American Express AXP

Reference table

1. The quarter, and what the headline hides

From the Form 10-Q filed 2026-07-24, Table 1:

Three observations, and the second and third are the ones that matter.

First, the revenue growth is genuine and currency-neutral. Total revenues net of interest expense grew 10% and 10% on an FX-adjusted basis — the growth is not a translation effect. First-half growth was 11%.

Second, roughly 62% of the pretax increase came from a credit release. Pretax income rose $521 million; provisions for credit losses fell $321 million, from $1.405 billion to $1.084 billion. That is 61.6% of the improvement, and it is a change in credit conditions rather than in the business. Expenses grew 12% against revenue growth of 10%, so the operating line was actually a modest negative. We flag this prominently because the headline "pretax income +15%" is materially better than the underlying operating performance.

Third, net income grew only 8% — BELOW the 10% revenue growth — because the tax rate rose. The income tax provision rose 45% on a 15% pretax increase, taking the effective rate from 18.7% to 23.6%. The 11% earnings-per-share growth reported in the earnings calendar is therefore net income growth of 8% plus a share count that fell roughly 3%.

The revenue composition, and this is where the quality is:

Net card fees are the best line in this company and they grew 15%. The 10-Q attributes it "primarily driven by growth in our premium card portfolios." At $2.862 billion a quarter — $11.4 billion annualised — this is a subscription business embedded inside a payments network. It arrives whether or not the cardholder spends, it is repriced upward periodically without visible attrition, and it has grown faster than billed business for several years. If one number distinguishes American Express from every other payments company, this is it.

And the one line that is deteriorating, disclosed plainly:

> "Discount revenue increased for both the three and six month periods, driven by increases in billed business of 9 percent and 10 percent, respectively, partially offset by lower average merchant discount rates primarily due to shifts in spend mix by geography and merchant categories."

Billed business grew 9% and discount revenue grew 9% — so the average discount rate was roughly flat to down. This is the slow-motion version of the structural constraint the knowledge base identifies: a closed loop charging higher merchant fees can only grow by pushing into merchant categories and geographies that will not pay the premium rate.

Segment and geographic mix, from the vendor blocks, which tie:

The geographic block ties to reported net revenue exactly. International Card Services grew 13.4%, faster than the company's 10.5% — the only structural diversification available against a 77.6% United States concentration.

2. The record — and the only underperformance in this batch

Four beats, one in-line, none large. American Express is the most predictable business in this batch — the beats average +3.6% against Goldman Sachs's 19.1% and Dell's swings from −8.8% to +64.2% — and that predictability is exactly what a 21x multiple is normally paying for.

Annual context:

Four consecutive years of 9-10% revenue growth and a diluted share count down 9.8% from 752 million to 678 million. Diluted EPS has compounded 18.9% a year from $9.84 to $16.53 — roughly half of it revenue, roughly a quarter margin, roughly a quarter buyback.

And now the market's verdict on all that:

American Express is the ONLY name in this batch of twelve that underperformed the index over twelve months, and the six-month gap is the widest negative reading in the batch by a factor of four. The knowledge base supplies the reason, and it is specific enough to be testable — see Section 5.

3. Balance sheet, capital returns, and two corrections

From bal_a at 2025-12-31:

Correction 1 — the vendor reports book value and TANGIBLE book value per share as identical, ignoring $4.873 billion of goodwill. bookValuePerShareTTM and tangibleBookValuePerShareTTM both read $50.56. Removing goodwill of $4.873 billion and intangibles of $0.090 billion from equity of $33.474 billion and dividing by 675.4 million shares gives tangible book value per share of approximately $42.34. Price to tangible book is therefore approximately 8.2x, not the vendor's 6.86x — a 19% understatement. This is the same error the vendor makes on all three banks in this batch and it is systematic.

Correction 2 — enterprise value is WITHHELD. enterpriseValueTTM of $248.037 billion less market capitalisation of $234.231 billion implies net debt of $13.806 billion; the corrected balance-sheet figure is $9.225 billion, a 49.6% residual — well above the 15% threshold. Separately, enterprise value is inapplicable: American Express's $57.759 billion of debt and a large part of its $170.806 billion of current liabilities are funding for a $221.053 billion card-member receivable and loan book, not acquisition leverage. evToSalesTTM (2.95x), evToEBITDATTM (10.03x) and netDebtToEBITDATTM (0.56x) are rejected. No enterprise-value figure appears anywhere in this dive.

Correction 3 — several ratios are lender artefacts and are not used. daysOfSalesOutstandingTTM of 969.5 days and a cashConversionCycleTTM of 541.1 days are measuring a loan book, not working capital. financialLeverageRatioTTM of 8.99x and debtToEquityRatioTTM of 1.72x are what a consumer lender is. The vendor's debtToEquityScore of 1 out of 5 is a lens artefact and is rejected; priceToBookScore of 1 at 6.86x book — 8.2x corrected — is a correct signal and is accepted.

