"139 days until the deadline" is a fact, not a plan. It tells you the size of the container and nothing about what fits. Planning backwards turns the fact into dates: subtract the phases from the deadline one at a time, and each subtraction produces the date the previous phase must finish - ending with the single most useful date in any project, the real start date, the day work must begin for the deadline to be reachable at all. This guide works a complete example and covers the two judgements that make backwards plans succeed or fail: where the buffer goes, and which unit each phase is counted in.
The worked example: a launch on April 10, 2027
Take a product launch fixed for Saturday, April 10, 2027. From today, November 22, 2026, that is 139 days away - a comfortable-sounding number. The work has three phases: design (21 days), build (60 days) and testing (14 days). Now subtract, in reverse order, using the same month-by-month care as any count on this site. Testing must finish at the launch, so it starts 14 days earlier: April 10 minus 14 days is March 27, 2027, the date the build must be complete. The build needs 60 days before that: March 27 minus 60 is January 26, 2027, the date design must be complete. Design needs 21 days before that: January 26 minus 21 is January 5, 2027 - the real start date.
Now the honest comparison. The phases total 95 days; the span holds 139; the difference, 44 days, is the plan's slack, and today sits 44 days before the real start date. That is a healthy plan - stated in dates, not adjectives. The days until calculator verifies every subtraction here (enter each phase date and check the counts), and the days between calculator on the homepage gives the total span and its breakdown in one step. Notice what the backwards pass bought: three intermediate deadlines and a start date, none of which existed when the plan was just "139 days, three phases".
Put the buffer at the end, in the open
The 44 days of slack should live as one visible block immediately before the deadline - or, in this example, before testing, since launch day itself is fixed by other people. Buffer hidden inside phases, a few padded days in each estimate, gets spent silently by every phase whether it needs it or not, and the plan reports itself on track until the moment it is not. A single end buffer can only be spent once, by whichever phase actually overruns, and its remaining size is a truthful progress report: when the buffer is untouched at the build's midpoint, the plan is genuinely healthy. Size it from the risk, not from optimism - a tenth of the total span is a floor for familiar work, a quarter for work with outside dependencies. In our example the 44 days are nearly a third of the span, which is what a launch with a fixed public date deserves.

Count each phase in the unit it consumes
The classic backwards-planning error is counting every phase in the same unit. Work phases consume working days - nobody builds on Christmas Day - while waiting phases consume calendar days: an approval that takes 10 business days at the other end, a delivery quoted at 14 calendar days, a notice period fixed by law. In the example above the phases were treated as calendar spans for simplicity, but if the 60-day build is really 60 working days, it spans about 84 calendar days, and the real start date jumps from January 5 back into mid-December - before today, which changes the whole conversation. Convert at the boundaries: when a work phase ends and a waiting phase begins, switch units deliberately, and write the unit beside every phase in the plan. Our guide on counting weekdays only gives the conversion method, and the milestone system in our milestone days guide layers checkpoints over the phase dates once they exist.
Finally, the uncomfortable case: if the real start date lands in the past, believe it. The count is telling you the plan cannot be done as specified, months before the deadline would have told you the same thing in public. Shorten a phase, overlap two that do not truly depend on each other, or move the date - those are the only honest moves, and all three are cheaper in November than in April.
Check the plan forwards before you trust it
A backwards plan deserves one forwards pass. Start at the real start date and add the phases back: January 5 plus 21 days of design lands on January 26; plus 60 days of build lands on March 27; plus 14 days of testing lands on April 10 - the deadline, exactly. The forwards pass is not redundant: subtraction and addition slip in different places, particularly across February, and a plan whose two directions disagree by a day has a slipped month in it somewhere. This is the same count-twice discipline used for every figure on this site, applied to plans instead of single spans. Once both directions agree, freeze the phase dates - not the day counts - in the plan document, because dates stay true as time passes while "60 days of build" quietly stops meaning the same thing. Then track against the phase dates: on any given morning, the only question is whether the current phase is ahead of or behind its finish date, which is a question with an answer, unlike "how is the project going".
Where the slack sits matters as much as how much there is. Keep it whole, at the end, just before the deadline - as in the worked example - rather than sprinkling a few days into every phase. Sprinkled slack gets spent invisibly, a little at a time, by each phase assuming the cushion belongs to it; a single block before the deadline can only be spent once, by the phase that actually runs late, and its consumption is visible to everyone. Plans rarely fail for lack of slack. They fail because the slack was spent three times on paper before the work began.
Frequently asked questions
How do I plan backwards from a deadline?
List the phases in order, give each a duration, then subtract them one at a time from the deadline date. Each subtraction gives the date the previous phase must finish. In the worked example, a launch on April 10, 2027 with 14 days of testing, 60 of building and 21 of design produces phase dates of March 27, January 26 and a real start date of January 5, 2027.
What is a real start date?
The date the first phase must begin for the plan to reach the deadline without borrowing time. It is found by subtracting all phases and the buffer from the deadline. Compare it with today: in the example, January 5 is 44 days after November 22, and those 44 days are the plan's true slack.
Where should the buffer go?
At the end, immediately before the deadline, as one visible block - not padded secretly into every phase. End-loaded buffer can be spent by whichever phase overruns; padding inside phases gets spent by all of them whether they need it or not, and hides the plan's real position.
Should phases be counted in calendar days or working days?
In the unit the phase actually consumes. Work phases consume working days; waiting phases - approvals, deliveries, curing, shipping - consume calendar days whether anyone works or not. Mixing them is the classic backwards-planning error: count each phase in its own unit and convert at the boundaries.
What if the real start date is in the past?
Then the plan as specified cannot be done in the time available, and the count has just done you a favour by saying so early. The options are the honest three: shorten phases, overlap phases that do not truly depend on each other, or move the deadline. Pretending the arithmetic away is the fourth option, and it is the one that fails in public.
