Long-Term Thesis
Long-Term Thesis — 5-to-10-Year View
Figures converted from Kazakhstani tenge at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.
1. Long-Term Thesis in One Page
The long-term thesis is that Kaspi can compound owner value at a mid-teens to low-twenties annual rate over the next 5 to 10 years only if the Kazakhstan stack — bank charter, 93% deposit funding, 6-second AI underwriting, 68% DAU/MAU habit, embedded government rails — keeps generating mid-30s ROE through whatever the National Bank of Kazakhstan does to fintech margins, and the Türkiye build converts a loss-making Hepsiburada into a second deposit-funded super-app within 36 months. This is not a long-duration compounder unless both halves work. Take the Kazakhstan stack away and the engine collapses to a regional bank trading at single-digit P/E; take Türkiye away and the home franchise still pays a 4–5% dividend on a 7x multiple but the reinvestment runway is capped at a 20-million-population country whose digital penetration is already 78%. The combination is what produces the compounding asymmetry — ROE held above 51% through an NBK tightening cycle at 6.0% Cost of Risk is the highest-confidence wide-moat reading in the file, and a Türkiye stack that turns from drag to engine inside 36 months is the only credible path to multi-country compounding. Failure modes are observable and few: open-loop QR mandate (PIX-style), a Krivenok class-action denial that confirms the disclosure tail, a Rabobank A.Ş. licence rejection, or founder Lomtadze departure. Skip the next quarter's print; underwrite these.
Thesis strength
Durability
Reinvestment runway
Evidence confidence
FY25 ROE (group)
FY25 Operating Margin
Loan book deposit-funded
FY25 Net Income ($M)
The thesis hinges on two questions, not twenty. (1) Does the Kazakhstan stack still earn software-grade margins on regulator-protected distribution five years from now? (2) Does Türkiye become a second-country super-app by the late 2020s? Bull and bear at the consolidated level both reduce to these two over a 5-to-10-year horizon — everything else is noise around them.
2. The 5-to-10-Year Underwriting Map
The driver that matters most is the first one. Drivers 2–6 are all conditional on the Kazakhstan stack holding — if KZ Payments take rate compresses to 0.7–0.8% on an open-loop-QR mandate, the deposit-funding advantage softens, and every other driver re-prices. The Türkiye option is the upside underwriting; the KZ stack is the base underwriting. A reader who only has time for one variable should track Kaspi Payments take rate (currently ~1.10% on $85B TPV) and Cost of Risk (currently 6.0% through an 18% NBK rate environment). Those two numbers, observed quarterly for five years, will tell you whether the moat is still wide.
3. Compounding Path
The compounding case rests on a simple multi-decade math: a deposit-funded lender with a daily-use app on top can grow its loan book in line with its deposit book without diluting equity, and the spread between deposit cost and loan yield earns mid-teens ROIC even after credit losses. Add operating leverage in Payments and Marketplace VAS and you get a ROE-led compounder that does not need leverage cycles to grow.
Revenue compounded at roughly 43% in USD terms over the eleven-year window (KZT depreciated from 221 to 520 per USD across the period, so native-currency growth was ~56%); net income at over 60% from the FY16 trough. The visible re-acceleration in FY25 is acquisition (Hepsiburada consolidation), and the Net Income line flattens precisely as the operating margin reset to 55% absorbed four exogenous shocks. The right way to read this chart for the long term is to ignore FY25 as a one-year regime change and trust that ROE is the durable metric — and ROE held at 76–97% for five consecutive years before the deal.
The single most load-bearing number in this report is the ROE line above. From FY19 through FY24, ROE moved in a 76–97% band while NBK base rate cycled between 9% and 16.75% — the moat survived full rate cycles before the Hepsiburada consolidation. The FY25 step-down to 51% is the Hepsiburada accounting consolidation (loss-making Türkiye drags reported equity yield), not the stack breaking. The compounding case asks the reader to believe that this 50-80% ROE band re-emerges as Türkiye turns from drag to engine; the bear case is that 51% is the new ceiling.
Three honest readings of this scenario set. First, the base case requires almost nothing aspirational — Kazakhstan grows at the cash-to-digital migration rate, Türkiye breaks even, and the multiple drifts from 7x to 9x. Second, the bear case still pays a real dividend yield because 93% deposit funding plus a 100% loan-to-deposit ratio mean no balance-sheet repair is needed even if growth slows. Third, the bull case is asymmetric on multiple expansion: a re-rate from 7x to 12x on growing earnings is +120% before any operating beat. The reinvestment runway is the swing factor — without Türkiye, only the bear and lower-base cases stay live.