Capital returns. Fiscal 2025 repurchases of $5.814 billion are 2.48% of market capitalisation, and the diluted share count has fallen from 752 million (FY2022) to 678 million — 9.8%. The dividend is $3.54 trailing, a 1.02% yield, at a 21.1% payout ratio. Total shareholder yield is approximately 3.5% — modest by the standards of the financials in this batch (Citigroup 8.5%, Goldman Sachs 7.0%), and it reflects a company retaining capital to fund loan growth rather than one with excess capital.

Free cash flow, with a caveat. FY2025 operating cash flow of $18.428 billion less capital expenditure of $2.425 billion gives $16.003 billion, a 6.4% yield on freeCashFlowYieldTTM. For a lender this figure is distorted — loan growth appears in investing rather than operating activities, so "free cash flow" overstates distributable cash considerably. We report it and do not use it in the valuation.

4. Valuation — priced in or room?

At $346.85 (market cap $234.23B, ~675.4M shares; no enterprise value quoted):

Estimate coverage is strong — 18 analysts on FY2026 and FY2027 revenue and 21 on both EPS lines, 14 on FY2028. The FY2029 row carries 8-9 analysts and is used only as a cross-check.

Per the data contract we tested the estimate block and found ONE broken row. netIncomeAvg is below ebitAvg in FY2026, FY2027, FY2028 and FY2029 — all consistent. But the FY2030 row is internally impossible: revenue FALLS from $103.981 billion (FY2029) to $76.423 billion — a 26.5% decline — while EPS RISES from $27.176 to $29.32, and netIncomeAvg of $20.407 billion EXCEEDS ebitAvg of $13.973 billion. A row showing revenue down 26% and earnings up 8% on 13 analysts is not a forecast. The FY2030 row is DISCARDED in full and no conclusion touches it.

One useful check the FY2026 row passes: first-half diluted EPS was $8.81 ($4.28 + $4.53), and the FY2026 consensus of $17.670 implies a second half of $8.86 — essentially flat with the first. Unlike almost every other name in this batch, the American Express consensus is neither above nor materially below the run-rate. It is simply in line, which is what a predictable business produces.

Peer context. The vendor peer set is a mixture — Capital One ($136.13B), Ally Financial ($13.69B), Mastercard ($571.09 per share) and Goldman Sachs are relevant in different ways; Caterpillar is not a comparable and its inclusion is a sector-code artefact. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. The structurally relevant comparison — and the one the knowledge base makes — is against the open-loop networks: American Express is a closed loop with roughly 10% share of an affluent niche, against the open-loop utilities whose economics are pure toll and whose multiples are correspondingly higher.

4a. What today's price assumes (the inversion)

At $346.85 — 17.2x FY2027 consensus and roughly 8.2x tangible book — the price embeds:

• EPS reaches $20.116 in FY2027 and $22.992 in FY2028. (Consensus; 21 and 14 analysts.) That is 13.8% and 14.3% growth on revenue growing 9.0% and 8.5%. Achievable on the record — diluted EPS has compounded 18.9% a year since FY2022 — and it requires operating leverage plus a continuing buyback.
• Provisions for credit losses do not normalise. (Filing-confirmed that they FELL 23% year on year and supplied ~62% of the pretax increase.) This is the most fragile assumption in the price. A return to the 2025 provision run-rate removes roughly $1.3 billion of annual pretax income — about 8% of pretax.
• Net card fee growth stays in the mid-teens. (Filing-confirmed at +15%.) At $11.4 billion annualised and 14.6% of net revenue, this line is the recurring core. If it decelerates toward the 9% billed-business rate, the quality argument for a 21x multiple weakens materially.
• The premium consumer holds. (Not testable from any filing; identified by the knowledge base as the specific fear.) 77.6% of revenue is United States and the cardholder base is affluent and small-business. The knowledge base records a 25% drawdown in March 2026 on artificial-intelligence white-collar-displacement fears. The June quarter shows no damage; six months of relative underperformance shows the market has not been persuaded.
• The market keeps paying 17-21x forward earnings for it. (Our number.) At 13x FY2027E the stock is $262; at 22x it is $443. The multiple band is worth $181 — 52% of the current price.

4b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+14.3%, from FY2027E $20.116 to FY2028E $22.992) + multiple drift (COMPRESSION, from 17.2x to roughly 16x, −7.0%) + shareholder yield (+3.5%) ≈ +10% to +11%.

Almost all of the expected return is earnings growth, which is the correct shape for a compounder and the reason a Hold here is a comfortable position rather than an uncomfortable one. Our base of $370 assumes mild multiple compression on the ordinary ground that consensus revenue growth decelerates from 10.1% to 8.5% across the forecast period.

Sensitivity, stated openly, on the FY2027 consensus EPS of $20.116: at 13x the shares are $262; at 15x, $302; at 18.4x, $370; at 20x, $402; at 22.9x, $460. The street's $378.09 implies 18.8x.

The bull case at $460 requires modest multiple expansion to 22.9x. The bear case at $265 requires no earnings miss at all — only a de-rating to 13x, which is roughly where a consumer lender with deteriorating credit would trade.

