What a Constraint Costs to Keep
Why every constraint has a price even when it is free to obey, how to work out what one costs, and what to do with the number once you have it.
What one site found when it priced four constraints it had never questioned
None of the four was a rule anybody could source, and together they cost more than the vacancy the site had been told it could not fill. This is one employer's own reckoning over a year, not a benchmark and not a recommendation about supervision.
Every constraint costs something to keep, including the ones that cost nothing to obey. The price is not the effort of following the rule; it is what the rule removes from the rota — the shifts that cannot be built, the people who cannot be used, the cover that has to be bought instead.
The test in “What a Constraint Costs to Keep” needs an unchanged baseline and a trustworthy record of what happened after the change. Teams exploring a transparent getting teams to meet deadlines process for getting teams to meet deadlines can strengthen that comparison while preserving corrections, exceptions and the decision owner’s explanation.
That price is almost never calculated, because the constraint arrives before the rota and feels like part of the terrain rather than part of the design.
As a separate benchmark for “What a Constraint Costs to Keep”, see Chicago Fair Workweek information. Keep the source rule, operational interpretation and scheduling decision distinct.
The three forms the price takes
Money, in overtime, agency cover or premium rates paid to work around the rule. Hours, in people doing something less useful than they could. And optionality, which is the least visible and often the largest: the arrangements that cannot be tried because the constraint forecloses them.
- Money: overtime, agency cover, premium rates and anything else bought to work around the rule.
- Hours: people doing something less useful than they could, including the time spent building the rota around it.
- Optionality: the arrangements that cannot be tried at all while the constraint stands.
Most constraint pricing stops at the first form, which is why the numbers come out implausibly small and the exercise gets abandoned.
Pricing one without a model
Take the constraint, remove it on paper for one representative fortnight, and rebuild that fortnight. Then compare the two versions on whatever the organisation actually pays for: filled shifts, overtime hours, agency shifts, premium hours, unfilled gaps.
The difference is the price. It is rough, it is specific to that fortnight, and it is far more useful than an estimate, because somebody can argue with it.
Why a representative fortnight beats a year
A year of data invites averaging, and averaging hides the thing you are looking for. Constraints bite unevenly: a rule about weekend rotation costs nothing in a quiet month and everything in the month with two sickness absences.
Pick the fortnight that actually hurt, price the constraint there, and say so. "In the worst fortnight of the quarter, this rule cost nine unfilled shifts" is a sentence somebody can act on.
What to do about the constraints that are free
Some constraints cost nothing: they bind, and the rota would have been built that way regardless. Price them anyway, briefly, and mark them zero.
A zero is a finding. It tells the next person not to spend time there, and it stops the constraint being used as an explanation for an outcome it did not cause.
The price of an assumption
An assumed constraint has a second cost on top of whatever it removes from the rota: nobody can weigh it, so it is never traded against anything.
That means assumptions accumulate. Each one is individually cheap and the set of them is not, and the set is what produces the familiar situation where a manager cannot name a single thing that would make the rota better.
Who should produce the number
Whoever builds the rota, not finance. The person with the spreadsheet can rebuild a fortnight in an afternoon; a finance partner cannot, because the constraint is operational and the rebuilding requires knowing who can cover what.
Finance should check the rates afterwards. Getting the order the wrong way round turns a two-hour exercise into a three-month project that produces an average.
What the number is for
Not for proving the constraint wrong. For putting it in a queue. Four constraints priced at £18,000, £6,000, £1,700 and zero tell a manager exactly where to spend the one conversation they are going to get this quarter.
Without the prices, that conversation goes to whichever constraint somebody complained about most recently, which is uncorrelated with what it costs.
Saying it out loud to the people affected
Pricing a constraint that protects people reads badly if the number appears without context, and some of the most expensive constraints are there to protect somebody.
Say what the price is and say what it buys, in the same sentence. "This costs about £18,000 a year and it is what keeps a supervisor available on every late shift" is a defensible position. Half of it is not.
The constraint that pays for itself
Occasionally the arithmetic runs the other way: a constraint that looks expensive turns out to prevent something costlier, and the rebuilt fortnight is worse rather than better.
That result is worth recording as carefully as the others, because an untested constraint that survives a test becomes a different kind of object. It can now be quoted with a number attached, and nobody has to have the argument again next year.
What usually comes out of it
Two or three constraints turn out to be carrying most of the cost, and at least one of those has no source. That combination — expensive and unattributed — is the one worth taking to whoever can decide.
Everything else goes on the list with its number, and the list is what makes next year's version of this exercise take an hour instead of a week.