Actionable insights — Private Credit's Clock is Ticking

17:59 1. Before evaluating a private deal, ask who is being paid to put it in front of you

1. When an unlisted opportunity arrives — a private fund, a direct deal, a "special" allocation — suspend the merits question entirely at first pass. Eisman: "the first question isn't, 'Is this a good deal?' It's 'Who's getting paid to show it to me?'"
2. Map the compensation chain end to end: who earns a placement fee, who earns carry, who earns a retrocession on the wrapper, who earns nothing. Then ask what that chain selects for — deals that are easy to sell, not deals that are hard to source.
3. Only after the chain is mapped, evaluate the asset — and weight the diligence toward the parts the seller has an incentive to be vague about (marks, fees, gates, liquidity terms).
4. Where possible, replace the seller's materials with peer diligence from people with no economics in the transaction.

• A product whose fastest-growing distribution channel is the least sophisticated one; commissions or platform fees that scale with gross sales rather than performance; "limited allocation" urgency attached to an evergreen fund.

4:37 2. The lazy-cash audit — find the line item the customer doesn't know they're paying

1. For any financial business, ignore the headline product and locate the most profitable revenue line. Ask the question Eisman forced here: what is the mechanic, in one sentence? ("Whatever is not invested… is swept into a money market fund or a bank, and the broker will make a spread on that cash.")
2. Establish who is on the other side of it and whether they know. Ask literally: if you polled the customers, would they say they are unhappy? If the honest answer is "they don't know," you have found an information-rent business, not a service business.
3. Estimate what fraction of profit disappears if the information asymmetry closes. Then ask what remains — is there a service the customer would pay for on purpose?
4. Rate the durability by switching friction, not satisfaction: how many products deep is the relationship, is it the direct-deposit account, would the customer "rather pluck [their] eyebrows out than actually switch"?
5. Look for the historical episode where the asymmetry was briefly closed and measure what actually left. That is your estimate of the un-sticky share.

• A firm that gives its headline product away free (zero commissions) while a disclosed-but-unread footnote carries the economics; deposit betas that lag peers; management describing an optimisation feature as something they may "allow" rather than build.

10:03 3. Convert a bull defense into a list of things that all have to be true

1. Take management's (or the analyst's) reassurance and decompose it into independent conditions rather than accepting it as one argument.
2. Assign each condition an owner: the customer, the company, the regulator, the technology. Conditions owned by someone outside the company are the ones to underwrite hardest.
3. Multiply. A thesis that needs four 80%-likely conditions is a coin flip.
4. Then invert: ask what single condition failing is sufficient to break it — and whether the company can respond, or only watch.

• Defenses that rest on a regulator's future behaviour; "we don't have to open the API" arguments in a market where one competitor's opening forces everyone's; the fee-replaces-spread pivot, which is usually smaller but far more defensible — model both.

13:56 4. Set an explicit error-rate threshold before delegating money to a machine

1. Don't ask "is the AI good?" Ask what accuracy is required for this specific delegation, and set the number first. Worthington's bar for handing over portfolio management: "not just the majority of the time, not even the vast majority of the time, but… like 99.99%."
2. Identify the failure mode, not just the failure rate. Here it is sycophancy: agents "tell investors what they want to hear based on the prompts that are given, not necessarily what is the optimal way to manage a portfolio." A model that agrees with you scores well on user satisfaction and badly on outcomes.
3. Test against a stated objective the agent cannot see in the prompt, so agreement can't substitute for correctness.
4. Scale the delegation to the reversibility of the error. Optimising a cash sweep is recoverable; discretionary allocation is not.
5. Re-run the threshold test periodically — this is a moving frontier, and the investable question is when it clears, not whether.

• Published agent-accuracy benchmarks on financial tasks; a major broker opening a cash-optimisation API to third parties; a regulator explicitly blessing (or forbidding) agent-initiated transfers; the first firm to monetise its own optimisation agent as a fee.

15:39 5. Read a fund complex through three separate flow numbers, never the net

1. Split flows into gross inflows, redemption requests, and the gate (what the fund is contractually allowed to pay out). Net flow hides all three.
2. Gross inflow is the demand signal and it moves first — "you can almost not see the line. The inflows, the gross sales inflows into the funds are infinitesimal."
3. Redemption requests versus the gate tells you whether the fund is in control: still "over the 5% per quarter limit" means requests exceed what can be paid.
4. Then ask a question the aggregate can't answer: are these new sellers or the same sellers re-queuing? "It's the same people asking for their money. There's not as many new people asking" is an easing; a widening roster is contagion.
5. Finally check the underlying: are the assets still performing? Redemptions driven by anxiety about a future event ("we're just predicting their demise someday") behave differently from redemptions driven by realised losses.

• Gates being hit for consecutive quarters; a fund raising debt or selling assets to fund redemptions; the roster of redeeming advisers widening; substitution into an adjacent product (real estate → credit → PE/infrastructure) stalling — "we're not at the point where you're seeing perfect substitution."