4c. Variant perception (where we differ, what would surprise)

• We differ from the street only marginally, and the street itself is at Hold. Consensus target $378.09 against our $370 — a 2.1% gap — on a distribution of 23 buy, 30 hold and 4 sell across 57 analysts, with a consensus rating of Hold. This is one of only two "Hold" consensus ratings in this batch (the other is Goldman Sachs), and we agree with it. We claim no directional edge.
• Where we do add something is the earnings-quality decomposition of the June quarter, and we have not seen it stated. Pretax income rose 15% while provisions for credit losses fell 23% — roughly 62% of the increase. Net income rose only 8%, BELOW revenue growth of 10%, because the effective tax rate rose from 18.7% to 23.6%. The reported 11% earnings-per-share growth is net income growth of 8% plus a 3% share-count reduction. The headline "pretax income +15%" is materially better than the operating reality, and expenses grew 12% against revenue growth of 10%. Watchable number: the provision line in the 2026-10-23 release. A return toward $1.4 billion removes most of the growth.
• We think the twelve-month underperformance is the most informative fact in this file and it is not an accident. This is the only one of twelve names to lag the index over twelve months, and the six-month gap is −12.8 points. The knowledge base identifies the cause precisely — artificial-intelligence displacement of white-collar employment hitting exactly this cardholder base — and four claims dated March 2026 call the resulting 25% drawdown an over-reaction. The June quarter supports them: billed business +9%, net card fees +15%, provisions falling. But the shares have not recovered, and the 50-day moving average remains below the 200-day. Either the market is wrong and this is the "fat pitch" the lane describes, or the market is early. We do not know which, and at 21x trailing earnings we are not paid to guess.
• We surface the one non-bullish claim in a lane of eleven bulls, because it is the structural bear case and nobody else in the lane makes it: that the closed-loop model "caps scale and forces higher merchant fees, limiting market share (~10%) to affluent niche versus open-loop utility." The 10-Q's disclosure that average merchant discount rates fell on "shifts in spend mix by geography and merchant categories" is the first quantitative evidence of that constraint operating.
• Positive surprise that would force a re-rate: net card fee growth above 15% with billed business accelerating above 10%; the 50-day moving average crossing back above the 200-day on a strong October print; International Card Services growth above 15%, reducing the United States concentration; or credit metrics improving further while spending accelerates, which would falsify the white-collar thesis outright.
• Negative surprise that would break it: provisions for credit losses returning above $1.4 billion in a quarter; billed business growth falling below 6%; net card fee growth decelerating below 10%; any disclosure of weakening spend among affluent or small-business cardholders; or the average merchant discount rate falling materially further.

Synthos fair values

All three anchors are multiples of the FY2027 consensus EPS of $20.116 (21 analysts — the best-covered forward year), cross-checked against book value and FY2028E.

• Bear ~$265 — 13.2x FY2027E. Cross-checks: 15.0x FY2026E; 11.5x FY2028E; 5.34x book and roughly 6.3x corrected tangible book; 9.3% below the 52-week low of $292.27. The scenario: credit provisions normalise and remove roughly $1.3 billion of pretax income, white-collar employment softens and billed business decelerates, net card fee growth falls toward single digits, and the multiple compresses to where a consumer lender trades. −23.6%.
• Base ~$370 — 18.4x FY2027E. Cross-checks: 20.9x FY2026E; 16.1x FY2028E; 7.32x book and roughly 8.7x corrected tangible book; 1.5% below the 52-week high of $384.89. Sensitivity: 15x gives $302, 20x gives $402, and the street's $378.09 implies 18.8x. The scenario: revenue keeps compounding at 9-10%, net card fees hold in the mid-teens, credit normalises gradually rather than abruptly, the buyback continues at 2.5% a year, and the multiple compresses modestly as growth decelerates. +6.7%, plus 3.5% of shareholder yield.
• Bull ~$460 — 22.9x FY2027E. Cross-check: 20.0x FY2028E; 9.1x book. This bull case requires modest multiple expansion, and its justification is the one thing that would produce it: the white-collar fear being decisively falsified. The scenario: billed business re-accelerates above 10%, net card fees hold at 15%, International Card Services broadens the base, and the market re-rates a 34%-return-on-equity compounder back toward where it traded before March 2026. +32.6%.

Base is 6.7% above spot; asymmetry roughly 1.38:1 to the upside (23.6% down, 32.6% up). That is an ordinary ratio on an ordinary setup. What makes it a Hold rather than a Watch is that the compounding is genuine and reliable — four consecutive years of 9-10% revenue growth and 18.9% compound EPS growth — so a holder is being carried by the business rather than waiting for a re-rating. What prevents a Buy is that the earnings quality in the most recent quarter was weaker than the headline, our fair value sits below the street's, and the street's own consensus rating is Hold.

5. Knowledge base — eleven bulls, no bears, and a gold miner

Raw entity hits: 12. Text matches: 5, of which THREE discarded as collisions. Independent voice-days: approximately 5.

