify what good looks like at Kaspi's economic engine.

What the peer set reveals. Kaspi sits in the top-right by margin but bottom-left by multiple — a combination that is rare and structural. MELI and SE generate single-digit operating margins because they run logistics-heavy, founder-subsidised commerce engines; the market still pays them 49–51x earnings because the runway is multi-country. Nu Holdings is the closest economic analog in mix (lending-spread plus payments) but trades at four times Kaspi's multiple on weaker ROE, because it operates in three large LatAm markets that together dwarf Kazakhstan's TAM. PagSeguro is the only peer that prints a Kaspi-like multiple, and it does so on a far weaker ROE and a far weaker market position — that combination is what the market currently treats Kaspi as, which is the central disagreement of the thesis.

The Hepsiburada line is the awkward one and worth reading carefully: Kaspi acquired 65% of it in January 2025 and now runs it; it lost money in FY25 (consolidated Türkiye net income –$184M), and management's own bar is "around EBITDA breakeven" for the Hepsiburada standalone. Hepsiburada is in the comp set as a strategic peer, not a financial benchmark — its consolidation drags the consolidated multiple down and adds optionality, not earnings.

3. Is This Business Cyclical?

Yes, but only one of the three engines is — and the cycle hits it through interest rates, not through volume. Payments and Marketplace are volume-led (cash-to-digital migration plus e-commerce penetration), so they ride a secular growth curve interrupted only by real-income shocks. Fintech is rate-led: when the National Bank of Kazakhstan tightens, deposit funding cost rises faster than the loan book reprices, NIM compresses for two to three quarters, and cost of risk follows 6–12 months later. FY2025 is the textbook example.

The visual makes the point: revenue grew double-digits across all three engines, but the net-income translation broke down only in Fintech and Marketplace — Fintech because of the rate move, Marketplace because of a one-off smartphone-import disruption and a 10% tax-on-liquidity levy. Adjusted for the four external shocks management itemises (smartphones, tax on government-securities revenue, minimum-reserve hikes, base-rate increases), consolidated FY25 net income would have grown about 18% rather than 1% — a useful sanity check that the engine itself is not broken.

Where the cycle is not. Kaspi's revenue base is not commodity-exposed (no oil, no metals), it is not tourism-exposed (domestic-services led), and it is not USD-funded (93% of deposits are KZT). It is, however, single-country: a Kazakhstan recession would touch all three engines at once — Payments TPV would slow, Marketplace GMV would contract, and Fintech cost of risk would climb. The diversification dividend works against external shocks; it does not work against a domestic one.

4. The Metrics That Actually Matter

The headline P and L line items are easy to read and easy to misread. The five metrics below explain Kaspi's value creation and its risk far better than revenue or EPS do, because they isolate the engagement-to-monetisation-to-credit-quality chain that the entire model rests on.

What the conventional ratios miss: a 51% ROE on a bank charter looks like a model error, but it isn't — it's the combination of (a) extremely cheap deposit funding from a captive app, (b) a 6-second AI-driven underwriting decision against in-app behavioural and bureau data, and (c) a payments business that earns spread on the same balance sheet. The day any of those three pillars degrades, the ROE will normalise — that is the right warning system, not the multiple.

5. What Is This Business Worth?

The right way to underwrite Kaspi is not one valuation lens. It is three weighted by segment net income, because the three engines have genuinely different multiples in any rational world. The market today applies a single bank-multiple anchor to all three, which is the source of the discount; the bull case is that as the segment mix shifts (Payments and Marketplace growing faster than Fintech), the blended multiple should drift toward platform multiples. The bear case is that single-country regulator risk caps the multiple regardless of mix.

Reading the sum-of-the-parts honestly. Take the three KZ engines at $700–860M of net income each, apply the right multiple to each, weight to consolidated, and the resulting fair-value range sits well above the current 7x consolidated multiple — but only if you give credit to the Payments and Marketplace engines as platforms rather than as fee lines on a bank. The single biggest valuation question is therefore whether the market will allow that separation. Today it does not.

The dividend matters here. Management resumed the dividend in FY25 ($1.68 per ADS proposed), and the payout policy is built on the same bank-charter cash generation that explains the high ROE. A 4–5% dividend yield at the current multiple is real cash return, and it puts a floor under the multiple while the Türkiye execution plays out. The question to ask is not what fair value is — it is what multiple expansion you need to make money, and the honest answer is: very little. A re-rating from 7x to 10x earnings on flat earnings is +43%; on growing earnings, more. That is the asymmetry.

6. What I'd Tell a Young Analyst

Stop fighting the multiple debate at the consolidated level. The 7x P/E is a sum, not a price tag — three engines with different right-multiples, weighted by net-income share, plus a Türkiye option, minus a single-country discount. If you cannot break it apart, you will trade the headline ratio and miss the thesis.

Watch the transaction frequency, deposit funding share, and cost of risk in that order — they are the early-warning chain that tells you whether the engine still works. The revenue and EPS lines arrive too late. If transactions per active consumer slip below 70 or DAU/MAU below 65%, the marketplace and lending lines tend to follow within two to three quarters; conversely, if those KPIs hold while the NBK starts cutting from 18%, the Fintech NIM expansion is a multi-quarter tailwind that the consolidated P and L typically shows late.

Read the FY2025 print correctly: 60% revenue growth, 1% reported net income growth, 18% normalised net income growth excluding four itemised regulatory and rate shocks. The right question is not "did earnings grow?" — it is "did the engine grow?" The engine grew. The shocks were external and rate-cycle reversible.

The Türkiye optionality is real but it is not free. Hepsiburada lost money in FY2025 and consolidates as a drag; the Rabobank A.Ş. banking licence has to clear before the deposit-funded lending model can be ported. Underwrite it as an option, not as cash flow. The Kazakhstan business pays for the option; do not pay full price for the option on top.

Finally — the bear case that should change your mind is not a Kazakhstan recession (that's known and bounded). It is regulator-mandated open-loop QR interoperability that compresses Payments take rate the way PIX did in Brazil, combined with a consumer-loan reserve hike that compresses Fintech NIM further, combined with a Türkiye execution stumble. Any one of those is digestible; the combination is what would force a real re-underwrite. Track all three independently.