What 22 Investment Post-Mortems Taught Me

Autopsy Season 1 is complete. The patterns that repeated across the failures (and successes), and the process improvements they demand.

Jul 10, 2026

I have spent 25 years watching intelligent, experienced investors lose money on investment theses that sounded reasonable at the time. A few of them were mine. I also observed others succeed in investments for reasons I didn’t understand.

That is what the Investment Autopsy series has been about. Every week I took one concluded investment thesis and asked the question I find more useful than any stock recommendation: what led to the success or failure of the investment thesis of an experienced investor, and how can we improve our own process as a result of what we learn?

I published Autopsy No. 22 last Thursday. With it, Season 1 is complete.

I am calling it a season on purpose. Each Autopsy is one data point. After 22 of them, the next case teaches less than the pattern already sitting in the ones I have written.

The same behavioral traps keep showing up. The same categories of mistakes recur, in businesses that otherwise have little in common. This is the moment to stop, look across all 22 at once, and pull out what they teach together. No single case can show that.

So, this week, in place of Autopsy No. 23, you are getting the synthesis: the patterns that repeat across all 22 post-mortems.

Two quick notes before we get to the patterns.

The archive stays where it is. All 22 cases remain in the paid library, and they are even more useful now than they were arriving one Thursday at a time. You can read them as a complete set, by failure type, when your own thesis starts to resemble one of them.

I am spending the rest of the summer reorganizing the paid library around the investing challenges many of you face, and building new tools to sit alongside it. Free essays continue as usual in the meantime.

The next new paid article comes out in August. Now, on to the patterns across the autopsy lessons.

Lessons across investment autopsies

Secular decline beats cheap valuation

The most common cause of death in the series: a statistically cheap stock facing a structural force pushing its revenue or economics down. It happened 8 times. In every case, the cheapness lost.

The mechanism repeats. Falling revenue hits a mostly fixed cost base, so profits fall faster than sales; Journal Register‘s profits halved on a 20% sales decline.

The market then correctly re-rates the company as a declining earnings stream, so the multiple compresses on top of the falling earnings; Lexmark fell another 44% from what was already its cheapest valuation ever.

Management responds with reasonable-sounding countermeasures, and they prove insufficient. If debt is present, it sets the deadline.

Why do investors this experienced keep making this mistake? Because value investors find ideas by screening for cheapness, and cheap screens are full of secularly challenged companies.

They are cheap for a reason. The behavioral pull is anchoring: a decade of strong historical margins makes the current weakness look like a deviation that will mean-revert. Mean reversion is a real force when the problem is cyclical. When it is secular, the mean itself is moving down.

Where it showed up:

Staples (No. 7): the write-up named Amazon and channel shift as risks, then valued the company as if neither would bind. Both did.

Kodak (No. 9): the short. Same force, opposite direction: the bet was that the visible decline would continue, and it did.

Blockbuster (No. 16): everyone was negative on it. Everyone was right. Contrarianism alone is not an edge.

Lexmark (No. 19): the other short that worked, for the same reason the longs kept failing.

Hain Celestial (No. 21): organic food went mainstream and competitive; the category matured out from under the thesis.

DXC Technology (No. 22): legacy IT contracts running off, disclosed in the filings all along.

Journal Register (No. 12) and Bed Bath & Beyond (No. 14, the second act) belong here too; debt finished what the decline started, and they appear again below.

The process check: classify the headwind before you value anything: cyclical, secular, or internal. If cyclical, the work is survival analysis and mid-cycle valuation; that is how Builders FirstSource became the biggest winner of the series. If secular, the burden of proof flips: demand demonstrated stabilization, a catalyst, or a healthy segment insulated from the decline. If secular and levered, pass.

Evidence beats potential, in both directions

This lesson separates the winners from the losers more cleanly than any other single factor. Almost every failed long bet on a future that nothing in the numbers yet supported:

eDiets (No. 10): needed broken subscriber economics fixed. The fix never came; the loss was near total.

Sleep Number (No. 20): needed sales growth to convert into profit growth. Sales grew 10% a year; operating income grew 1%.

