American Express Company (NYSE: AXP) — A Lender Wearing a Network’s Clothes, Priced at the Seam

An independent fundamental research note. Date: 2026-06-11 · Price: $313.34 (2026-06-10 close) · Market cap: ~$214B · Diluted shares: ~682M Sector: Financials — Consumer Finance / Closed-Loop Payment Network · FY end: December

⚡ Claude’s Take

This block is the author’s own subjective opinion and general information only — not investment advice. The analysis that follows deliberately carries no recommendation and no price target; this opening block is the single exception.

Verdict: HOLD a wonderful business at a fair-to-full price; accumulate on weakness toward ~$250–270 (≈14–15x forward EPS), trim conviction above ~$370 (≈21x). Not a short — the franchise quality forbids it. Conviction: medium-high on quality, medium on entry.

American Express is one of the dozen-best franchises in financial services: a ~34% ROE earned on almost-all-tangible equity, a closed-loop data moat the open networks structurally cannot copy, demonstrated pricing power (annual fee per card $92→$117 in two years with retention intact through the hikes), and a genuinely valuable demographic land-grab — the majority of new accounts are now Gen-Z/millennial, acquired on premium fee-paying products, with credit better than the industry’s older cohorts. The +18%-compounding net-card-fee annuity is slowly turning a cyclical lender into something more network-like. That is the bull case, and it is real.

The reason I land on HOLD rather than BUY is entry, not quality. At ~19–20x trailing and ~18x forward, AXP sits at the 76th percentile of its own ten-year valuation history even after a 19% pullback — the opposite of Visa/Mastercard, which trade in the bottom decile of theirs. You are buying a balance-sheet consumer lender at a benign point in the credit cycle (net write-offs 2.3%, below 2019) on a rich-vs-own-history multiple. That is the textbook late-cycle double-risk: if the affluent consumer rolls over, current earnings are modestly inflated by sub-normal provisions and the multiple re-rates toward the lender bracket — both compress together, and my bear case is a ~35% drawdown. The fee annuity cushions that fall; it does not eliminate it. Framing: a quality-compounder-at-a-price — own it, pay up only grudgingly, and let the credit cycle, not the narrative, set your entry. The single thing that flips me bullish: write-offs peaking below ~3.5% through a genuine consumer slowdown while card fees keep compounding double-digit (proof the de-cyclicalization is real, which earns the network-adjacent multiple). The single thing that flips me bearish: 30+ delinquencies breaking above ~2% with billed-business growth turning negative — the tell that the premium book is cracking and the multiple has nowhere to go but down. Tag: a great house bought at the top of its own street.

1. Executive Summary

American Express is a closed-loop, “spend-centric” payments and lending franchise that is simultaneously the card issuer, the network, and (for most of its volume) the merchant acquirer — a “three-party” model structurally distinct from the four-party Visa/Mastercard rails. That architecture lets AXP keep the entire merchant discount (~2.24% of billed business in FY2025), roughly ten times a pure network’s net take, in exchange for funding rich rewards, bearing consumer credit risk, and holding regulatory capital. The result is a hybrid: part high-margin toll-taker, part balance-sheet lender, and the central analytical question — in the business and in the valuation — is where on that spectrum the company truly sits.

The franchise is executing at a high level. FY2025 delivered total revenues net of interest expense of $72.2B (+10%), diluted EPS of $15.38 (+10% reported, +15% excluding the prior-year Accertify gain), and an ROE of ~34% on a clean, almost-all-tangible equity base (goodwill is only $4.9B). Billed business reached $1,669.8B (+8%). The standout is the net-card-fee annuity — $9,993M, +18%, growing double-digit for 30 consecutive quarters — a recurring, prepaid, recession-resilient revenue stream whose average fee per card rose from $92 (FY2023) to $117 (FY2025) while retention held through the price increases. That is the cleanest moat evidence in the file: a franchise that raises prices and keeps its customers.

The growth engine beneath the headline is demographic. Gen-Z and millennial customers are now the largest share of U.S. consumer spending and AXP’s fastest-growing cohort (Q1-2026 cohort spend: Gen-Z +38%, millennials +13%), acquired at an average age of 33 (Platinum) / 29 (Gold) on fee-paying premium products, with credit performance better than the industry’s Gen-X and boomer base. International (ICS) has grown double-digit FX-adjusted for 20 straight quarters on infrastructure that is already built. Management guides FY2026 revenue +9–10% and EPS $17.30–$17.90 (~13–16% growth), and frames a long-term algorithm of “10%+ revenue, mid-teens EPS.”

The skeptical counterweights are three. First, cyclicality: ~76% of revenue (discount revenue + net interest income) is spend- and balance-driven, and a meaningful slice is affluent discretionary spend (luxury, travel) that compresses in a recession; AXP is a real lender ($224.8B of loans and receivables) whose current EPS is flattered by sub-2019 write-offs. Second, valuation: the stock trades at the 76th percentile of its own ten-year history despite a 19% pullback — peak-benign credit times a rich multiple. Third, two modest governance/capital knocks: buybacks are mechanical rather than valuation-sensitive (AXP repurchased into all-time highs at an average $312.87 in FY2025), and the annual bonus scorecard is growth-tilted (revenue growth is 30% of the scorecard vs. ROE 15%), though the long-term plan is properly anchored to relative ROE.

Capital allocation is above-average for a large-cap financial: a disciplined framework (hold CET1 ~10.5%, pay a 20–25% dividend, sweep the rest to buybacks) that has retired ~7% of shares since 2022 and grown the dividend >80%, M&A confined to value-accretive tuck-ins (Center, Resy/Tock, Nipendo) with non-core divestitures (Accertify), and a structural endorsement in Berkshire Hathaway’s passive 22.1% stake, which compounds upward as buybacks shrink the denominator. The business quality is not in dispute. The debate is entirely about the price of admission at this point in the credit cycle — the subject the valuation and variant-perception sections below adjudicate, without rendering a recommendation.

2. Business Overview

What American Express actually is. AXP is an integrated, closed-loop payments and lending company — what the industry calls a “three-party” network. In a Visa or Mastercard transaction, four distinct parties share the economics: the card-issuing bank, the network (Visa/MA), the merchant’s acquiring bank, and the merchant. American Express collapses three of those four roles into a single entity: it issues the card (in its proprietary business), it operates the network, and it acquires most of its own merchants. The FY2025 10-K states it directly — AXP has “access to information at both ends of the card transaction, distinguishing our integrated payments platform from the bankcard networks.” Management’s shorthand is the “end-to-end integrated payments platform” running a “spend- and fee-centric model” (Q1-2026 call, 2026-04-24). (FACT)

How it makes money. Because AXP owns the merchant relationship, it keeps the entire merchant discount rather than splitting interchange with a third-party issuer. FY2025 total revenues net of interest expense were $72,229M (+10%), composed as follows (FACT, FY2025 10-K, Consolidated Results of Operations):

The economic engine is discount revenue — roughly half of all revenue — a toll on cardholder spending that ran 2.24% of billed business in FY2025 (down from 2.27% in 2024 and 2.29% in 2023). This is the single most important comparison in the thesis: AXP keeps ~2.24% gross of every dollar spent versus Visa’s ~0.24% net take-rate (per Visa’s reported network economics). AXP captures ~10x the take per transaction because it is paid for issuing, lending, and acquiring — but in exchange it funds rewards, services, and the consumer credit risk the pure networks never touch. (FACT / INTERPRETATION)

The four reportable segments (FACT, FY2025 10-K, Business Segment Results):