The search covered AXP, American Express and Amex across the entity field of all 51,928 distilled claims, plus a text sweep on american express|\bamex\b.

The collisions first, because "Amex" is a genuinely dangerous token. Three of the five text matches are discarded:

• 2026-06-26 — a mining-sector relative-valuation claim citing "~US$150 EV/oz vs Snowline $180, New Found Gold $226, Amex $217". That is Amex Exploration, a gold-mining company. It has nothing to do with American Express. A hard collision, and the clearest example in this batch after Merck KGaA.
• 2025-06-11 — a claim describing a crypto staking platform as "the 'Amex of crypto'". A metaphor, not a claim about the security.
• 2026-05-11 — a claim about a private-equity take-private of American Express Global Business Travel for $6.3 billion. GBTG is a separately listed company in which American Express holds a minority interest; the claim is about GBTG, not AXP.

All three discarded. The remaining two text hits are a 2012 mobile-payments principle claim and a 2026 insurance-sector claim mentioning American Express only as a Berkshire holding; neither is used.

The entity lane: 11 bullish, 1 neutral, 0 bearish. On a stock that has underperformed the index for twelve months and is the only such name in this batch, a lane with no bearish claims at all should itself provoke suspicion, and we treat it accordingly.

Four of the twelve — a third of the lane — are one argument restated across two days by one channel:

> 2026-03-03 · bullish · conviction 55 and 60 · channel compound_and_friends (two claims)

> "Down 20% on white-collar AI-unemployment fears; secular winner for the upper-K got oversold."

> 2026-03-31 · bullish · conviction 70 and 75 · channel compound_and_friends (two claims)

> "Amex down 25% on AI white-collar-displacement fears with no news; premium consumer (mirrored by Delta's healthy bookings) remains strong... a fat pitch worth buying."

Counted as two voice-days. But the substance is the most useful thing in the lane, because it names the cause of the underperformance in this dive's central technical finding. A card business whose revenue is premium consumer and small-business spending, 77.6% concentrated in the United States and skewed to affluent white-collar cardholders, is a direct expression of white-collar employment. The claims argue the 25% drawdown was an over-reaction and cite airline booking data as corroboration.

The only named speaker in the lane makes the same call with a stated falsifier:

> 2026-03-17 · bullish · conviction 50 · thesis · speaker Josh Brown · speaker_role independent

> "American Express, premier card for upper-K luxury shoppers, is in a 20% drawdown — probably a great buying opportunity; won't fall 30% without a genuine softening in [employment]."

"Won't fall 30% without a genuine softening in employment" is a testable conditional and it is the right frame. The shares bottomed at $292.27 and are now $346.85 — up 18.6% from the low — so the call was directionally right. But twelve-month performance is still +17.84% against SPY's +24.26%, so the recovery has not closed the gap.

And the single non-bullish claim, which is the structural bear case and the most analytically valuable item in the lane:

> 2026-02-12 · NEUTRAL · conviction 55 · principle · channel we_study_billionaires

> "Amex's closed-loop model caps scale and forces higher merchant fees, limiting market share (~10%) to affluent niche versus open-loop utility of Visa/MA."

This is corroborated in the 10-Q from an unexpected direction: "lower average merchant discount rates primarily due to shifts in spend mix by geography and merchant categories." A closed loop charging premium merchant fees can only grow billed business by moving into categories and geographies that will not pay the premium rate — which is exactly what a falling average discount rate looks like.

The remaining claims are principle-level and mostly historical: four from we_study_billionaires describing the classic Buffett purchase during a temporary scandal, and two from invest_like_the_best — one arguing that "vertical integration (customer, merchant and bank in one) delivers a superior, lower-friction customer experience", and one, notably, that is about a private business-to-business competitor winning corporate customers from incumbents on user experience. That last one is a competitive warning tagged to American Express and we record it as such.

Conclusion. The knowledge base has a positively-signed, low-breadth and structurally one-sided view of American Express: eleven bulls, no bears, five independent voice-days, one named speaker, and one neutral claim that carries the entire structural bear case. Its genuine contribution is identifying the specific mechanism behind the only twelve-month underperformance in this batch — artificial-intelligence displacement of the white-collar consumer — and calling it an over-reaction. The June quarter supports them; the six-month share price does not yet. We use it in Section 4c and we do not let a zero-bear lane move the fair value on a stock the market has been marking down for six months.

6. Data integrity — what we rejected from the vendor file and why

Nine findings.

1. bookValuePerShareTTM and tangibleBookValuePerShareTTM are IDENTICAL at $50.56, ignoring $4.873 billion of goodwill — CORRECTED. Removing goodwill of $4.873 billion and intangibles of $0.090 billion from equity of $33.474 billion and dividing by ~675.4 million shares gives tangible book value per share of approximately $42.34. Price to tangible book is therefore roughly 8.2x, not the vendor's 6.86x — a 19% understatement. The identical error appears on Morgan Stanley, Goldman Sachs and Citigroup in this batch; it is systematic across financial companies in this vendor file.

