How forecasting replaces estimation

If Metriq never asks you to estimate, how does it tell you when work will be done? The answer is forecasting: a probabilistic projection built from your team's real delivery history. This page explains the mechanism — what data goes in, what the simulation does with it, and how to read the number that comes out.

The raw material: your throughput

As your team finishes work, Metriq counts what actually comes off the board — your throughput, the amount of work you really complete. It isn't an opinion about how hard the work looked; it's how much your team genuinely gets done.

Do that for a while and you don't have one throughput number, you have a distribution: stretches where work flew, stretches where it stalled. That spread is the honest signature of how your team delivers — including the interruptions, the reviews, the blocked patches. Crucially, none of it was guessed. It was observed.

(Metriq also measures each card's cycle time — how long it took from start to done — but that feeds the cycle-time scatterplot, the SLE target line, and aging-WIP heat. The forecast itself samples your throughput history, not individual cycle times.)

The engine: Monte Carlo simulation

A single average would be a poor forecast, because real work isn't average — it's variable, and the variability is the whole story. So Metriq uses Monte Carlo simulation instead.

The idea is simpler than the name. To forecast the next ten cards, Metriq does this thousands of times over:

Draw a random sample from your throughput history — a chunk of the work your team has really completed.
Knock that off the ten cards you're forecasting, then draw again, and again, until the pile is empty. How much delivery that took is one possible future.
Throw it away and do it again. And again. Thousands of times.

Each run is a plausible future assembled entirely from throughput your team has actually delivered. Metriq turns each one into a projected finish date; together they give you not a single answer but a full picture of how the next ten cards might go — the lucky fast futures, the unlucky slow ones, and everything between. Because the simulation samples from your real spread of throughput, it naturally captures your team's variability instead of pretending it away.

The forecast needs a baseline of completed history before it can run; until then Metriq shows nothing rather than a number built on too little data.

Reading the result: what p85 means

Monte Carlo gives you a range of outcomes, so Metriq reports a forecast as a probability rather than a date carved in stone. The headline number you'll see most is p85.

In plain language: p85 is the result you hit or beat 85% of the time. If the p85 forecast for finishing your ten cards is "two weeks," it means that in 85 out of every 100 simulated futures, the work was done within two weeks. There's still a 15% tail where things ran longer — and Metriq shows you that honestly rather than hiding it.

Why 85 and not 100? Because a 100% forecast would have to include the single worst run the simulation ever produced — a useless, paralysingly conservative number. p85 is a deliberate, well-calibrated bet: confident enough to plan around, honest enough to admit the tail exists. You can think of it the way you read a weather forecast — "85% chance of being done by Thursday" is a far more useful planning statement than a flat "Thursday."

Why this beats a guess

A traditional estimate is a forecast with a sample size of zero — one person's intuition, stated as a fact. A Monte Carlo forecast is a forecast with a sample size of thousands, every one of them drawn from work your team really did. When reality changes — the team speeds up, or hits a rough patch — the forecast moves on its own, because the underlying throughput data moved. You never re-estimate. You just keep finishing work, and the forecast keeps recalibrating itself against the truth.

That's the whole trade. You give up the comfort of a made-up number. In return you get a forecast that is built from evidence, carries its own uncertainty, and updates itself for free.