Note that GMNS, at just 10.7% of revenue, throws off $3,968M of pretax income — the second-highest of any segment and ~25% of reportable pretax profit — because it is the high-margin, low-credit-risk network/merchant layer. The credit-bearing consumer and commercial businesses are revenue-heavy but margin-diluted by rewards and provisions. This internal split is the quantitative proof that a real, Visa-like toll-road sits inside American Express. (FACT / INTERPRETATION)

Proprietary vs. network (GNS) cards. Worldwide billed business on AXP-issued (proprietary) cards was $1,670B across 86.6M proprietary cards-in-force in FY2025. Separately, the Global Network Services (GNS) business licenses banks in ~110 countries to issue Amex-branded cards and acquire merchants; worldwide processed volume on third-party-issued Amex cards plus alternative-payment volume was only $227.2B. Total cards-in-force including network partners were 152.8M; proprietary were 86.6M. (FACT, FY2025 10-K) AXP is therefore overwhelmingly a proprietary closed-loop issuer, not an open network — GNS is a strategic appendage providing international coverage and fee income, not the core engine. (INTERPRETATION)

Charge cards vs. lending cards. AXP runs both. Charge cards (the historical Platinum/Gold core) carry no preset spending limit and are generally paid in full each cycle — a payment product, not a credit product. Lending cards generate the $17.4B of net interest income (24% of revenue) on Card Member loans. The mix matters for cyclicality: the charge book is near-zero-loss, while the lending book carries the consumer credit risk. (FACT)

Recurring vs. cyclical. The most durable revenue is net card fees ($9,993M, +18%) — a contractual annuity that recurs regardless of spend volume and grows fastest. Discount revenue and net interest income are spend- and balance-driven, therefore pro-cyclical — they fall in a recession as billed business and loan balances contract. The card-fee annuity is the recession ballast; the discount/NII body (76% of revenue) is the cyclical mass. (INTERPRETATION)

Verdict (Business Overview). A genuinely differentiated, vertically integrated payments-and-lending model whose closed-loop architecture is the source of both its ~10x take-per-transaction and its credit-bearing cyclicality. The business is not one thing — it is a high-margin network bolted onto a premium consumer lender, and the quality of the whole rests on the affluent customer base that makes both halves work.

3. Industry Dynamics

The payments value chain and where AXP sits. Electronic payments is a secular-growth, oligopolistic, high-return industry — management pegs category growth at “about 8% every year” (Q4-2025 call), driven by the multi-decade migration of cash and check to card and digital. The structure splits along two models. In the open-loop / four-party model (Visa, Mastercard), a merchant pays a merchant discount rate composed of three pieces: interchange — the largest slice, set by the network but paid to the issuing bank (the network keeps none of it); a small network/scheme fee (what Visa/MA actually keep, ~0.24% net); and the acquirer’s markup. In the closed-loop / three-party model (American Express, and now Discover under Capital One), one entity is issuer, network, and acquirer at once, so there is no interchange leg to hand off — AXP keeps the entire ~2.24% gross discount, but funds rewards and bears credit risk out of it. (FACT)

This is the structural reason AXP’s take is ~10x a network’s and why its margins are far lower than Visa’s ~50%-plus net margin: it is a bank + network + acquirer in one, not a pure toll-road. The same architecture, however, gives AXP something the open networks lack — a proprietary data advantage from seeing both ends of every transaction, which underpins best-in-class fraud and credit outcomes (see the relevant section). (FACT / INTERPRETATION)

Regulation — AXP’s relative shelter is a genuine structural edge. The regulatory threats that hang over the open networks largely miss the three-party model:

• Durbin Amendment / interchange caps. Durbin caps U.S. debit interchange paid to issuers in four-party networks. Because AXP is a three-party network with no separate interchange leg, it is largely outside the direct interchange-cap regime — a real, durable advantage. (FACT)
• Credit Card Competition Act (CCCA). The proposed routing-competition bill would force a second unaffiliated network onto the largest Visa/MA credit cards. As a closed loop with no separate routing leg, AXP is again largely insulated, though final bill text could in principle sweep in three-party networks above a size threshold — an open question to monitor. (FACT / OPEN QUESTION)
• CFPB late-fee rule. The $8 late-fee cap was vacated / under litigation through 2025; AXP’s affluent base makes late fees a small revenue contributor, so the impact is modest regardless of outcome. (FACT / INTERPRETATION)
• International interchange caps. The EU (0.2%/0.3%), Australia, and Canada have historically pressured AXP’s GNS economics in those markets and contributed to coverage/merchant-discount mix shifts; the 10-K flags ongoing interchange/MDR regulation in those geographies as a focus area. (FACT)

The acceptance gap — the industry’s defining AXP-specific feature, and it is closing. Historically, AXP’s higher merchant discount meant fewer merchants accepted it — a real demand-side handicap versus near-universal Visa/Mastercard. Management now asserts near-parity acceptance in the U.S.: “you did not have parity coverage in the United States… now every place takes it” (Bernstein, 2026-05-28), with the network reaching “over 170 million merchant locations worldwide” (Q4-2025 call). International coverage is the remaining gap — and therefore the stated growth lever (see the relevant section). The closure of the domestic acceptance gap removes the single biggest historical reason a premium consumer would not carry an Amex as a primary card. (FACT / INTERPRETATION)

Competitive intensity. The premium-card and small-business-card arenas are intensely competitive. In premium consumer, Chase Sapphire Reserve, Citi, and Capital One all chase the affluent rewards customer; management concedes “the competitive dynamic in consumer is as tough as it’s ever been” (Q4-2025 call). In commercial, software-led fintechs (Ramp, Brex, and Capital One’s acquisition of Brex-adjacent capabilities) are attacking the SME expense-management-plus-payments space. Squeri’s framing is that the category has been fiercely contested “since the financial crisis” precisely because it is “a great return on assets… a great way to deploy capital” that grows ~8% a year — i.e., the high returns attract capital, the classic Marathon capital-cycle dynamic. (FACT)

The Capital One–Discover merger (closed 2025) — a third closed loop. This is the most credible structural change to the industry in years: Capital One now owns the Discover network, creating a third vertically integrated issuer-network with its own rails and two-sided data. Discover’s network is sub-scale internationally and skews mass-market rather than affluent, so near-term it competes more with AXP’s cash-back/lending flank than its premium core — but over a multi-year horizon a well-capitalized closed-loop competitor with merchant data is a genuine erosion risk, not a benign one, and belongs in the risk matrix (see the relevant section). (INTERPRETATION)

Verdict (Industry Dynamics): structurally good industry, AXP in a structurally privileged niche. Payments is a secular grower with high barriers and elite returns, and AXP occupies the most regulation-sheltered position within it — the three-party model that sidesteps the interchange caps and routing mandates that threaten the open networks. The offsets are that AXP competes in the most contested sub-segment (premium rewards + SME) and, unlike Visa/MA, carries consumer credit risk and regulatory capital. Net: a structurally good industry, but a more cyclical, more capital-intensive, lower-margin seat within it than the pure networks occupy.

4. Competitive Position

American Express’s moat is not one thing — it is a reinforcing stack of four mechanisms, each of which must be pressure-tested against the requirement that a moat show up in financial outcomes that would deteriorate without it.