2. enterpriseValueTTM carries a 49.6% residual and is conceptually inapplicable — WITHHELD. The vendor's $248.037 billion less market capitalisation of $234.231 billion implies net debt of $13.806 billion; the corrected balance-sheet figure is $9.225 billion. Separately, American Express's $57.759 billion of debt and much of its $170.806 billion of current liabilities fund a $221.053 billion card-member receivable and loan book. evToSalesTTM, evToEBITDATTM and netDebtToEBITDATTM all rejected.

3. netDebt omits $826 million of short-term investments — corrected. The vendor reports $10.051 billion; the correct figure is $9.225 billion. Small in absolute terms and the same mechanism that produced an $8.78 billion error on Arista and an $8.2 billion error on Palantir in this batch.

4. The quarterly operatingIncome series is internally impossible — REJECTED. The June 2026 row reads operatingIncome: $9,338M against incomeBeforeTax: $4,071M — a $5.27 billion gap that interest expense of $1.958 billion does not explain. The series reads $3,550M, $3,825M, $3,090M, $6,598M, $9,338M across five quarters on revenue that moved 10%. operatingIncome, grossProfit, operatingProfitMarginTTM (27.2%) and grossProfitMarginTTM (84.3%) are all rejected. We use the 10-Q's Table 1 — revenues net of interest expense, provisions, expenses, pretax income and net income — throughout.

5. The quarterly revenue field is GROSS while the reported metric is NET — basis established. The June 2026 row reads $21.595 billion; the 10-Q and the earnings calendar both report $19.637 billion of "total revenues net of interest expense", and the difference is exactly the $1.958 billion of interest expense. Every revenue figure in this dive is net of interest expense. priceToSalesRatioTTM of 2.79x, struck on the gross figure, is rejected.

6. The FY2030 estimate row is internally impossible — DISCARDED. Revenue FALLS from $103.981 billion (FY2029) to $76.423 billion — a 26.5% decline — while EPS RISES from $27.176 to $29.32, and netIncomeAvg of $20.407 billion EXCEEDS ebitAvg of $13.973 billion. A row showing revenue down 26% and earnings up 8%, on 13 revenue analysts, is not a forecast. The FY2026 through FY2029 rows all pass the arithmetic screens and are used.

7. Several ratios are lender artefacts and are not used. daysOfSalesOutstandingTTM of 969.5 days, cashConversionCycleTTM of 541.1 days, receivablesTurnoverTTM of 0.376 and operatingCycleTTM of 969.5 days are all measuring a card-member loan book rather than working capital. financialLeverageRatioTTM of 8.99x and debtToEquityRatioTTM of 1.72x are what a consumer lender is.

8. The vendor composite rating of B / 3 is half artefact — partially accepted. debtToEquityScore of 1 reflects the lender's funding structure and is rejected as a lens. priceToBookScore of 1 at 6.86x book — approximately 8.2x on the corrected tangible figure — is a CORRECT signal and we agree with it. returnOnEquityScore of 5 at 34.1% is correct; discountedCashFlowScore of 4 rests on a free-cash-flow figure that is distorted for a lender and is not used.

9. quote.yearHigh/yearLow disagree with tech.hi52/lo52 — we use tech. The quote block reports $387.49 / $290.63; the computed block reports $384.89 / $292.27. Discrepancies of 0.7% and 0.6%. We use tech throughout for consistency with the moving averages and drawdown figures.

Not defects, correctly reported and independently confirmed: the seg_geo block ties to reported net revenue exactly for FY2025 ($56.015B + $7.073B + $5.218B + $4.194B − $0.271B = $72.229B); the seg_prod block reconciles with a small residual; the gross-to-net revenue identity holds exactly against the 10-Q; the implied share count of ~675.4 million is consistent with the 678 million diluted weighted average with no share-class complication; and dividendPayoutRatioTTM of 21.1% is consistent with the dividend and earnings figures.

Non-equity tripwire — checked and passed. AXP is common stock, NYSE-listed, one class. Beta 1.045; a 52-week range of $292.27 to $384.89 — a 31.7% spread, the NARROWEST in this batch, consistent with the most predictable earnings stream here; a variable, rising dividend; volume of 1.71M shares (roughly $594M of turnover, the lightest in this batch). This is common equity.

7. Technicals

• Price $346.85. −9.91% from the 52-week high of $384.89; +18.64% above the 52-week low of $292.27. The 31.7% annual range is the narrowest in this batch. Position within the annual range: 59th percentile.
• Moving averages, and this is the finding: +3.5% above a 50-day average of $334.98 and +2.1% above a 200-day average of $339.58 — but the 50-day sits BELOW the 200-day. This is the only such configuration among the twelve names in this batch, and it is the mechanical signature of a stock that fell hard enough earlier in the year to pull its short-term average through its long-term one and has not yet fully repaired it.
• RSI 41.9 and MACD +0.25 — weak-to-neutral.
• Maximum drawdown from peak over the trailing year: −9.91% — the current price is the low point of a shallow drawdown, though the intra-year low of $292.27 represents a 24.1% fall from the high, consistent with the knowledge base's account of a 20-25% March drawdown.
• Relative performance — the most important table in this section: 3-month +8.63% vs SPY +7.59% and QQQ +7.67%; 6-month −1.72% vs SPY +11.09% and QQQ +15.71%; 12-month +17.84% vs SPY +24.26% and QQQ +30.80%. American Express is the ONLY name in this batch that underperformed the index over twelve months, and the six-month gap of −12.8 points is the widest negative reading of the twelve.
• Sentiment: 23 buy, 30 hold, 4 sell across 57 analysts — consensus rating "Hold", one of only two in this batch alongside Goldman Sachs. Consensus target $378.09 (+9.0%), median $384, high $415 (+19.6%), low $315 (−9.2%).