Advance Auto Parts (No. 4): my own thesis. It needed a margin gap to O’Reilly and AutoZone to close before there was any evidence of closing. The gap widened. My exit was profitable, and it was luck.

Sprint Nextel (No. 18): mine as well, with Bill Miller for company. Needed announced merger synergies to show up. What showed up was a $30 billion impairment.

Omnicare (No. 5): fundamentals stalled under a revolving door of CEOs; a premium CVS takeout rescued the result.

The successful theses were based on change that was already observable. Gartner (No. 13): the product shift and resilient renewal rates were in the reported numbers when Ubben presented. TJX (No. 17): the returning CEO had already run the same business well once, and did it again. Be careful if your thesis is based on expecting a future that is different from what has actually been happening to the company thus far.

McKesson (No. 15) makes this a 2-way rule. I passed on it over a theoretical worry, undifferentiated distributors competing away profits, that had no real-world evidence behind it. It never materialized. Talking yourself out of a good investment with an untested fear is the same error as talking yourself into a bad one with an untested hope, run in reverse.

The process check: separate what is already happening from what must change for the thesis to work. Have a very high bar for betting on future developments that aren’t yet supported by current fundamentals.

Debt plus fixed costs plus falling sales is lethal

The combination kills, not any single element. A company can be undervalued on correct mid-cycle earnings and still hand shareholders a zero, because the trough arrives before the mid-cycle does and creditors own the company by the time the value shows up.

Lear (No. 1): the cleanest example. The whole normalized-EPS debate became irrelevant once Chapter 11 wiped out the equity.

Journal Register (No. 12): the same math, faster. A fast exit contained the damage.

Blockbuster (No. 16): debt plus a shrinking business, with the substitutes already visible.

Bed Bath & Beyond (No. 14): the self-inflicted version. Borrowing heavily to buy back shares while fundamentals decline assembles the lethal combination voluntarily.

Hain Celestial (No. 21): debt removed the time the turnaround needed.

The counterexample earns its place. Builders FirstSource (No. 8) carried real distress risk into a housing depression and returned 46% a year for 5 years. Robotti did explicit survival math: cash on hand, burn rate, dilution scenarios, and whether the shareholder base could fund a raise. He sized the path, not just the destination.

The process check: before valuing the destination, ask whether the balance sheet survives the worst plausible 2 to 3 years. If the thesis needs years and the balance sheet grants quarters, the valuation work may be moot.

Approximately right on a few variables

Almost every winner was wrong on its point forecasts. Wyden’s Ferrari (No. 2) EBIT estimate missed by 40%. Tarasoff’s Amazon (No. 6) e-commerce penetration guess came in high, and his retail margins landed at the low end. Robotti‘s 10% EBITDA margin never arrived.

None of it mattered, because each thesis was right on the direction and rough magnitude of the 1 to 3 variables that actually controlled value.

American Tower (No. 11) is the purest version: one variable (tenants per tower), one driver (cellular data demand), one consequence (profits growing off a near-fixed cost base). No 500-line model. McKesson and TJX share the shape: a handful of levers, each one checkable.

I supplied the counterexample myself. My Sprint model was detailed, precise, and wrong, and its precision is what let me rationalize away the subscriber losses that were breaking the thesis. Complexity gave comfort. It did not reduce risk.

The process check: name the 2 or 3 key variables that drive the value and hold an approximate view on each supported by the evidence. If the thesis needs precision, pass.

Selling is a separate skill from buying

The series shows this from every angle, and the spread of outcomes is enormous.

Holding a winner: up 42% in 9 months on Ferrari (No. 2), Wyden re-underwrote the thesis, saw further large upside, and held. That one decision captured most of the eventual return.

Selling a winner too early: Robotti (No. 8) sold half of what became a 50-bagger between $4 and $7, anchored to his $2 entry. Ubben exited Gartner (No. 13) in early 2012; it compounded at 25% a year for the next decade.