(a) The spend-based “virtuous circle” — Greenwald: customer captivity interlocked with economies of scale. This is the real core. Rich rewards and benefits attract affluent, high-spending cardholders → high billed business per card → a loyal high-spend base that is worth a premium merchant discount → discount revenue funds still-richer rewards → repeat. Average proprietary basic Card Member spending was $25,453 in FY2025 (up from ~$24,608 in 2024). The 10-K makes the merchant-side argument explicit: AXP’s spend “is higher on average on a per-card basis versus our network competitors, offer[ing] superior value to merchants in the form of loyal customers and larger transactions.” This is a genuine two-sided flywheel, and the hardest piece to replicate, because it requires simultaneously an affluent base, merchant scale, and a rewards engine funded by the spread. A new entrant cannot bootstrap any one leg without the other two. (FACT / INTERPRETATION)

(b) The closed-loop data advantage — a proprietary intangible/scale asset. AXP sees both ends of every proprietary transaction (cardholder and merchant), which the bankcard networks structurally cannot. The strongest evidence is fraud: management states “our fraud is about 1/3 of what the networks are because of the data that we have,” and AXP has run machine learning on structured transaction data “since 2010” (Bernstein, 2026-05-28). The same data feeds best-in-class underwriting (see the credit metrics in the relevant section). The data moat is real and underpins measurable cost advantages today; its incremental value in an AI/agentic-commerce world (the “ACE Developer Kit,” Agent Purchase Protection, Q1-2026 call) is speculative — an option, not a proven earnings driver — and should be treated as upside, not base case. (FACT / INTERPRETATION)

© The premium brand and affluent base — Greenwald: intangible (brand) feeding customer captivity. The Amex brand carries demonstrable pricing power. Net card fees grew +18% in FY2025, average fee per card rose to $117 (from $103, then $92 two years prior), and after the U.S. Consumer Platinum fee increase, retention held at pre-refresh levels with roughly a quarter of the portfolio already billed at the higher fee (Q1-2026 call). A franchise that raises prices and keeps its customers has the cleanest possible moat evidence — pricing power that flows straight to the financial statements. (FACT)

(d) Scale in marketing and membership assets — economies of scale. AXP spends ~$6B/year on marketing and maintains an unmatched membership-asset stack — 90+ Centurion and partner lounges, the Resy and Tock dining platforms (25,000+ restaurants), ~3,400 Fine Hotels & Resorts properties — that smaller issuers cannot match on a per-card basis (Bernstein, 2026-05-28). These are fixed-cost assets amortized across a large base; Squeri’s pointed comparison to Chase Sapphire’s launch was that “we didn’t have Resy… the lounge network… fine hotels and resorts” — capabilities a pure issuer cannot replicate quickly. (FACT / INTERPRETATION)

Switching costs — moderate, not high. Cards are not high-switching-cost products in the enterprise-software sense; a consumer can leave. AXP’s stickiness is behavioral and loyalty-driven — accumulated Membership Rewards balances, status, ecosystem lock-in via lounges and Resy, plus the sunk annual fee — rather than a contractual switching cost. The proof of stickiness is the retention held through fee hikes, which is real but softer than a true switching-cost moat. (INTERPRETATION)

Network effects — real but asymmetric. The model is genuinely two-sided (more cardmembers → more merchant value → more acceptance → more cardmember value), but AXP’s network is thinner than Visa/Mastercard’s because the historically higher discount capped merchant adoption. The network effect is potent within AXP’s affluent niche; it is not the near-universal effect the open networks enjoy. (INTERPRETATION)

Direct comparison vs. the relevant competitors:

• vs. Visa / Mastercard: AXP keeps ~2.24% gross versus ~0.24% net — far higher take per transaction, plus a data advantage the open networks lack. But AXP bears rewards, credit risk, servicing, and regulatory capital, and earns roughly half the margin. Visa/MA are higher-quality, lower-risk, more universal businesses; AXP wins on take and data, they win on margin, capital intensity, regulatory simplicity, and acceptance breadth.
• vs. Chase / Capital One in premium lending: AXP’s differentiation is the closed-loop merchant-funded offers plus the membership-asset ecosystem a pure issuer cannot replicate — the moat is the integrated platform, not the card itself.

The credit-cycle caveat on the moat. Much of the virtuous circle rests on affluent discretionary spend — FY2025 strength was led by luxury retail (+18%), front-of-cabin air travel (+12%), and Fine Hotels & Resorts (+50% engagement) (Q1-2026 call). In a genuine consumer recession, billed business and loan balances contract, discount revenue and net interest income fall, and provisions rise — the flywheel runs in reverse on the spend dimension even as the fee annuity holds. The moat is durable in structure, but the earnings stream it protects is meaningfully pro-cyclical. The fee annuity (14% of revenue, +18%) is the recession ballast; the discount/NII body (76%) is the cyclical mass. (INTERPRETATION)

Verdict (Competitive Position): durable, multi-layered advantage — one of the better moats in financials, but narrower and more cyclical than the pure networks. The Greenwald test passes unambiguously: the ~2.24% gross take is sustained, pricing power is proven by +18% card-fee growth through fee hikes with stable retention, fraud runs at ~1/3 of network rates, and the lifetime-value skew toward affluent younger cohorts is real. The closed-loop data-and-spend flywheel is genuinely hard to replicate. The qualifications are honest: (1) acceptance was a real handicap, now largely closed in the U.S. but still a lever internationally; (2) Capital One–Discover is a credible third closed loop; (3) the protected earnings are pro-cyclical and credit-bearing in a way Visa/MA’s are not.

5. Growth History and Forward Opportunities

Headline growth. Billed business reached $1,669.8B in 2025 (+8%), after $1,550.9B (2024, +6%) and $1,459.6B (2023). Total revenues net of interest expense compounded from $42.4B (FY2021) to $72.2B (FY2025), a ~14.3% CAGR, and grew +10% in FY2025 (+9% FX-adjusted). Diluted EPS reached $15.38 (+15% excluding the FY2024 Accertify gain). Q1-2026 accelerated: spend +10% — “the highest quarterly growth in 3 years” — with revenue +11% and EPS $4.28 (+18% YoY). Proprietary cards-in-force grew to 86.6M (+4%), with 12.5M proprietary new cards acquired in 2025. (FACT, FY2025 10-K MD&A; Q1-2026 call)

The Gen-Z / millennial mix shift — the central growth engine. This is the most valuable and least-appreciated part of the story:

• Over 70% of new accounts are on fee-paying products (Q1-2026 call); globally, millennial and Gen-Z customers are AXP’s fastest-growing cohorts.
• As of Q4-2025, millennial and Gen-Z customers made up the largest share of U.S. consumer spending and remained the fastest-growing cohort (Q4-2025 call).
• Average age of new customers: 33 on U.S. Consumer Platinum, 29 on U.S. Consumer Gold (Q4-2025 call) — a multi-decade runway as these customers’ spend and income rise with age.
• Q1-2026 cohort spend growth: Gen-Z +38%, millennials +13%, Gen-X +8%, boomers +4% — growth is structurally front-loaded to the young.
• Critically, the credit quality of these younger cohorts is better than the industry’s Gen-X and boomer performance (Q1-2026 call) — AXP is acquiring “the cream of the crop,” not the median young consumer. The high-yield-savings franchise corroborates: younger cohorts are more than half of new deposit accounts.