Balance-sheet capacity matters for the long-term thesis because reinvestment options widen when the bank's own funding is cheap and the parent has unused debt headroom. Net debt to EBITDA at 0.16x and BBB- investment grade (Fitch) mean Kaspi can issue another $1–2 billion of senior unsecured paper without breaking covenants. That headroom is what funds a multi-country build without equity dilution — share count has only moved 0.4% over two years, even through the Hepsiburada deal.
4. Durability and Moat Tests
Five tests for whether the moat holds. Each has a competitive, financial, or regulatory load-bearing pillar; all five have observable signals.
Two tests carry asymmetric weight. Test 1 (Payments take rate) is the external binary — a regulator-mandated PIX-equivalent rule would commoditise the rail in a single ruling, exactly as it did to PagSeguro in Brazil (revenue growth slowed to 8%, op margin to 11%). Test 4 (Türkiye stack portability) is the internal binary — without a Turkish banking licence the Hepsiburada P&L stays a marketplace P&L and the reinvestment runway closes. Tests 2 and 3 are confirmatory rather than binary: they tell you whether the stack still works on the same engine, not whether the engine itself was replaced.
5. Management and Capital Allocation Over a Cycle
The five-person management board has been together for 17–19 years, with CEO Lomtadze in the chair continuously since 2007 and four members holding Harvard Business School degrees. That continuity is the single most under-priced moat in the file: every product (Kaspi Pay, Postomats, e-Grocery JV with Magnum, Kaspi Alaqan pay-by-palm) was conceived, shipped and iterated by the same operator team, and the cadence has not stalled through two NBK rate cycles, a KZT devaluation from 221 to 520 per USD, a global pandemic, and a US class-action complaint.
The capital-allocation record splits into two regimes. From FY20–FY24, Kaspi paid out roughly $3.7 billion of cumulative dividends — a near-100% LSE-era payout consistent with a mature compounder treating shareholders as the residual claimant. FY25 broke that pattern in a single year: dividends went to zero, capex jumped fourfold to $352M, and $1.1B went out the door for Hepsiburada (funded partly by a $600M Eurobond placed at BBB-). Then Q1 2026 reinstated dividends at $1.74/ADS, restoring the 4–5% yield floor. The optics of the dividend skip in the same year insiders accelerated selling are governance-poor; the substance — redirecting cash to a strategic deal at 1x revenue (versus MELI at 51x) — is exactly the long-term-thesis-aligned move you want to see.
The pattern over twenty-four years is clean: deliberate sequencing of charter → digital platform → cross-sell stack → public-market currency → cross-border step. No equity-funded acquisitions; no dilutive secondaries; no impulsive M&A; no overpayment for growth. Hepsiburada was bought at roughly 1x revenue with proceeds from the dividend pool plus a hard-currency bond — the deal economics line up with the long-term thesis even if FY25 P&L did not. The three open governance questions are real: (1) no disclosed succession plan; (2) related-party flows with Magnum (Kim) and Kolesa (Lomtadze) are small in absolute terms but structurally awkward; (3) the unresolved Krivenok litigation and ICFR material weakness sit on top of the founder-controlled structure. Each of these can be tested against management decisions over years — what cannot be tested over a single quarter.
6. Failure Modes
The bear case that should change a long-term holder's mind is not a Kazakhstan recession (that's known and bounded) and not a single bad quarter. It is one of the five thesis-breakers below.
The combination of failure modes 1, 2 and 6 inside a single 12-month window is the genuine thesis-breaker. Any one alone is digestible — Kazakhstan still pays 4-5% dividend at bank multiples even on a re-rated ROE; Hepsiburada still has option value at 1x revenue. Two together force a partial re-underwrite. Three together break the 5-to-10-year compounding case and the right response is to size down, not to wait for confirmation.
7. What To Watch Over Years, Not Just Quarters
Five multi-year signals will tell you, with high confidence, whether the long-term thesis is widening or fading. Each is observable in primary filings or central-bank releases at quarterly frequency, but the signal lives at the multi-year level — single-quarter noise should be discounted.
The long-term thesis changes most if Rabobank A.Ş. closes and Hepsiburada engagement starts visibly compounding toward Kazakhstan levels — that single multi-year signal converts the Türkiye option from "free in the price" to a second deposit-funded super-app, lifts the consolidated reinvestment runway from a 20-million-population country to a combined 100-million-plus, and is the only path by which the bank-multiple discount collapses on a fundamental basis rather than a sentiment basis.