By Matt Wampler, CEO of ClearCOGS
Every brand that runs limited time offers has a version of the same week. The new item drops. The marketing is live. And somewhere in a kitchen, a manager is standing in front of a walk-in trying to decide how much of something nobody has ever sold to prep for tomorrow.
The forecast, if there is one, usually came from a meeting. Someone said it would do about what the last one did. Somebody else said that one underperformed, so add twenty percent. The number went out to the field. Two days later half the system is out and the other half is throwing product away, and the conclusion everyone draws is that new items are just unpredictable.
They are harder. They are not unpredictable. The problem is that most brands are trying to forecast the wrong thing.
The model has nothing to look at
Forecasting an established item is a well-understood exercise. You have years of transactions. You know what Tuesdays look like, what the item does in August versus February, how it moves when it rains. The past is a reasonable guide to next week.
A new item has none of that. On launch day its sales history is an empty column. This is not a restaurant problem, it is a structural one. Researchers studying new product forecasting describe it plainly: for existing products, historical sales give an indicator of future sales, but that data simply does not exist for a new product. Their proposed answer is to build the forecast from the data of comparable products introduced in the past (Baardman, Levin, Perakis, and Singhvi, Production and Operations Management, 2018).
That is the reframe worth sitting with. You are not forecasting the item. You cannot. You are forecasting the shape that items like it have followed before.
The information is in the building, just not in operations
Here is what makes this frustrating rather than merely hard. Most brands already own everything they need to build that comparable.
You have run limited time offers before. You know which ones were supported by a national push and which ones were a menu board change and a social post. You know which ones landed in a heavy season and which ones launched into a slow March. You know the price point, the daypart, the category. You know whether it was a new format of something familiar or something genuinely unlike the rest of the menu.
All of that lives in a marketing calendar. In most organizations, the marketing calendar is not something operations sees until the item is already in the pipeline, and it is almost never something the forecasting system sees at all. The single highest-leverage change many brands can make is boring: give the people building production numbers the promotional calendar in advance, including the ones from last year.
A forecast that knows a promotion is coming behaves very differently from one that discovers it on the second day of the promotion.
A comparable is not just the closest menu item
When operators hear “use a similar item,” they usually reach for the nearest thing on the menu. That is a start, but it misses most of the value.
The useful comparable is the closest past launch, not the closest current item. What you want to know is how a launch of this type behaved: how steep the first week was, whether it kept climbing or spiked and settled, how long the lift lasted, and what happened when the marketing support ended.
And then there is the part almost nobody builds into the launch plan. A new item does not add demand out of nowhere. Some of it comes from guests who would have ordered something else. If the new sandwich pulls from the existing sandwich, your total protein need barely moves, but the mix underneath it shifts hard. Prep the new item up without pulling the old one down and you have simply moved your waste from one container to another.
The question is not only “how much of the new thing will we sell.” It is “what does the new thing do to everything around it.” A brand that has tracked that on three past launches knows the answer within a reasonable range. A brand that has not will rediscover it every time.
The first week is a correction problem
Even with good comparables, the launch forecast will be the least accurate number you produce all year. That is fine, as long as you plan for it.
The first two or three days of real sales are worth more than any pre-launch estimate, because they are the only data that reflects this item, this menu, these guests, right now. The operational question is how fast that signal gets turned into a revised number.
In most brands the answer is a week or more, because the correction runs through a reporting cycle and a meeting. By the time the adjustment lands, the promotion is a third over and the damage is done.
The teams that handle launches well treat the first week differently from the rest of the calendar. They expect to be wrong on day one. They look at actuals daily rather than weekly. They push a revised number to the field mid-week instead of waiting for the period to close. The forecast stabilizes quickly once real sales exist, which means the value is almost entirely in how fast you are willing to update.
Directionally right beats precisely wrong
There is one more thing worth saying, because it determines whether any of this matters.
An operator once described running a prep tool at a large brand and made a point that has stuck with me: nobody follows the number one hundred percent, and they should not. What matters is whether the number is directionally right often enough that people start trusting it, and whether they can adjust when they know something the system does not.
That is especially true on launches. A manager who has worked a store for six years may know that this particular location gets a different crowd for this kind of item. That instinct is real information. The goal is not to replace it. The goal is to give that manager a defensible starting point instead of a blank page, so the judgment gets spent on the exception rather than on the entire number.
When a launch forecast is wildly wrong, managers stop reading it, and the next three launches get built on gut again. Accuracy on new items is not really about the new item. It is about whether the system keeps its credibility for everything else.
Where this leaves you
New menu items will always carry more uncertainty than a Tuesday chicken order. But the gap between brands that handle launches well and brands that do not has less to do with modeling sophistication than with three fairly ordinary habits.
Keep structured records of past launches, including what they did to the rest of the menu. Get the promotional calendar into the hands of the people producing forecasts, before the launch rather than during it. And build a correction loop that moves in days, not weeks.
This is the kind of problem we spend our days on at ClearCOGS: turning the data a restaurant already has, including the promotions it has already run, into a number a manager can act on before the shift starts.
Sources
- Baardman, Lennart, Levin, Igor, Perakis, Georgia, and Singhvi, Divya. Leveraging Comparables for New Product Sales Forecasting. Production and Operations Management, 27(12), 2340–2343. December 2018. journals.sagepub.com