This is the rare growth story where the new customers are younger, higher-lifetime-value, fee-paying, and better credits — the opposite of the “grow the book by loosening underwriting” trap. (FACT / INTERPRETATION)

International runway. ICS is “our fastest-growing segment,” with billings up double digits FX-adjusted for the 20th consecutive quarter (Q1-2026 call; Q4-2025 international spend +12% FX-adjusted). Management’s framing is that the expensive global network and merchant infrastructure is already built and at scale, so incremental international growth carries little new fixed cost (Bernstein, 2026-05-28) — operating leverage embedded in a runway that is years long. (FACT / INTERPRETATION)

The premium refresh cycle. AXP refreshed the U.S. Consumer and Small Business Platinum in 2025 (plus refreshes in roughly a dozen countries). The result: U.S. Consumer Platinum spend accelerated ~6 percentage points, “the majority… coming from tenured Card Members” — i.e., the existing back book, not just new acquisition — with retention holding through the fee increase (Q1-2026 call). Card-fee growth is guided to “exit 2026 in the high teens.” The refresh playbook (add benefits, raise the fee, retain the customer, lift spend) is a repeatable lever AXP can pull product-by-product, geography-by-geography. (FACT)

The fee annuity as durable growth. Net card fees (+18% in FY2025, the fastest-growing revenue line, double-digit for 30 straight quarters) are the highest-quality component of the growth story: contractual, prepaid, recession-resilient, and growing on both card count and per-card fee. This is the line that, over time, structurally lowers AXP’s cyclicality and earns it a multiple above the lender bracket. (FACT / INTERPRETATION)

Verdict (Growth): high-quality growth, with one cyclical caveat. The growth is organic, demographically front-loaded, premium-mix-accretive, and increasingly fee-annuity-driven — not financial engineering or low-quality lending growth. The Gen-Z/millennial capture is genuinely valuable: high share-of-wallet at acquisition, superior credit, and decades of rising spend ahead. International and the refresh cycle provide multi-year runway on already-built infrastructure. The caveat is that a large slice of the spend-driven revenue is affluent discretionary spend that compresses in a recession; the fee annuity and credit quality are the offsets. Net: high-quality, durable growth that is somewhat pro-cyclical on the volume lines but anchored by a fast-growing, recession-resilient fee annuity.

6. Financial Quality

Revenue quality — three engines, each with different quality. Decomposing the top line over FY2021–FY2025 (FACT, 10-K MD&A; XBRL):

(FY21–22 service-fee figures were recast in later filings and are not perfectly like-for-like; the discount, card-fee, NII and total lines are clean.)

1. Discount revenue (52% of revenue) is the capital-light closed-loop annuity, tied to spend rather than credit — the highest-quality line. The one yellow flag is the slow erosion in the realized discount rate (2.29%→2.24% over three years, on geographic and merchant-mix shift toward lower-rate categories), which means discount revenue structurally lags billed-business growth (billed business +8% in FY2025, discount revenue +6%). (FACT / INTERPRETATION)
2. Net card fees (14% of revenue) — the standout annuity. Compounded ~17.8% over FY21–FY25 to a record $9,993M; average fee per card rose $92→$103→$117. This is fee-per-card and card-count both rising — genuine pricing power, the closest thing to a subscription annuity in the P&L. (FACT)
3. Net interest income (24% of revenue) is the lender’s engine; net interest yield on Card Member loans was 8.1% in FY2025 (7.9% FY2024, 7.3% FY2023). The 2022–23 rate tailwind is largely spent — from here NII growth must come from balance growth and revolving-loan mix, not a new rate uplift. (FACT / INTERPRETATION)

Margin and operating leverage — real, but partly a “spend-to-grow” reinvestment machine. AXP’s income statement runs revenue net of interest expense → less provisions → less total expenses → pretax income. Total expenses of $53,178M (FY2025) decompose into two very different buckets (FACT, 10-K):

The central question is whether operating leverage genuinely exists or whether this is a treadmill. The honest answer is both, on different lines:

• VCE (~52% of revenue) scales with the franchise by design. Rewards (+11%) and Card Member services (+27%) are “generally driven by volumes and usage” and rise as AXP sweetens premium value propositions — they do not deliver leverage; management holds the VCE-to-revenue ratio roughly flat (guided ~44% on its narrower definition for 2026) and reinvests incremental scale into growth rather than letting it fall to margin. (FACT)
• Operating expense is where leverage shows up. Management states “OpEx as a percentage of revenue is down 4 points since 2022, even as we increased technology spend by 11%” (Q4-2025 call); salaries grew ~9.6% CAGR vs. ~14.3% revenue CAGR. Genuine leverage on the fixed/semi-fixed base. (FACT)
• Marketing (~$6B) is growth capex in disguise — fully expensed customer-acquisition spend (“lowest cost of acquisition for Platinum in the last 2 years,” Q4-2025) that builds a multi-decade fee-and-spend annuity per premium card. Because AXP expenses rather than capitalizes it, reported earnings are understated relative to the economic asset — a conservative, not aggressive, accounting posture. (INTERPRETATION)

Pretax margin expanded modestly from ~17.4% (FY2023) toward ~19% (FY2024–25, ex-distortions) — real but bounded leverage, because management deliberately reinvests most of it. This is a quality-compounder that grows revenue at 10%+ and EPS in the mid-teens by reinvesting scale, not an ever-expanding-margin network. (INTERPRETATION)

Credit quality — best-in-class and stable. AXP is a real lender: Card Member loans of $145.9B net (of a $5.9B reserve) plus charge-card receivables of $61.9B net, total $224.8B (+8%). The credit metrics are the best in U.S. consumer lending (FACT, 10-K Selected Metrics):

A 2.3% net write-off rate and 1.3% 30+ delinquency are roughly half typical large-bank card levels — the structural payoff of the affluent, fee-paying base. Both metrics remain below 2019 (pre-pandemic) levels (Q4-2025 / Q1-2026 calls): credit has normalized but not deteriorated past pre-COVID norms. The provision line confirms a stable cycle: total provisions were $5,256M in FY2025, up only 1% despite 8% loan growth, and the reserve build on loans shrank to just $198M (from $1,353M in FY2023) — meaning provisions are increasingly just write-offs, the signature of stabilized credit. Reserve coverage actually declined to 3.89% under CECL, consistent with a benign forward view. The charge-card book is near-zero-loss ($61.9B of receivables carry only $180M of reserves). (FACT / INTERPRETATION) The standing risk: current EPS is flattered by sub-2019 provisions, and the affluent book’s resilience at its current Gen-Z-heavy, fee-maximized mix has not been tested in a real recession. (OPEN QUESTION)

Balance sheet and capital — strong. Total assets $300B; customer deposits $152.5B (~51% of the balance sheet) — a low-cost, sticky, granular direct-to-consumer funding base that is a genuine advantage over wholesale-funded card lenders; long-term debt $56.4B; cash $47.8B (a fortress 16% liquidity buffer). CET1 of 10.5% sits well above the 7.0% regulatory minimum (4.5% + 2.5% Stress Capital Buffer floor) — lean but appropriate for short-duration, high-yield, low-loss card assets, and the source of the “excess capital we return” framing. Goodwill is only $4.9B (~15% of equity), so there is no goodwill cushion masking weak returns: ROE was 33.9% in FY2025, and ROTCE is ~40%+ on a clean tangible base. (FACT / INTERPRETATION)

Quality of earnings — clean. The only material one-time item in the window is the Q2-2024 Accertify sale ($531M pretax / $479M after-tax / $0.66 per share gain), which inflated the FY2024 base and means clean FY2025 growth (+15% ex-Accertify) was higher than the +10% headline — a case where the one-time item understates momentum and management’s disclosure is transparent. Operating cash flow of $18.4B comfortably exceeds net income of $10.8B (provisions and rewards accruals are non-cash) — cash generation exceeds reported earnings, the opposite of an earnings-quality red flag. Decomposing FY2025 EPS growth: of ~9.8% diluted-EPS growth, ~7.0 points was net-income growth and ~2.8 points was share-count reduction — EPS growth is ~70–75% operating, ~25–30% buyback, not financially engineered. (FACT / INTERPRETATION)

Verdict (Financial Quality): economics genuinely improve with scale — on the returns and OpEx lines. The proof of moat is unambiguous: ~34% ROE / ~40%+ ROTCE on clean tangible equity, best-in-class credit that is a structural cost advantage, a low-cost deposit funding moat, and an ~18%-compounding fee annuity with demonstrated pricing power. The honest caveat: this is not an asset-light, ever-expanding-margin model — ~52% of costs are engineered to grow with the franchise, so margin expansion is modest and deliberate. AXP compounds value through reinvested growth at high ROIC, not through margin inflation — with the standing risk that the premium consumer is cyclical and the realized discount rate is slowly eroding.

