Ask a manager why next week's rota changed three times and the answer is almost always some version of "we can't predict how busy we'll be".

It is a good answer. It is also, in the one place anybody has properly measured it, wrong.

What the trial found

Researchers from UC Hastings, the University of Chicago and the University of North Carolina ran a randomised controlled trial with Gap, Inc. - some stores got a package of stability measures, matched stores carried on as normal. It is still the best evidence available on this, because it is an experiment rather than a survey.

Two findings matter to anyone building a rota.

The first is the business result. Stabilising the schedule raised median sales in the treated stores by 7% and labour productivity by 5%. In retail, where a chain will restructure itself for one or two points, seven is an enormous number. The intervention cost about $31,000 to run and the researchers estimated it returned $2.9 million over the 35 weeks it was in the field.

The second is more useful, and it is the one nobody quotes. When the researchers asked store managers what was driving the instability, the managers said customer demand. When they looked at what was actually causing shifts to move, the main causes were inaccurate shipment information, last-minute changes to promotions, and visits from senior leaders.

None of those are customers. All of them are internal, and every one of them was knowable earlier by somebody in the same company.

Why this is good news

If your instability is caused by demand, there is not much to do about it. Demand is genuinely uncertain, and you are stuck absorbing it with short notice and goodwill.

If it is caused by information arriving late from people you can talk to, that is a different problem, and a much cheaper one.

So the useful exercise is not "how do we forecast better". It is: for the last four weeks, why did each shift change? Write the actual reason next to each one. Not the category - the reason.

You are looking for the difference between:

  • "Saturday was busier than expected" - genuine demand variance
  • "The delivery slot moved to Tuesday so we needed two on goods-in" - someone knew on Thursday
  • "Head office confirmed the promotion on the Friday" - someone knew a fortnight ago
  • "The function sheet changed" - the client changed it, but sales knew before the kitchen did
  • "A group booking dropped" - real, but was it visible in the system two days before anyone looked?

In most operations the first category is a minority of the changes. It is just the one that gets remembered, because it is the one nobody could have prevented.

The sector versions of the same problem

In retail, it is the delivery and the promotion. Stock arriving on a different day changes the labour requirement for that day completely, and promotional set-up is a labour event that the person building the rota is often the last to hear about.

In hotels, it is the group booking and the function sheet. A 40-room group confirming late changes housekeeping's morning and the breakfast cover, and both of those rotas were published days ago. The information existed in reservations before it reached the people staffing against it.

In bars and restaurants, it is the large booking, the delivery that did not arrive complete, and the event down the road that everyone found out about from customers.

The pattern is identical: the information existed inside the business before it reached the rota.

What to actually change

Find the one upstream source that causes the most changes and fix the handover, not the forecast. If deliveries move, the person who knows should be telling the person building the rota as a matter of routine, not when it becomes a problem.

Give the rota a cut-off, and make the cut-off mean something. If the schedule is published on Wednesday for the following week, then Wednesday is the deadline for everything upstream of it. A promotion confirmed on Friday goes into the following week's plan.

Log the reason for every change. This costs almost nothing and it is the only way the previous two get taken seriously, because it converts "the rota keeps changing" into a list with names on it.

Then stop apologising for the demand variance. Once the self-inflicted changes are gone, what is left is genuine, it is smaller than you think, and it is much easier to absorb with a bit of deliberate slack than it was when it was buried under everything else.

The uncomfortable part

The trial's finding has an implication that is worth sitting with. If most of your schedule instability originates inside your own organisation, then the cost of it - the turnover, the last-minute cover, the goodwill - is a cost you are choosing, one small late decision at a time.

That is not a comfortable thought. It is a considerably more actionable one than "customers are unpredictable".