Today's move

AXP closed 2026-08-04 at $346.85, up 0.62% or $2.13 from a $344.72 close, on 1.71M shares against a 3.21M average — 53% of normal volume and the lightest turnover in this batch. No company-specific news is in this file for the date; the last company event was the 2026-07-24 second-quarter release and 10-Q, eleven days earlier. While Citigroup rose 2.43%, Morgan Stanley 2.75% and Goldman Sachs 2.52% in the same session, American Express managed 0.62% — it did not participate in the financials move. That relative weakness on a day the sector rallied is a small but consistent datapoint alongside the six-month underperformance.

8. Insiders — a fractional director accrual

The most recent transaction in the file is an A-Award of 119.746 Share Equivalent Units to director Christopher David Young, filed 2026-07-02 for a transaction dated 2026-06-30, at a price of $0, leaving 20,991.92 units held.

The reading. Fractional share-equivalent units awarded at a zero cost basis on a quarter-end date are a deferred-compensation accrual, not a decision.

What the file contains is therefore: zero open-market purchases, zero open-market sales, and no officer transactions of any kind. Neutral, and reported as neutral. For a stock that has been in a 20-25% drawdown during the period covered by this file, the complete absence of any insider purchase is a mild disappointment — if the "fat pitch" the knowledge base describes were obvious to management, one would expect to see it here — but the absence of sales is equally notable and the net read is genuinely nil.

9. Verdict, kill-criteria and flip conditions

Hold.

What is genuinely good: total revenues net of interest expense of $19.637 billion in the June quarter, up 10% and 10% currency-adjusted, with first-half growth of 11%; net card fees of $2.862 billion, up 15%, "primarily driven by growth in our premium card portfolios" — an $11.4 billion annualised subscription business inside a payments network, growing faster than the payments volume; billed business up 9%; net interest income up 11%; International Card Services growing 13.4% in fiscal 2025 against a company rate of 10.5%; a 34.1% return on equity produced by a closed loop that captures the merchant discount, the annual fee and the net interest margin from one relationship; four consecutive years of 9-10% revenue growth and 18.9% compound diluted-EPS growth since fiscal 2022; a diluted share count down 9.8%; and the narrowest 52-week range and most predictable earnings stream in this batch — five quarters of beats averaging just +3.6%.

What prevents a Buy: a June quarter in which roughly 62% of the 15% pretax income increase came from provisions for credit losses falling 23%, with net income growing only 8% — below the 10% revenue growth — because the effective tax rate rose from 18.7% to 23.6%, and expenses growing 12% against revenue growth of 10%; a valuation of 21.0x trailing diluted earnings, 17.2x FY2027 consensus, 6.86x book and roughly 8.2x tangible book once the vendor's identical book and tangible-book figures are corrected for $4.873 billion of goodwill; 77.6% United States revenue concentration in a cardholder base explicitly exposed to white-collar employment; a base fair value of $370 that is 2.1% BELOW the street's $378.09, with a street consensus rating of Hold; a knowledge-base lane of eleven bulls and zero bears on a stock the market has been marking down for six months; and the only 50-day-below-200-day moving-average configuration in this batch.

The distinction that matters. Nothing here is broken. American Express is the most predictable business in this batch and it is compounding revenue at 10% and earnings per share at nearly 19%. The issue is that the market has spent six months disagreeing with that — the shares returned −1.72% while the index returned +11.09% — for a reason the knowledge base names precisely and no filing can settle: that artificial intelligence displaces exactly the affluent white-collar cardholders this franchise depends on. The June quarter gives that fear no support. The share price has not yet conceded. At 21x trailing earnings we are not paid enough to take a side on it, and a holder is meanwhile carried along by a business growing earnings 14% a year. That is a Hold.