7. Capital Allocation

A disciplined, ROE-anchored framework. AXP runs a repeatable capital-return formula: grow the dividend toward a 20–25% target payout, hold CET1 inside ~10–11%, and sweep essentially all remaining excess capital into buybacks. In FY2025 it returned $7.6B (~71% of net income): $5.3B of buybacks + $2.3B of dividends (~75% on a trailing three-year basis). The dividend grew from $2.08 (FY2022) to $3.28 (FY2025) — up >80% since 2022 — and is guided +16% to $0.95/quarter ($3.80 annualized) for 2026. Buybacks retired ~7% of shares since 2022 (diluted shares ~736M FY2022 → ~696M average FY2025 → 682M on the April-2026 cover). (FACT, 10-K; Q4-2025 call)

The one soft spot — buybacks are mechanical, not opportunistic. AXP repurchased 16.8M shares at an average $312.87 in FY2025, and the monthly cadence shows it buying more aggressively into a rising, near-all-time-high stock ($358–366 in Q4-2025). Management’s process is to “return excess capital each quarter as it is generated,” not to flex repurchase with valuation. Through the Marathon Capital Returns lens this fails the counter-cyclical test — AXP buys steadily regardless of multiple. The partial defense is that (a) even at $360 the ~6–7% earnings yield keeps buybacks accretive, and (b) a bank holding company’s repurchase capacity is gated by CCAR/SCB and the CET1 floor, so true counter-cyclical hoarding is partly precluded by regulation. Capital-return discipline is high; buyback price discipline is mechanical — a genuine, if modest, knock. (FACT / INTERPRETATION)

M&A — correctly scaled tuck-ins, not empire-building. AXP’s deals are small, strategic, capability-oriented, and plug into the closed-loop value proposition: Center (2025, expense-management software to defend the SME flank against Ramp/Brex), Resy and Tock (dining), Rooster, Nipendo (B2B payments), and Kabbage (2020, SMB lending tech — later wound down via “KServicing,” the one historical M&A scar). It divests non-core: Accertify (2024) and the earlier Global Business Travel spinout (now the GBTG equity-method investee). The far larger “capital allocation” decision is the ~$6B/year of marketing plus rising technology spend run through the income statement — value-accretive on the evidence (ROE held ~34–35% through heavy reinvestment; the LTI plan’s relative ROE landed at the 91st percentile over 2023–2025), though AXP does not disclose cohort-level marketing ROI, so external validation rests on aggregate returns. The asset-growth-anomaly red flag (chasing balance-sheet growth at the cycle top) is not present — growth is funded by premium spend and fee economics, not reckless RWA expansion. (FACT / INTERPRETATION)

Incentive alignment — a growth-tilted bonus, but a properly anchored long-term plan. The 2026 proxy’s annual scorecard weights Shareholder 60% (within it: Revenue Growth 50%, EPS 25%, ROE 25%), Customer 10%, Colleague 10%, Strategic ~20%. The dominant single metric is therefore Revenue Growth at ~30% of the total scorecard vs. EPS and ROE at ~15% each — a growth tilt that partly explains the appetite for heavy reinvestment over near-term buyback. The long-term plan is the better-aligned anchor: awards are 80% performance RSUs vesting on three-year relative ROE (top-quartile pays up) with a relative-TSR ±20% modifier, and 20% options subject to positive cumulative net income; 2023–2025 NEOs vested at the 120% maximum (ROE 91st percentile, TSR 77th). CEO Stephen Squeri (Chairman & CEO since 2018, 40±year veteran) earned $46.2M in 2025 (heavily variable, mostly equity). Two yellow flags: say-on-pay support softened to 92.9% (below the mid-90s of a well-regarded board), and the 2026 design raised the maximum annual-incentive payout from 187.5% to 200% of target — mild pay-quantum drift worth watching. (FACT / INTERPRETATION)

The Berkshire dynamic — a structural endorsement. Berkshire Hathaway held 151,610,700 shares = 22.1% of AXP at 12/31/2025 (up from the long-quoted ~21%), having bought no shares since the 1990s — AXP’s buybacks shrink the denominator, so Berkshire’s stake compounds upward passively. A 1995 passivity/standstill agreement governs the relationship (Berkshire votes with the Board up to 17% while Squeri is CEO). This is exactly the per-share-value-focused behavior the Marathon framework rewards: buybacks quietly hand a permanent, aligned 22% owner an ever-larger slice. (FACT / INTERPRETATION)

Verdict (Capital Allocation): above-average for a large-cap financial — positive. A clear, repeatable, ROE-anchored framework that converts a mid-30s ROE into shareholder returns, correctly scaled tuck-in M&A, value-accretive reinvestment, a long-term plan anchored to relative ROE, and the structural Berkshire endorsement. Let down only by valuation-insensitive buybacks and a growth-weighted annual bonus. Not empire-building.

8. Changes and Headwinds — Last Two Years

Strategic and product changes (mostly thesis-strengthening).

• The 2025 premium refresh cycle. The U.S. Consumer and Small Business Platinum refresh — added benefits and a higher annual fee, phased in over FY2025–FY2026 on renewal anniversaries — lifted U.S. Consumer Platinum spend ~6 points (majority from tenured members) with retention intact, and underwrites the “card fees exit 2026 in the high teens” guide. The refresh playbook is now a repeatable lever across products and geographies. (FACT — strengthens thesis)
• The Gen-Z/millennial inflection. These cohorts became the largest share of U.S. consumer spend and the fastest-growing cohort, acquired on fee-paying premium products with superior credit — a structural improvement in the customer base and the growth runway. (FACT — strengthens thesis)
• The Center acquisition (2025) added expense-management software to defend the SME flank against Ramp/Brex/Capital One, expected to launch mid-2026. (FACT — defensive, modest positive)
• U.S. merchant acceptance reached ~parity (“170M+ merchant locations”), removing the historical reason a premium consumer would not carry Amex as a primary card. (FACT — strengthens thesis)

Regulatory, litigation, and leadership developments.

• Small-business sales-practices settlement (2025). AXP “entered into agreements to resolve governmental investigations related to historical sales practices for certain U.S. small business customers” (FY2025 10-K) — the resolution of a multi-year DOJ/OCC/Fed inquiry into legacy SME card-sales tactics. A contained, now-settled legacy matter, not a forward capital constraint, but a governance/compliance scar. (FACT — modest negative, resolved)
• Capital One–Discover closed (2025), creating a third closed-loop competitor — a multi-year erosion risk, not an acute one (see the relevant section). (FACT — long-term negative)
• Stress Capital Buffer regime change. A regulatory change moved the SCB effective date from October 1 to January 1; AXP’s SCB has historically sat near the 2.5% floor, reflecting its low-asset-risk model. (FACT — neutral)
• No CEO transition. Squeri remains Chairman & CEO; Christophe Le Caillec is CFO. Leadership is stable. (FACT — neutral/positive)

Headwinds carried into 2026.

