a p50 date is not a commitment

there's a date at the bottom of most project plans, and it's usually the sum of the "most likely" numbers along the longest chain of work. it's an honest number. it's also, almost by construction, one you'll probably miss.

this post is about why, and about the idea jollygig scheduling is built on: a plan is a spread of possible dates, and the honest thing to do is say so.

what a p50 is

run a project a thousand times, with every activity taking a slightly different amount of time on each run, and write down the date it finishes every time. line the dates up. the p50 is the date half of them beat. the p75 is the date three runs in four beat, and the p95 the date nineteen in twenty beat.

that makes a p50 a fair thing to plan around inside the team. as a promise to someone outside it, it's a coin flip. say "we'll be done by nov 14" when nov 14 is your p50 and you've told them there's an even chance you won't be.

where the single date comes from

most plans add up the most likely durations. but most likely isn't the same as average. work runs over more often than it runs under, and by more, so the most likely value sits to the left of the average. the overruns don't cancel the underruns, they outweigh them. add up a handful of those and you land well short of the middle.

here's a small example from our own engine. six activities in a row, each estimated at 3, 5 and 10 workdays: best plausible, most likely, worst plausible. the most likely numbers add up to 30 workdays. run the plan and the p50 comes out near 33 and the p95 near 38. only about one run in five finishes in 30 days or fewer.

nobody in that plan was careless. the single date was just the wrong summary of what they knew.

what we built

• three numbers per activity, in workdays, on your team's calendar with its holidays and partial days.
• thousands of runs that turn those into a whole distribution of finish dates, with the p50, p75 and p95, and the chance of making a date you've committed to.
• what's carrying the date: how often each activity sat on the critical path and how much it moves the finish, said in words rather than as a score.
• correlation you elicit, because activities run long together. there's a whole post on why independent estimates lie to you.
• actuals that feed back. move a card to in progress or done, the forecast re-runs on what's left, and you can watch the chance of your date move from run to run.

how to use the numbers

there's no right percentile. it depends on what a miss costs. an internal milestone that can slide a week can sit at the p50. a date you give a customer is a better fit for the p75 or the p95, or for a p50 with the chance said out loud: "about even odds by the first week of november, about 95% by early december."

the point isn't to be pessimistic. it's that choosing a percentile is now a decision someone made, instead of an accident of arithmetic.

what it isn't

it isn't a scorecard. variance is a property of the work. there is no per-person view, no estimate-accuracy ranking, and we don't plan to build one. an estimate that turned out wrong tells you something about the plan, not about whoever gave it. if a forecast ever gets used to pressure someone, we've built it wrong.

it isn't a crystal ball. a forecast is only as good as the estimates and dependencies you give it. a wide spread isn't a crisis; it's information, and often the most useful thing the tool tells you. we round the chances we show to the nearest 5% below 90, and never say a flat 0 or 100, because anything finer would be pretending.

and the team doing the work

jollygig began as a team health tool, and the two products share one belief: say how sure you are. it turns out to matter for schedules too. estimates made while a team is strained tend to run optimistic. link a workspace to a team in team health and, when the team is showing signs of strain, jollygig offers to widen the pessimistic end of its estimates by 10%. it's one click, you can undo it, and it never says who or why.