24:37 6. Reprice an unmarked private asset off the cleanest public comparable — then ignore the cash-flow reassurance

1. Find the highest-quality listed business in the same category as the private assets you're worried about. Quality matters: if the good one is down, the milked one is worse.
2. Apply its peak-to-trough de-rating to the private cohort as a first-pass mark: "any software company owned by private equity has to be down something like 50% from… where it was bought because that's what the public markets have done."
3. Now separate the two questions people conflate. Is the borrower performing? and is the collateral still worth the loan? Managers answer the first when asked the second.
4. Refuse the substitution explicitly. Eisman: "I'll give them the statement that the companies that they own are doing okay. I don't think it matters… It's just that the valuation of the company is half."
5. Restate the refinancing as the lender sees it: same loan, half the equity cushion → "put up some more money."

• Private marks that haven't moved while the listed comp de-rated; risk buckets ("85% low risk") defined by current payment status rather than collateral value; a manager quoting borrower EBITDA growth in answer to a question about loan-to-value.

24:12 7. You cannot date the outcome — but you can date the moment someone is forced to act

1. Stop trying to forecast whether the bad thing happens. Instead find the contractual event that removes everyone's option to wait: a maturity, a covenant test, a gate, a lock-up expiry, a reset.
2. Size the cohort at that event ("270 billion of software maturities held by financial sponsors… in 2028 and 29") and then back up the calendar by the normal refinancing lead time: "it's somewhere in the next two quarters, three quarters where they'd normal course of business be refinancing."
3. State the resulting date out loud and let it discipline the position size and the patience: "we are a year away from finding out what we're going to really need to find out… until then, you could say whatever you want."
4. Between now and that date, watch process evidence (who is negotiating what) rather than expecting price evidence.
5. Re-derive the date when the cohort changes — extensions, amend-and-pretend and LMEs push it out, and the pushed-out version is itself information.

• The refinancing lead time shortening (a sign lenders are stalling); a wave of maturity extensions or liability-management exercises that reset the clock; the first cluster of sponsor equity injections or keys-back handovers — either resolves the stalemate and dates the marks.

28:05 8. Count the tails of a negotiation, not the average — a stalemate is data

1. For any standoff, define the two extreme resolutions. Here: the sponsor hands the lender the keys, or the sponsor writes a fresh equity cheque.
2. Go count how many of each have actually happened — through practitioners on the other side of the table (restructuring lawyers, workout desks), not through the principals' commentary.
3. If both tails are near zero, you are in a stalemate, and a stalemate means nobody has been forced to price it yet. Treat every reported mark in that window as untested.
4. Track the ratio over time. The tail that starts filling first tells you who blinked — and by extension what the marks should be.

• The first cluster in either tail; restructuring-lawyer hiring and workout-desk staffing ("you're not going to see them for a few years"); opportunistic credit funds actually deploying rather than accumulating dry powder.

29:44 9. Expect credit stress to arrive as terms, not as defaults — and price the terms

1. Ask whether there is a bid of last resort. Where large opportunistic-credit pools are raised and idle, a refusal to roll doesn't create a default — it creates a transfer.
2. Then read the terms of the transfer as the real mark: "put in a lever layer of credit with more favorable position, scratch back better terms, don't let them do LMEs, don't let them do certain EBITDA adjustments, do it at a 300, 200 basis point higher yield."
3. Convert those terms into the implied write-down for the incumbent lender: "they'll happen at worse terms… it'll eventually bring markdowns that you haven't seen yet."
4. Cross-check against the return stream, which moves before the marks do: direct lending has already gone from "mid-teens plus" to "high single digits, mid single digits," and "you're seeing this flow through the valuations already. Gradually."
5. Set expectations accordingly: gradual repricing, not an event. Position sizing and patience follow from that, not from the headline.

• New-issue spreads on sponsor software paper widening 200-300bp; covenant packages tightening (EBITDA add-back limits, LME blockers); reported direct-lending returns compressing further; unitranche deals being split into senior + opportunistic layers.

21:06 10. The sell-into-strength test — a seller who can't monetise at the top is telling you the mark is wrong

1. Establish that the exit environment is as good as it gets: open IPO window, active M&A, record equity prices. Remove the excuse before you ask the question.
2. Now ask who is and isn't selling. "Despite better IPO market, better M&A market, and record highs in the equity markets, they can't sell some of the stuff they own."
3. Treat inability to sell into that backdrop as a valuation statement, not a timing preference: "it makes people question how good of an asset is it or did you just pay too much and you have to wait several more years for earnings to grow into that valuation."
4. Use dispersion within the peer group as the control — if one manager prints record monetisations in the same quarter, the constraint is asset-specific, not market-wide.
5. Overlay the vintage: money "put to work at pretty high multiples with zero interest rates" in 2020-21 is now at the age where it should be sold. Age plus silence is the signal.