• Credit normalization off a benign, sub-2019 base — the dominant cyclical variable.
• Realized discount-rate erosion (2.29%→2.24%) structurally caps discount-revenue growth below billed-business growth.
• Intense competition in premium consumer (“as tough as it’s ever been”) and software-led SME payments.
• A rich-vs-own-history multiple entering a late-cycle window (see the relevant section).

Verdict (Changes): on balance, thesis-strengthening. The premium refresh, the demographic inflection, and acceptance parity are durable positives that outweigh the contained legacy settlement and the still-distant Capital One–Discover threat. The genuine open headwind is macro/credit, not company-specific execution.

9. Risk Analysis

Overall risk posture. The risk profile is cyclical, not structural — the franchise is durable; the earnings it protects are pro-cyclical and currently benefit from benign credit. Risks 1 and 2 (credit normalization and multiple compression) are correlated and dominate the downside; they are the substance of the bear case. There is no realistic catastrophic-loss scenario short of systemic crisis.

10. Valuation Discussion (Embedded Expectations)

No price target and no recommendation. This section frames the price as embedded expectations and scenarios.

Where a hybrid belongs. AXP trades at 19.3x TTM EPS ($16.22), ~20.4x FY2025 EPS ($15.38), and ~17.6–18.1x forward (2026 guide $17.30–$17.90), with P/B of 6.32x on ~34% ROE. The peer bracket frames the placement (live multiples 2026-06-10; cross-checked to Visa, Mastercard and JPMorgan public filings):

AXP sits almost exactly at the midpoint between the ~28x network multiple and the ~7–13x lender/bank bracket — the correct conceptual location for a business that is structurally part toll-taker, part balance-sheet lender. The network/fee half (net card fees +18%, recurring, high-margin) deserves a premium in isolation; the lender half (on-balance-sheet credit risk, regulatory capital, cyclicality) deserves a discount. The market blends them at ~18x forward — a ~3.5–4 turn discount to the networks that is the credit-and-cyclicality penalty. (FACT / INTERPRETATION)

P/B-versus-ROE. On a Gordon-growth frame (warranted P/B = (ROE − g)/(CoE − g)), a 34% ROE at ~10% cost of equity and ~7–8% sustainable growth would justify a P/B above 10x — yet AXP trades at 6.3x. The market is deliberately not fully capitalizing the 34% ROE into book value, which is rational: it discounts that ROE as partly benign-credit-inflated and not fully durable through a cycle. 6.3x P/B is a “high ROE, but I don’t trust all of it” multiple. (INTERPRETATION)

The own-history tension. Despite the 19% pullback, AXP’s own-history valuation index sits at the 76th percentile composite (P/E 67th, P/B 86th, P/S 73rd) — rich versus its own ten-year history. This is the mirror image of Visa, which trades near the bottom decile of its own history. The networks have de-rated on regulatory fear; AXP has not. The market is paying up for AXP’s recent execution at the same moment it is paying down for the networks — an asymmetry that is itself a variant-perception flag. (FACT / INTERPRETATION)

Embedded expectations. At $313 and ~$17.60 midpoint forward EPS, the forward earnings yield is ~5.6%. Discounting an EPS stream growing ~10–12% for five years, fading to ~7% terminal, at a ~9.5% cost of equity with a ~16x exit multiple, lands near today’s price. The market is underwriting roughly “low-double-digit, decelerating-from-mid-teens” EPS growth — not the full mid-teens algorithm in perpetuity, and not network-quality durability. It is pricing 2026 EPS in the guided band, ~9–10% revenue compounding, the real fee annuity, and benign near-term credit correctly. What it may be mispricing — in either direction — is (a) whether mid-teens EPS is durable through a credit cycle, and (b) whether the fee annuity de-cyclicalizes earnings enough to re-rate the multiple up toward the networks. (INTERPRETATION / OPEN QUESTION)

Scenarios (3-year, to 2028). All start from FY2025 EPS $15.38 / 2026 midpoint ~$17.60. Implied price = 2028 EPS × applied multiple; IRR is ~3-year annualized price return from $313.34 (the ~1.2% dividend adds ~1pt/yr). (ASSUMPTION-driven throughout.)

The payoff is constructive in the base case but asymmetric in the tail. The bear combines two hits — EPS ~17% below the base path and ~5 turns of multiple compression — because peak-benign-credit earnings and a rich-vs-own-history multiple unwind together. The base case offers a respectable ~10%/yr with no heroic assumptions; the bull requires the market to re-rate a hybrid toward a network, which it has historically resisted. (INTERPRETATION)

Cyclicality overlay — peak earnings × peak multiple? AXP is a lender at a benign credit point (write-offs 2.3%, below 2019) trading at the 76th percentile of its own valuation history — the classic late-cycle double-risk. The counterweight is genuine structural de-cyclicalization: the fee annuity (+18%, recurring, spend-independent) and the premium/Gen-Z mix lower through-cycle earnings volatility versus a monoline. But this is a matter of degree, not a removal of credit risk — current EPS is modestly cyclically elevated, and the multiple prices little cyclical margin of safety. The fee annuity cushions the fall; it does not eliminate it. (INTERPRETATION / OPEN QUESTION)

11. Variant Perception

(a) Consensus belief. AXP is a high-quality, premium-brand spend-and-lend franchise executing exceptionally — record card fees, best-in-class credit, ~34% ROE, a credible mid-teens EPS algorithm, and a successful pivot to younger affluent cohorts. Consensus appears to treat the mid-teens algorithm as broadly durable and awards a midpoint hybrid multiple (~18x forward) — premium to banks/lenders, discount to networks. (INTERPRETATION)

(b) Strongest bull case. The fee annuity (+18%) plus the Gen-Z/millennial acquisition flywheel structurally lower cyclicality and lengthen the growth runway, making AXP’s earnings more network-like than the market credits. Closed-loop data is a genuine moat the open networks lack and an under-priced option in agentic commerce. If mid-teens EPS proves durable through a mild cycle, the multiple should re-rate toward the networks (low-20s), compounding price ~20%/yr.

© Strongest bear case. At its core AXP is a consumer-credit lender whose current EPS is inflated by sub-2019 write-offs and whose multiple sits at the 76th percentile of its own history. The affluent consumer is not immune to recession; a normalization to ~4%+ write-offs would stall EPS and trigger a re-rating toward the lender bracket — a 35%+ drawdown. The 19% pullback to date is insufficient given the late-cycle setup; “expensive vs. own history at peak credit” is a value trap.

(d) The assumptions that matter most. (1) Credit durability — does the write-off rate stay below ~3% through 2028 or spike to 4%+? (2) Fee-annuity de-cyclicalization — does +18% net-card-fee growth genuinely lower through-cycle earnings volatility, justifying a network-adjacent multiple? (3) EPS-algorithm durability — does mid-teens EPS hold, or revert to high-single-digits as provisions normalize and the base grows? (4) Multiple sustainability — can a 76th-percentile own-history multiple hold if growth decelerates? (5) Affluent-cohort recession resilience — untested at the current Gen-Z-heavy, fee-maximized mix.

(e) What would falsify each side. Falsifies the bull: write-offs break above ~3.5% with rising 30+ delinquency; net-card-fee growth decelerates toward single-digits; billed business turns negative YoY; EPS guidance cut below the mid-teens path. Falsifies the bear: AXP delivers through a credit scare with write-offs peaking below ~3.5%, fee revenue compounding double-digit through the stress, and EPS staying positive — proving the de-cyclicalization thesis and earning the network-adjacent multiple.