Pre-registered KILL criteria — what would take this to Avoid:

• Provisions for credit losses returning above $1.4 billion in a quarter, which would remove roughly the entirety of the current pretax growth.
• Billed business growth falling below 6%, which would be the first hard evidence that the premium-consumer thesis is breaking.
• Net card fee growth decelerating below 10%. This is the highest-quality revenue line and the reason a 21x multiple is defensible.
• Any disclosure of weakening spend concentrated among affluent or small-business cardholders, which would confirm the artificial-intelligence displacement fear.
• Average merchant discount rates falling materially further, which would confirm the closed-loop scale constraint the knowledge base identifies.
• The 200-day moving average turning down with the price below it, ending the long-term uptrend that is currently intact by 2.1%.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price below approximately $300 — 14.9x FY2027 consensus and near the 52-week low of $292.27 — at which the base case is +23% and the risk/reward inverts. This is 13.5% below today's close and inside the range this stock traded in March 2026.
• The 2026-10-23 print with billed business accelerating above 10% and net card fees holding at 15%, which would falsify the white-collar thesis on the two lines that matter and would likely take the 50-day average back above the 200-day.
• Credit metrics improving further while spending accelerates, which is the specific combination that would prove the fear wrong rather than merely premature.
• International Card Services growth above 15%, reducing a 77.6% single-country concentration that is the largest structural risk here.
• Genuine bearish knowledge-base coverage appearing. A lane of eleven bulls and zero bears on a stock that has underperformed for twelve months is not corroboration — it is an absence of scrutiny, and it currently caps rather than supports the conviction.

Where AXP fits in the Synthos Framework Portfolio. The financials / quality-compounder sleeve, held rather than added to, at a 2% position with a 3.5% target on a fill near $300. On batch overlap: American Express, Citigroup, Morgan Stanley and Goldman Sachs are four financials in these twelve names, but American Express is the least like the others — its revenue is consumer spending and card fees rather than trading, advisory or net interest spread, its beta is 1.045 against Morgan Stanley's 1.218 and Goldman's 1.292, and its earnings are far more predictable. It is the closest thing in this batch to a consumer-quality compounder wearing a financial's clothing, and its correlation with the other three is lower than their correlation with each other. Within the financials group, Citigroup is the deep-value expression, Morgan Stanley the fee-annuity expression, Goldman Sachs the cycle expression and American Express the consumer expression. Logged as a tracked Synthos call (Hold) as of 2026-08-04 at $346.85.

Single biggest risk: the premium consumer, and specifically white-collar employment. 77.6% of fiscal 2025 revenue was United States and the cardholder base is affluent and small-business. The knowledge base records a 25% drawdown in March 2026 "on AI white-collar-displacement fears with no news", and the resulting relative performance is visible in this file: the only twelve-month underperformance of the twelve names here, and a six-month return of −1.72% against the index's +11.09%. No filing can settle this in advance. The June quarter shows billed business up 9%, net card fees up 15% and provisions falling — no damage at all. But the mechanism is real: if a material share of well-paid office employment is displaced, the discretionary spending, the willingness to pay a several-hundred-dollar annual fee, and the credit quality of this specific customer base all deteriorate together. At 21x trailing earnings and roughly 8.2x tangible book, there is no cushion for that, and the $265 bear case is what it looks like.

Most fragile assumption in the price: that credit provisions do not normalise. Provisions for credit losses fell 23% year on year in the June quarter, from $1.405 billion to $1.084 billion, and supplied roughly 62% of the $521 million pretax income increase. A return to the prior-year run-rate removes approximately $1.3 billion of annual pretax income — about 8% of pretax and roughly $1.50 of annual earnings per share. At 17.2x that is $26 of share price, or 7.5%, and it would arrive at exactly the moment the market decides the white-collar consumer story is real rather than imagined. That is the specific way this position loses money while revenue continues to grow 10%.