Selling a loser fast: Rogers exited Journal Register (No. 12) within weeks of presenting it, converting a would-be wipeout into a merely bad outcome. Welling’s firm escaped Hain (No. 21) in 2021, before the 97% decline.

Refusing to sell: Pzena held more than half of his Lear (No. 1) position nearly to zero, sustained by a belief the bankruptcy process falsified.

Exiting when the insight is exhausted: American Tower (No. 11) lagged the market from 2010 on, once the tenancy story was priced. The business stayed excellent. The mispricing was gone.

The rule that unifies all of these: update the value estimate on new information, and anchor to nothing else. Entry price is a sunk cost.

The process check: Is your current value estimate based on your original facts or the latest version? If the former, you must rethink your thesis, update your value and act accordingly.

How to use the library from here

The 22 cases are indexed below. My suggestion: read them by symptom.

When one of your own holdings starts producing the same arguments you have seen in these cases, when you catch yourself extending the timeline or ignoring the balance sheet, go find the case that matches and see how it ended.

That is what a post-mortem library is for. You study the failure while you still have time to act on your own investment to avoid it.

Thank you for reading this season, and to the paid members especially, for supporting work that takes the slow route. If one of the 22 cases stayed with you, or if there is a research problem you want the paid tier to do more work on, leave a comment or send me a message. I read every one.

I am working on finishing some exciting new tools and resources that I plan to share with you starting in August. Until then, here is to better investment process, not more opinions.

A Request

Help grow our community of thoughtful investors by restacking this article.

The index: the cases by pattern

Most cases teach more than one pattern; each appears under every pattern it fits.

Secular decline beats cheap valuation: Staples (No. 7), Kodak (No. 9), Journal Register (No. 12), Bed Bath & Beyond (No. 14), Blockbuster (No. 16), Lexmark (No. 19), Hain Celestial (No. 21), DXC Technology (No. 22)

Evidence beats potential, in both directions: Nestlé (No. 3), Advance Auto Parts (No. 4), Omnicare (No. 5), eDiets (No. 10), Gartner (No. 13), McKesson (No. 15), TJX (No. 17), Sprint Nextel (No. 18), Sleep Number (No. 20)

Debt plus fixed costs plus falling sales: Lear (No. 1), Builders FirstSource (No. 8, the survivor), Journal Register (No. 12), Bed Bath & Beyond (No. 14), Blockbuster (No. 16), Hain Celestial (No. 21)

Approximately right on a few variables: Ferrari (No. 2), Amazon (No. 6), Builders FirstSource (No. 8), American Tower (No. 11), McKesson (No. 15), TJX (No. 17), and Sprint Nextel (No. 18, the counterexample)

Selling is a separate skill from buying: Lear (No. 1), Ferrari (No. 2), Builders FirstSource (No. 8), American Tower (No. 11), Journal Register (No. 12), Gartner (No. 13), Hain Celestial (No. 21)

Management and culture change take years: Advance Auto Parts (No. 4), Omnicare (No. 5), Gartner (No. 13), Bed Bath & Beyond (No. 14), TJX (No. 17), DXC Technology (No. 22)

Large acquisitions carry bad base rates: Advance Auto Parts (No. 4), Omnicare (No. 5), Builders FirstSource (No. 8, the disciplined exception), Sprint Nextel (No. 18)

History-based normalization fails when the business has changed: Lear (No. 1), Nestlé (No. 3), Lexmark (No. 19)

A risk listed is not a risk weighted: Staples (No. 7), Blockbuster (No. 16), Hain Celestial (No. 21)

Unit economics govern everything above them: eDiets (No. 10), American Tower (No. 11)

Well-followed companies demand a differentiated view: Nestlé (No. 3), Blockbuster (No. 16)

Appendix: the 22 cases in order

Tickers as covered in each case; several have since been acquired or delisted.

Gary Mishuris, CFA is the Managing Partner and Chief Investment Officer of Silver Ring Value Partners, an investment firm that seeks to apply its intrinsic value approach to safely compound capital over the long-term. He also teaches the Value Investing Seminar at the F.W. Olin Graduate School of Business.

Discussion about this post

Ready for more?