Where the analyst diverges from consensus. The genuinely under-appreciated variable is the fee annuity’s de-cyclicalizing effect — consensus models it as growth but likely under-weights how much a recurring +18% membership-fee stream stabilizes through-cycle earnings, which over time should compress AXP’s discount to the networks (pro-bull on a 5-year view). Simultaneously, consensus may under-weight the double-compression tail — the specific danger that peak-benign-credit EPS and a 76th-percentile multiple unwind together, making the drawdown deeper than the modest pullback suggests (pro-bear on a 1–2-year, late-cycle view). The defensible variant view: the business is de-risking structurally, but the current entry multiple prices little cyclical margin of safety — a great franchise where the timing/entry question dominates the quality question. The market is probably right on quality and possibly complacent on the credit-cycle tail.

12. Fact vs. Interpretation Table

13. Open Questions

1. Is current ~34% ROE / 2.3% write-off the normal for a premium book, or a benign-cycle peak? The entire valuation hinges on which — and there is no clean precedent for AXP’s current cohort/fee mix through a real recession.
2. How resilient is affluent discretionary spend (luxury, travel, front-cabin air) in a genuine downturn, given it drove FY2025 strength?
3. Could a final CCCA include three-party networks above a size threshold, eroding AXP’s regulatory shelter?
4. What is Capital One–Discover’s actual merchant-data/acceptance build-out trajectory — currently dismissed by management; independent verification needed before treating as benign.
5. What is the cohort-level marketing ROI/payback? AXP discloses only aggregate ROE, so marketing-spend accretion is inferred, not proven at the unit level.
6. Does the buyback pace need managing as Berkshire crosses 25% (bank-holding-company control thresholds / standstill terms)?
7. Exact 2025-cycle Stress Capital Buffer basis points (reconcile to the Fed’s mid-2025 SCB letter) and the resulting precise dividend/buyback capacity.
8. Form 4 completeness: bodies were not exhaustively mirrored; a definitive insider-purchase audit (--all-form4) would confirm the absence of any discretionary open-market buy.

14. What Must Be True

For the bull case to be right (the de-cyclicalization thesis):

• Net card fees keep compounding double-digit through a credit scare, and the fee annuity grows as a share of revenue — structurally lowering earnings volatility.
• Net write-offs peak below ~3.5% in the next downturn, proving the premium/Gen-Z book is genuinely more resilient than a monoline.
• Gen-Z/millennial cohorts mature into rising-spend, high-LTV, low-loss customers as they age, extending the growth runway a decade-plus.
• The market accepts this durability and re-rates the multiple toward the networks (low-20s).
• Falsification test: write-offs break above ~3.5% with rising 30+ delinquency, or net-card-fee growth decelerates to single-digits, or billed business turns negative YoY. Any one breaks the bull.

For the bear case to be right (the value-trap thesis):

• A consumer recession reaches the affluent cohort; billed business goes flat-to-negative and write-offs spike toward ~4%+.
• Provisions surge, stalling or shrinking EPS off a base that was flattered by sub-2019 credit.
• The multiple re-rates from the 76th percentile toward the lender bracket (~13–14x), compounding the EPS decline into a ~35% drawdown.
• Falsification test: AXP delivers a full cycle (or a genuine credit scare) with write-offs peaking below ~3.5%, fee revenue compounding double-digit through the stress, and EPS positive/growing. That proves the de-cyclicalization and breaks the bear.

The two cases share a single fulcrum: the behavior of credit and the fee annuity through the next downturn. Everything else — the moat, the growth, the capital allocation — is already largely settled in the bull’s favor. What is unsettled is whether American Express has genuinely become a more network-like, less cyclical business, or whether it remains a superb consumer lender whose current earnings and multiple both sit near a cyclical high.

15. Source Appendix

A full source appendix is maintained separately at AXP_source_appendix.md (Appendix B of the combined report). Primary sources: American Express FY2025 Form 10-K (filed 2026-02-06) and FY2024 10-K; the Q1-2026 (2026-04-24), Q4-2025 (2026-01-30), Q3-2025, and Q2-2025 earnings-call transcripts; the 2026 DEF 14A proxy; the 8-K and Form 3/4/5 corpus 2021–2026; SEC EDGAR XBRL (CIK 0000004962); conference presentations (Bernstein 2026-05-28, Goldman/UBS/KBW/Barclays 2025–2026); and public filings of Visa (2026-06-09), Mastercard (2026-06-10), and JPMorgan (2026-06-09) for peer framing. Live market data via yfinance and the AZI fundamentals feed (2026-06-10). All non-obvious facts are cited inline by filing section or transcript date.

APPENDIX A — Standard Diligence Questionnaire

American Express Company (NYSE: AXP) — as of 2026-06-11

Supplemental to the research memo. Answers grounded in the FY2025 10-K, 2026 proxy, earnings transcripts, and EDGAR XBRL. Fact / Interpretation / Assumption labeled where it matters.

General

What thoughtful questions have other investors asked about this company? The recurring debates: (1) Is AXP a “network” deserving a Visa-like multiple or a “lender” deserving a bank multiple? (2) Is current EPS at a benign-credit cyclical peak? (3) Can the premium positioning survive a recession in affluent discretionary spend? (4) Does the Gen-Z/millennial pivot structurally lower cyclicality? (5) Will Capital One–Discover erode the closed-loop advantage? (6) Are buybacks at all-time highs value-destructive? Each is addressed in the memo (the relevant section, the relevant section, the relevant section, the relevant section).

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? (Interpretation) Modestly cyclically elevated — net write-offs (2.3%) sit below 2019 levels, so provisions are sub-normal and flatter current EPS. Not an extreme peak, but the credit tailwind is unlikely to persist indefinitely. Driven by the external environment or internal actions? Both — internal (premium refresh, Gen-Z acquisition, fee increases, international expansion) drives the durable growth; external (benign credit, resilient affluent spend, rate environment) flatters the current level. How stable are revenues? ~76% (discount + NII) is spend/balance-driven and pro-cyclical; ~14% (net card fees) is a recurring, recession-resilient annuity growing +18%. Increasingly stable at the margin, but the body remains cyclical. Outlook for products/services? (Fact) Management guides FY2026 revenue +9–10%, EPS $17.30–$17.90; long-term “10%+ revenue / mid-teens EPS.” How big is the market — growing, shrinking, domestic or international? (Fact) Global electronic payments grows ~8%/yr (cash-to-card migration); AXP is global with international (ICS) its fastest-growing segment (double-digit FX-adjusted for 20 straight quarters), so the runway is large and growing, weighted internationally for incremental growth.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More — premium consumer “as tough as it’s ever been”; SME contested by Ramp/Brex/Capital One; Capital One–Discover adds a third closed loop. How profitable is the business (ROIC, ROE)? (Fact) ROE 33.9% FY2025; ROTCE ~40%+ on near-all-tangible equity (goodwill only $4.9B). Elite. How profitable is the industry — competitors, barriers to entry? High returns industry; barriers are high (network scale, brand, data, regulatory licensing, capital). AXP’s three-party model is the most regulation-sheltered niche. Can the business be easily understood? Mostly — a closed-loop card issuer/network/lender. The complexity is the unusual income statement (revenue net of interest expense; VCE vs. OpEx split) and the credit/reserve dynamics. Can it be undermined by foreign low-cost labor? No — it is a network/brand/data/balance-sheet business, not labor-arbitrage-exposed. Do brands matter? Decisively yes — the Amex brand is core to pricing power (fee/card $92→$117 with retention intact). Nature of competition? Rewards richness, brand/status, acceptance breadth, merchant economics, ecosystem (lounges/Resy/FHR), and underwriting. Customers’ switching costs? (Interpretation) Moderate — behavioral/loyalty (rewards balances, status, ecosystem) plus sunk annual fee, not contractual lock-in. Proven by retention through fee hikes.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? (Interpretation) Yes — the brand and the customer-acquisition asset built by ~$6B/yr of fully expensed marketing; reported earnings understate economic value. Off-balance-sheet liabilities? Securitization VIEs are largely consolidated; rewards liabilities are accrued (Membership Rewards). No alarming off-balance-sheet items identified; merchant-litigation contingencies disclosed. How conservative is the accounting? Conservative — marketing/acquisition expensed not capitalized; CECL reserves built ahead; OCF > NI. The realized discount-rate disclosure is transparent. How CapEx-hungry is the business? Low physical CapEx; the “investment” is marketing + technology run through the income statement, plus regulatory capital held against loans.

