What Is Monte Carlo Forecasting?

Ask a team when something will be done and you usually get a single date. It sounds confident. It is almost always wrong. Monte Carlo forecasting replaces that brittle guess with something far more useful: a range of outcomes, each with a probability attached.

The problem with a single date

Engineering estimates miss their targets by 30–50%. The reason isn't that teams are bad at guessing — it's that a single date can't carry the one thing a stakeholder actually needs to know: how likely is it? "The 14th" tells you nothing about risk. "80% chance by the 18th" tells you everything.

How a Monte Carlo forecast works

A Monte Carlo forecast doesn't try to be clever about the future. It simply replays the past, thousands of times:

1. Take your real throughput. Count how many items your team finished each week over its recent history.
2. Simulate the future. For each of a thousand runs, randomly draw weekly throughput from that history and accumulate it until the work is done.
3. Read the spread. Sort those thousand outcomes and look at the percentiles — that's your forecast.

The only input is your own history. No story points, no planning poker, no optimism. It's the same technique used in finance and engineering risk analysis, pointed at delivery.

From a date to a range: P50, P85, P95

The output is a distribution, usually expressed as percentiles. P50 is the median — a coin flip, risky to promise. P85 is the commitment most teams should actually make. P95 is for when failure isn't an option. If a stakeholder wants the P50 date but it only carries a 50% chance, you now know that before you commit — not at the retro afterwards.

Why it beats story-point estimates

Story points feel precise and deliver false confidence. They compress every unknown into one negotiated number, and that number can be argued down under pressure. A Monte Carlo forecast can't be talked into optimism — it only knows what your team has actually done. You spend your planning meetings deciding what to do about the odds instead of arguing about the estimate.

The part most forecasts miss: behaviour

Here's the catch. A Monte Carlo forecast tells you the range. It does not tell you whether you'll land inside it — because the future isn't only a function of throughput. It's a function of behaviour.

Outcome = Capability × Behaviour. Your throughput history captures capability. But the moment focus fragments, work-in-progress balloons, or quiet dependencies stack up, your real delivery rate drifts away from the history the forecast was built on — and the range quietly stops being true. The forecast gives you the odds; behavioural signals tell you whether the odds still hold. That's the argument at the heart of The First Red.

Doing it on your real boards

Running a forecast once by hand is educational. Running it every week, across every board, is not realistic — so it's worth letting software read your history for you. If your team works in monday.com, IMIRT's Delivery Intelligence runs Monte Carlo forecasting on your board automatically, returning the P50/P85/P95 answer to how long, when, how much and how likely — and watching the behavioural signals that decide whether you'll hit the range. It's the first probabilistic forecasting app on the monday.com marketplace.

Common questions