• Realisation ratios (distributions ÷ NAV) diverging across managers in the same quarter; continuation vehicles and NAV loans substituting for genuine exits; hold periods extending past six years without a stated reason.

32:44 11. Falsify an asset's story with its own trading behaviour

1. Write down the single stated reason the asset is owned. If you can't state it in one sentence, that is already the finding.
2. Derive what that reason predicts about behaviour on specific days: if it is a hedge against currency debasement, it must rise on frightening days and lag on calm, risk-on days.
3. Check the actual pattern. "My issue is that crypto trades inversely to its own thesis" — rallying on the good days, falling on the frightening ones.
4. If behaviour contradicts the story, the asset is owned for an unstated reason. Go find it, or decline to own it: "I've never heard a thesis other than this fiat currency hedge."
5. Invite the counter-thesis explicitly and judge whether it is coherent, separately from whether you agree — Eisman accepts Worthington's use-case framework as "actually coherent" while still declining Bitcoin.

• Any asset whose correlation flips sign against the reason given for owning it; a story that migrates ("replacement currency" → "hedge" → "digital gold") each time the previous one fails; the competing outlet for the same speculative demand (prediction markets — "all the young people who used to trade Bitcoin are on Kalshi").

37:58 12. Audit the concentration inside the asset class you're bullish on

1. State the source of value you believe in — here, usage of the underlying network.
2. Now measure where the market value actually sits, and check whether the two line up. "There is an awful lot of value in a small number of tokens… half the market's Bitcoin."
3. If the largest constituent is the one with the least of the thing you say creates value, your bullish case and your index exposure are in conflict. Own the thesis, not the aggregate.
4. Express the view as a rotation rather than a directional bet: "over time we transition from this concentration in Bitcoin to other stuff."
5. Apply the same audit to any thematic exposure — an index whose weight sits in the least-thematic name is a different bet from the one you intended.

• Bitcoin dominance falling as chains and tokens proliferate; new chains launched by the venues themselves (Coinbase, Circle) drawing real activity; a thematic ETF whose top holding doesn't fit the theme.

40:55 13. The incumbent-awake test — check whether the giant has already responded before underwriting a disruption story

1. Every disruption story implicitly assumes incumbent passivity. Name that assumption and test it directly: has the incumbent shipped a competing product, joined a consortium, or bought a position?
2. Use the base case both ways. Netflix worked because "all the entertainment companies acted like idiots and let them grow like crazy" — so the question is whether these incumbents are those incumbents.
3. Grade management quality specifically on paranoia, not margins: V and MA are "among the smartest… they're not asleep. They're very active in crypto technology."
4. Count the coalition. A consortium of 140 financial institutions plus the banks' own interoperable tokenised deposits is not a competitor, it is a market structure.
5. Re-underwrite the challenger on what it can build without winning the contested market — Worthington's dollarization wallets, cross-border transfers and a 24/7 settlement layer are real businesses; the payments prize may not be available.
6. If the challenger's only path to the target valuation runs through the contested market, the rational action is to sell to a bigger balance sheet: "I think Circle needs to sell to somebody who's got bigger pockets."

• Incumbent consortium announcements (the tell that the disruption has been absorbed); merchant-side adoption of stablecoin rails as an interchange-avoidance play; the challenger pivoting its own messaging from "replace" to "settle"; a bid arriving for the challenger.

48:13 14. Forecast trading revenue from the dispersion of opinion, not from volumes

1. Schorr's rule: "When everybody agrees on something, trading's not going to be so good. And when we don't agree trading's going to be pretty good." Consensus kills two-way flow.
2. Inventory the live disagreements — rate path (cuts priced out, hikes back in the conversation), geopolitics, the software/AI debate — and treat a long list as a revenue forecast.
3. Separate revenue from volume before extrapolating. Q2 equity revenues +68% on volumes up only 9-10% means the money came from spreads, financing and margin balances ("the money on borrow was up 50"), not from activity.
4. Apply the known seasonal: "trading is historically down 17% second half from the first half" — subtract it before treating a first-half run-rate as a base.

• Consensus converging on a single rate path or a single AI verdict (bearish for trading); margin balances and financing revenue as the quality tell inside a strong trading print; deposit competition raising the cost of funds and squeezing the boring half of the bank.

50:21 15. Treat "everything is humming" as a risk reading, not a confirmation

1. When every line of a business is simultaneously at or near best-case, stop and count how many things must stay right to hold the run-rate.
2. Note that estimates have already risen to meet the result — "the earnings estimates have all gone higher" — so the good news is now the base case, not the upside.
3. Ask what breaks first if any single line normalises (trading seasonality, IB dependence on one financing theme, deposit costs).
4. Don't convert the observation into a short. Eisman doesn't: the discipline is to hold the caveat alongside the position, "but for now, everything is really great."

• Consensus estimates rising to the point where in-line is a disappointment; every division above trend at once; the first single line to normalise (usually the seasonal one) and whether the multiple reacts to it.