Provenance & disclosures

• Traceability: 12 tagged knowledge-base claims name American Express by entity, plus 5 text matches — THREE of the five text matches DISCARDED as collisions. The most striking is a 2026-06-26 mining-sector relative-valuation claim citing "Amex $217" alongside Snowline and New Found Gold — that is Amex Exploration, a gold-mining company, not American Express, and it is the clearest homograph in this batch after Merck KGaA. The other two discards are a metaphorical reference to a crypto platform as "the 'Amex of crypto'" and a claim about American Express Global Business Travel, a separately listed entity. On the entity lane: ELEVEN of twelve claims are bullish and ZERO are bearish — on a stock that has underperformed the index for twelve months, which we treat as an absence of scrutiny rather than as corroboration. FOUR of the twelve — a third of the lane — are a single argument restated across two days by one channel (compound_and_friends, 2026-03-03 and 2026-03-31): that Amex "down 20-25% on AI white-collar-displacement fears" is "a fat pitch." Counted as two voice-days. Only ONE claim carries a named speaker: Josh Brown, 2026-03-17, who states a testable conditional — the stock "won't fall 30% without a genuine softening in" employment. The single most analytically valuable item is the ONLY non-bullish claim in the lane (2026-02-12, neutral): that "Amex's closed-loop model caps scale and forces higher merchant fees, limiting market share (~10%) to affluent niche versus open-loop utility of Visa/MA" — which the 10-Q corroborates from an unexpected direction with its disclosure that average merchant discount rates fell "due to shifts in spend mix by geography and merchant categories." Breadth 5, net conviction positive-medium. No management voice appears. All quotes verbatim from stored claim text.
• Data as-of: income statement, revenue composition and segment detail for the quarter and six months ended 2026-06-30, from the Form 10-Q filed 2026-07-24; fiscal 2025 from the 10-K filed 2026-02-06 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873598 = 2026-08-04T19:59:58Z ($346.85, +0.62%; 50-DMA $334.98; 200-DMA $339.58; RSI 41.9; MACD +0.25) · knowledge-base claims 2026-08-04. American Express's fiscal year is the calendar year. All figures come from the Synthos vendor data file for AXP or from the SEC filings in the AXP archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-02-06 (fiscal 2025, tables preserved); 10-Q filed 2026-04-23 (March 2026 quarter); 10-Q filed 2026-07-24 (June 2026 quarter — the primary source for this dive); 8-Ks filed 2026-06-25, 2026-07-15 and 2026-07-24. Tables were available on all substantive documents, and the June 10-Q supplied the Table 1 summary of financial performance and the revenue-composition table used throughout.
• Where the filings corrected or extended the vendor (detailed in Section 6): book value and TANGIBLE book value per share reported as IDENTICAL at $50.56, ignoring $4.873 billion of goodwill — corrected to approximately $42.34 of tangible book per share, taking price to tangible book from 6.86x to roughly 8.2x, a 19% understatement, and the same error appears on all three banks in this batch; enterpriseValueTTM WITHHELD on a 49.6% residual and on the separate ground that enterprise value is inapplicable to a lender funding a $221.053 billion receivable book; netDebt corrected from $10.051 billion to $9.225 billion for the omission of $826 million of short-term investments; the entire quarterly operatingIncome and grossProfit series REJECTED as internally impossible — the June row shows $9,338M of operating income against $4,071M of pretax income, a gap that interest expense does not explain — and replaced with the 10-Q's Table 1; the quarterly revenue field identified as GROSS against a reported metric of revenues net of interest expense, with the identity confirmed exactly ($21,595M less $1,958M equals $19,637M) and priceToSalesRatioTTM rejected; the FY2030 estimate row DISCARDED as internally impossible (revenue falling 26.5% while EPS rises 8%, with net income exceeding EBIT); several receivable and cash-conversion ratios identified as lender artefacts (969.5 days of sales outstanding); the vendor's debtToEquityScore of 1 rejected as a lens while priceToBookScore of 1 is accepted as a correct signal; and quote.yearHigh/yearLow rejected in favour of tech.hi52/lo52. Where vendor and filing AGREED — worth recording: the seg_geo block ties to reported net revenue EXACTLY for FY2025; the gross-to-net revenue identity holds to the dollar; the implied share count is consistent with the diluted weighted average; and the earnings-calendar actuals reconcile to the 10-Q.
• Basis note: American Express reports GAAP results only and publishes no adjusted earnings measure; the earn_cal actuals match reported diluted EPS. All revenue figures in this dive are "total revenues net of interest expense", the company's own reported metric, derived from the vendor's gross figure by subtracting interest expense and verified exactly against the 10-Q. Trailing and forward multiples are all GAAP.
• Estimate coverage: 18 analysts on FY2026 and FY2027 revenue, 21 on both EPS lines; 14 and 14 on FY2028; 9 and 8 on FY2029. The FY2030 row is discarded outright as internally impossible. The FY2026 through FY2029 rows all pass the arithmetic defect screens. Notably, the FY2026 consensus of $17.670 is consistent with the first-half actual of $8.81 — implying a second half of $8.86, essentially flat — so unlike most names in this batch the American Express consensus is neither above nor below the run-rate.
• Peer note: the vendor peer set is mixed — Capital One ($136.13B), Ally Financial ($13.69B), Mastercard and Goldman Sachs are relevant in different ways, while Caterpillar's inclusion is a sector-code artefact. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. The structurally relevant comparison, made by the knowledge base rather than the payload, is against the open-loop networks: a closed loop with roughly 10% share of an affluent niche versus an open-loop utility.
• Fair-value caveat: the $265 / $370 / $460 anchors are multiples of the FY2027 consensus EPS of $20.116 — 13.2x, 18.4x and 22.9x — cross-checked against FY2026E, FY2028E and book value. Stated arithmetic, not a discounted cash flow, which would be inappropriate for a lender whose free-cash-flow figure is distorted by loan growth appearing in investing activities. Sensitivity: 15x gives $302, 20x gives $402; the street's $378.09 implies 18.8x. The bear case requires no earnings miss — only a de-rating to 13x on unchanged consensus; the bull case requires modest multiple expansion whose justification would be the decisive falsification of the white-collar-employment fear. No case attempts to quantify that fear, because no filing can test it in advance and we do not pretend otherwise.
• Timing: second-quarter 2026 results were released 2026-07-24 and the 10-Q filed the same day, eleven days before this dive, beating consensus EPS by 2.7%. The next print is 2026-10-23, 80 days away, with consensus of $4.58 EPS on $20.111 billion of revenue. This is the only name in this batch to have underperformed the index over twelve months (+17.84% against SPY's +24.26%), and the only one whose 50-day moving average sits below its 200-day — both facts are explicit inputs to the verdict.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.