Capital Allocation & Management

How much FCF, and how is it used? (Fact) OCF $18.4B FY2025; capital return $7.6B (~71% of NI): $5.3B buybacks + $2.3B dividends, the rest funding loan/balance-sheet growth. Philosophy: 20–25% dividend payout, CET1 ~10.5%, sweep residual to buybacks. Significant acquisitions recently? Tuck-ins only — Center (2025, expense-mgmt SW), Resy/Tock, Nipendo. Divested Accertify (2024). No scale deals. Buying back shares? Yes — ~7% of shares retired since 2022; FY2025 16.8M shares at avg $312.87 (mechanical, not price-disciplined — buying into all-time highs). Issuing large amounts of stock to insiders? No — net share count is falling; equity comp is a normal portion of NEO pay. Compensation policy of directors/management? (Fact) CEO Squeri $46.2M (2025), heavily variable/equity. Annual scorecard growth-tilted (revenue 30% of total); LTI properly anchored to relative ROE + TSR. Say-on-pay 92.9% (soft); max AIA raised to 200%. Motivations of management? Long-term, growth-oriented, ROE-conscious; the Berkshire 22% anchor reinforces per-share-value alignment. Mild growth-tilt in the bonus is the watch item.

Valuation & Market Data

ADR, MLP, or K-1 issuer? No — U.S. domestic C-corp, NYSE-listed common; standard 1099 dividend. Dividend policy? (Fact) 20–25% payout; DPS $3.28 (2025), guided +16% to $3.80 (2026); ~1.2% yield. Buybacks are the larger return vehicle. How profitable is the business? Among the most profitable in financials — ~34% ROE. Is net income diverging from cash from operations? No — OCF ($18.4B) comfortably exceeds NI ($10.8B); cash-backed earnings.

Risks & Downside

What would cause the stock to decline? Credit normalization/recession hitting affluent spend (the dominant risk); multiple de-rating from a rich own-history level; competitive/regulatory shocks. See the relevant section risk matrix. Risk of catastrophic loss? Very low — deposit-funded, 10.5% CET1, ~34% ROE, 22% Berkshire anchor; only a systemic crisis or major fraud/cyber event qualifies. Chance of total loss? Negligible for a $214B, well-capitalized, profitable franchise.

Recent News & Events

Has the business environment changed recently? (Fact) 2025 premium Platinum refresh; Gen-Z/millennial became the largest share of U.S. consumer spend; U.S. acceptance reached ~parity; small-business sales-practices investigations settled (2025); Capital One–Discover closed (2025). (The AZI news feed returned no scored articles — routine for a mega-cap; timeline built from 8-Ks + transcripts.) Significant acquisitions? Center (2025). Change in accounting policies? None material; SCB effective-date regulatory change (Oct 1 → Jan 1). Recent changes — markets, facilities, management? International expansion ongoing; leadership stable (Squeri CEO, Le Caillec CFO).

APPENDIX B — Source Appendix

American Express Company (NYSE: AXP) — as of 2026-06-11

Primary sources (SEC filings — EDGAR CIK 0000004962)

• Form 10-K for FY2025 (filed 2026-02-06; axp-20251231.htm) — Business; Risk Factors; MD&A (Consolidated and Segment Results of Operations); Selected Statistical Information (discount rate, billed business, cards-in-force, average spend, net write-off/delinquency rates, net interest yield, CET1); financial statements and credit/reserve footnotes; capital strategy; legal proceedings (small-business sales-practices settlement). https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000004962
• Form 10-K for FY2024 (filed 2025-02-07; axp-20241231.htm) — trend comparatives; Accertify gain disclosure.
• Form 10-Q for Q1-2026 — interim results, credit metrics, capital.
• DEF 14A proxy (2026) — executive compensation scorecard and metric weights; LTIA design (relative ROE/TSR); CEO pay; say-on-pay; Berkshire ownership (151,610,700 shares / 22.1%) and the 1995 passivity agreement.
• 8-K corpus 2021–2026 — buyback authorization (Mar-2023, 120M shares; ~58.3M remaining at 12/31/2025), dividend actions, Stress Capital Buffer, master-trust charge-off/delinquency data, debt issuance.
• Form 3/4/5 corpus 2021–2026 — insider transactions (NEO 10b5-1 exercise-and-sell pattern; no discretionary open-market purchases in the sampled set; director deferred-comp accruals).
• SEC EDGAR XBRL (data.sec.gov, CIK 0000004962) — net income, diluted EPS, stockholders’ equity, dividends declared, share repurchases, noninterest expense, shares outstanding. Accessed 2026-06-11.

Earnings-call and event transcripts

• Q1-2026 earnings call (2026-04-24) — spend +10%, revenue +11%, EPS $4.28 (+18%); guidance reaffirmed; credit, fee-annuity, Gen-Z cohort, Platinum-refresh commentary.
• Q4-2025 earnings call (2026-01-30) — FY2025 results; FY2026 guidance (revenue +9–10%, EPS $17.30–$17.90); VCE/OpEx ratios; capital return; card-fee streak.
• Q3-2025 (2025-10-17), Q2-2025 (2025-07-18) earnings calls.
• Conference presentations: Bernstein 42nd Strategic Decisions (2026-05-28; closed-loop/fraud/acceptance/data framing), UBS (2026-02-10), Goldman Sachs (2025-12-10), KBW (2025-11-12), Barclays (2025-09-09), Morgan Stanley (2025-06-11).

Market & third-party data (color; reconciled to filings)

• yfinance (2026-06-10) — price $313.34, market cap, shares, 52-week range. Unofficial; reconciled to the 10-Q cover and XBRL.
• AZI fundamentals feed (2026-06-10) — snapshot, ownership/short interest, own-history valuation index (P/E 67th, P/B 86th, P/S 73rd, composite 76th percentile). Third-party aggregated; valuation percentiles are own-history only.
• AZI news feed — returned no scored articles for AXP (routine mega-cap pattern); recent-events timeline built from 8-Ks and transcripts.

Peer cross-reads (public filings — attributed as prior internal work)

• Visa (V) public filings — open-loop four-party economics, ~0.24% net take-rate, network multiple (~28x trailing, bottom-decile own history).
• Mastercard (MA) public filings — network comp.
• JPMorgan (JPM) public filings — money-center bank comp (~14x).
• Card-lender multiples (SYF, COF) via live market data for the lender-bracket comparison.

Analytical frameworks

• Greenwald & Kahn, Competition Demystified — moat taxonomy (customer captivity + economies of scale; the spend-flywheel), barriers-to-entry and ROIC tests.
• Marathon / Chancellor, Capital Returns — supply-side capital-cycle lens on capital allocation (buyback price-discipline test; asset-growth-anomaly check).

All non-obvious facts in the memo are cited inline by filing section or transcript date. Management commentary is treated as a hypothesis validated against filings, financials, and external evidence, not as evidence in itself.