Most investment memos are written backwards. They open with the market, move to the thesis, build a model that compounds for five years, and then arrive at a risks section near the end — usually three bullets, usually generic, usually written last by whoever had the least to say. By the time anyone reads that section, the decision has already been made emotionally. The risks are there to be acknowledged and then stepped over.
I have come to believe the order matters more than the content. Good downside risk analysis in investing is not a step you perform; it is the frame you start inside. If you begin with what would have to go wrong for this to be a bad outcome, and only then ask what it is worth if things go well, you end up with a different position size, a different price, and often a different answer entirely.
The loss case is a story, not a number
Asking "what is the downside?" usually produces a number — a percentage drawdown, a bear-case multiple, a haircut on the exit assumption. That number is almost always derived from the upside model with the growth rate turned down. It carries all the same structural assumptions. If the thesis depends on a customer segment behaving a certain way, the bear case usually assumes that segment behaves the same way, just less of it.
A real loss case is a narrative. It has a sequence. The large customer renews but at worse terms, which compresses gross margin, which makes the hiring plan unaffordable, which slows the product that was supposed to win the next segment, which gives a competitor an opening it would not otherwise have had. That chain is what actually destroys capital. No single link in it looks fatal in a sensitivity table.
So I make myself write the story. Not a sentence — a paragraph, with causation in it. If I cannot write a plausible paragraph explaining how this investment loses most of its money, I do not understand the business well enough to own it. That is a different failure from being wrong. Being wrong is survivable. Not knowing how you could be wrong is not.
Size against the loss, not the win
The upside case is the worst possible basis for position sizing, because it is the part of the analysis where you have the least information and the most enthusiasm. Upside is unbounded in imagination and bounded in reality. Downside is the reverse: bounded in reality at zero, but routinely underestimated in imagination.
The practical form of this is simple. Decide what the position is worth if the loss story happens, then ask whether you can absorb that outcome without changing anything else you do. Not whether it would hurt — of course it would hurt. Whether it would force a sale somewhere else, interrupt a commitment you have made, or change your behaviour on the next six decisions. If the answer is yes, the position is too large regardless of how good the opportunity is.
This is where margin of safety earns its keep as a practice rather than a slogan. The margin is not only in the price. It is in the size. You can buy a wonderful asset at a fair price and still be wiped out by owning too much of it at the wrong moment. Price protects you against being wrong about value. Sizing protects you against being wrong about timing, and timing is the thing almost nobody gets right.
Separate the permanent from the temporary
There is a distinction I apply to every risk I can name: does this destroy capital permanently, or does it only make the next eighteen months unpleasant? These are not the same risk, and treating them identically is how investors end up terrified of volatility and relaxed about leverage.
Temporary impairment is a multiple that contracts, a cycle that turns, a quarter that disappoints. It is painful and it is recoverable, provided you are not forced to transact during it. Permanent impairment is a business whose customers have found a better answer, a balance sheet that hands the equity to the lenders, a regulatory change that removes the product, a founder who leaves with the relationships. Those do not come back when the cycle does.
The question that converts a temporary risk into a permanent one is almost always about forced action. Who can make us sell, and when? Debt covenants can. Redemption terms can. A partner's liquidity needs can. Our own panic can. I spend more diligence time on the mechanics that could force a transaction at the bottom than on refining the growth forecast, because the forecast is a guess and the covenant is a contract.
Where the downside usually actually lives
After enough cycles you notice that losses rarely come from the risk at the top of the risk register. The named risk gets watched. The unnamed one is what gets you. In my experience the recurring sources are concentration, dependency and governance — and all three are visible before you invest if you are willing to look at them unflatteringly.
Concentration is the obvious one, and it is broader than customer concentration. It includes a single channel that produces most of the leads, a single engineer who understands the core system, a single geography, a single regulatory regime. Dependency is concentration you do not control: a platform whose terms can change, a supplier with pricing power, a distribution partner who could decide to compete. Governance is the one most often skipped, because it feels like a soft issue until the moment it is the only issue. A board that cannot have a hard conversation will not raise an alarm early enough to matter. I have written separately about the three questions I take to any risk on a board agenda — concentration, reversibility and detection — and the same frame works on the way in, before the money moves.
Detection deserves particular attention in an investment context. The relevant question is not only how bad the loss case is, but how long it would take to know it was happening. A business with monthly recurring revenue and a short sales cycle tells you quickly. A business with long contracts and lumpy renewals can be two years into decay before the financials show it. All else equal, I will accept a worse loss case that I can see coming over a milder one that stays hidden until it is finished.
Diligence that is trying to find the no
The reason downside work gets done badly is that diligence is usually organised to confirm a decision rather than to stress it. The deal team wants the deal. The seller has prepared the materials. Everyone is working from the same data room, which is by construction the seller's version of the story.
The fix is less about process documents and more about who is assigned to disbelieve. Someone in the room has to be responsible for the loss case and evaluated on how well they argued it, not on whether the deal closed. That role has to be real. If the person who raises the inconvenient question is the person who slows everyone down, the role dies quietly within two deals. Some of this is cultural, and some of it is simply being deliberate about which questions you ask and what the answers tend to predict.
I also look hard at the question of what the business does when it is under pressure. Not what management says it would do — what it has already done. Past behaviour during a bad quarter is more informative than any plan. Did they cut the right things? Did they tell the board early or late? Did customers find out before the board did? A team that handled one downturn candidly will probably handle the next one candidly.
What the discipline costs
I should be honest about the price of this approach, because it is not free. Underwriting the downside first means you will pass on things that go on to work beautifully. You will sit out opportunities where the loss story was genuinely plausible and simply did not happen. You will feel slow in markets where speed is being rewarded, and you will occasionally be wrong in the direction of caution, which is less visible than being wrong in the direction of enthusiasm but is still being wrong.
What you buy with that cost is the ability to keep playing. Capital preservation is not a defensive posture; it is what makes the next decision possible. An investor who avoids the permanent loss gets to compound judgment as well as capital, because they are still in the room when the good opportunity arrives. An investor who takes one unrecoverable hit has their whole future decided by a single bad paragraph they never wrote down.
There is also a second-order benefit that is hard to quantify. Starting with the loss case changes the conversations you have with management after you invest. You already know what you are afraid of, you have said it out loud, and the people running the business know it too. That makes the quarterly review honest from the first one, rather than a negotiation about whether anything is actually wrong. It also tends to produce better relationships, because it is clear you are watching the same things they are. The work that follows a closing goes considerably better when both sides have already agreed on what failure would look like.
Write the bad paragraph first
The single habit I would recommend to anyone doing investment risk assessment is also the cheapest. Before the model, before the market map, before the call with the founder you already like — write the paragraph explaining how this loses money. Date it. Keep it. Read it again in a year.
Most of the time it will be wrong in its specifics and right in its structure. The thing that actually hurt you will be a cousin of the thing you wrote down, and you will recognise it earlier because you were already looking in that direction. That recognition is most of what good investing is. We do not get paid for being optimistic; plenty of people are optimistic for free. We get paid for knowing, in advance and in writing, exactly what we would have to be wrong about.