Your plan says 87%. Here is what that number counts.
A couple is at the table looking at a report. The headline number is 87%.
One of them asks the only sensible question: "So there's a 13% chance we run out of money?"
No. And the reason why is worth thirty seconds of your time in that room, because the honest answer changes what the whole conversation is about.
The number is a count
Nothing in a Monte Carlo result is predicting your life. The engine builds a complete simulated lifetime, start to finish, then does it again, then does it a thousand more times. In Foundry the default is 1,000 trials, which is the industry convention. Each trial gets its own sequence of market returns. Everything else about the plan stays exactly as you entered it.
At the end, each trial is stamped pass or fail. The percentage is the passes divided by the trials. That is the entire calculation. 870 of 1,000 cleared, so the report says 87%.
Which means the number is not a probability about this household in the way people hear it. It is a tally of how a particular model behaved under a particular definition of "cleared," run a particular number of times. Michael Kitces has written about this at length: "We can never assess the quality of a forecast based on a single observation." Your client gets exactly one lifetime. The 87% never applies to it directly.
Worth saying out loud: the trials are also reproducible. Foundry draws returns from a seeded generator, so running the same plan twice returns the same 87%. That stability is a property of the seed. It is not the future settling down.
Three choices hide inside every percentage
Here is the part almost nobody looks at. Before a tool can count passes, somebody had to decide three things. All three are judgment calls, and all three move the number without changing anything about the household.
What counts as money. Foundry scores the liquid portfolio only: taxable, cash, and retirement accounts. Real estate, business interests, and life insurance are deliberately excluded, on the reasoning that you cannot cleanly liquidate a house or a closely held company to cover a shortfall in a specific year. That is a defensible choice, and it is still a choice. A tool that counts home equity as spendable will clear the bar more often, with identical facts.
What counts as falling short. In Foundry a trial fails if the liquid portfolio goes below zero in any simulated year, or if the ending balance lands under the plan's minimum asset level. Two checks. That minimum defaults to zero, and it is a setting. Set it to a legacy floor because a client wants to leave something behind, and the same thousand trials return a lower number. Nothing about the household changed. The bar moved.
What varies across the trials. In Foundry, market returns vary, drawn as correlated lognormal draws per asset class from the return and standard deviation on each one, with the correlation matrix respected so equities and bonds do not wander independently. Inflation can be randomized too, if the plan is configured that way. What does not vary is everything else: the spending, the claiming ages, the plan horizon, the tax law. All thousand futures assume your client lives exactly as long as the plan says and spends exactly what the plan says. The full year by year tax computation does run inside every trial, which is not universal, but the tax law is the same in all thousand.
So the honest framing of the number is narrow: given these assets, this definition of falling short, and a thousand different market paths, this share of them cleared.
Two honest numbers that disagree
The cleanest demonstration that the counting rule is a choice is that we ship two of them, on purpose, on the same screen.
Foundry's plan level metric fails a trial on either check above: negative in any year, or ending under the minimum. The ending distribution chart carries its own separate stat, and that one counts only the ending value.
They produce different numbers, and both are right.
A trial that dips below zero in 2041, recovers, and ends comfortably above the minimum is a shortfall under the first rule and a clean pass under the second. The boundary is asymmetric too: a trial that lands exactly on the minimum passes the plan level test and counts as below on the chart.
The temptation, when you notice this, is to pick one and collapse them. We didn't, because they answer different questions. "Did this plan ever get into trouble" and "where did this plan end up" are both worth knowing, and a client who has been through a bad decade cares a great deal about the difference. What we did instead was refuse to give them the same name. The plan level number is Shortfall Risk. The chart's stat is Ended below minimum, which is precisely what it measures. There is a test in the codebase whose only job is to fail the build if those two labels ever collide.
If your tool reports a single failure number, it made this same choice. It just made it somewhere you cannot see.
What the 13% is not
The complement is where the most damage gets done at the table, because "13%" invites exactly one interpretation and it is the wrong one.
Thirteen percent of trials fell short of a bar. That is all. It does not say by how much, and it does not say when.
A trial that ends four thousand dollars under the line counts once. A trial that runs dry at 74 with sixteen years to go counts once. Same weight, same tally. eMoney, whose tool a third of the industry uses, says this plainly about their own engine: "Every trial within a Monte Carlo analysis is either a pass or a fail... A trial that falls $1,000 short is labeled a failure just the same as the trial that falls $2,000,000 short."
That is not a knock on the method. It is the arithmetic of counting. But it means the percentage has thrown away the two things you would actually act on. Magnitude and year are the planning information. The tally is the summary that discarded them.
Which is why the follow up question matters more than the headline. Not "is 87% good enough," but "in the trials that fell short, when did it get tight, and by how much?" That question has an answer, it points at a specific year, and it usually points at a lever: a claiming age, a conversion, a spending band, a bond ladder covering a particular stretch.
So we stopped calling it success
We retired success and failure from every screen in Foundry this summer.
Probability of Success is inherited vocabulary. eMoney, MoneyGuidePro, and RightCapital all use it, so it has the feel of a technical term rather than a rhetorical one. It isn't. Paired with a Probability of Failure tile, it turns a distribution of simulated market paths into a pass or fail verdict on a person's life, delivered as a grade. A client who hears "13% chance of failure" is not hearing a counting rule. They are hearing that there is a one in eight chance they end up destitute, which is not what the model said.
The replacements are plain: Plan Confidence for the headline, Shortfall Risk for its complement. Both are ordinary advisor terms. Neither renders a verdict. And the precise definition now sits in a tooltip on the label itself, in the words the engine actually implements:
The share of simulated trials where the portfolio went negative in any year, or ended below the plan's minimum asset level.
The footnote under the number changed too. It used to read "N of M trials ran out of money," which was simply inaccurate, since it also counted trials that merely ended under the minimum without ever hitting zero. It now reads "N of M trials fell short."
None of that is an engine change. The math is identical. It is a copy change, and it is one of the higher leverage things we shipped this quarter, because the words are what the client takes home.
What to say at the table
The version that works is shorter than you would think, and it is mostly just describing what happened.
"We ran your plan through a thousand different market histories. In 870 of them, the money outlasted you. In the other 130, it got tight, and I can show you exactly when and by how much."
Then show them. The year it got tight is a real year with a real age attached, and the gap is a real number. That is a conversation about a decision. "You're at 87%" is a conversation about a grade.
The number is still worth having. It is a good thermometer, as we've written before, and it is genuinely useful for comparing two versions of the same plan against each other. It is just not a forecast, not a verdict, and not the plan. It is a count of how one model behaved under rules somebody chose. Knowing which rules, and being willing to say them out loud, is the difference between a number that informs a decision and a number that ends the conversation.
Foundry's Monte Carlo engine runs the full tax computation inside every trial and shows you the year by year distribution behind the headline, not just the headline. The solver works the other direction, searching for the savings rate, retirement age, or spending level that reaches a confidence target you set. If you want to see what that looks like on a real household, pricing is here.
Related reading: what a financial plan is actually for, cash flow planning versus portfolio modeling, and risk profiles, where we made a similar call about refusing to average away a binding constraint.
Working notes on planning, written for the people who do the work. If there's a planning conversation you'd like covered, support@foundryplanning.com.