Launching or Cutting a Product: Reading the Effect Before and After
Before you cut or launch a product, size the customers who would leave or switch; after, read the result against a no-change range. A real backtest and Taco Bell reviews.
Cutting a product loses the sales of customers who bought it and do not switch to something else, and launching one adds the sales of customers it brings in, minus what it takes from your other products. Both can be read before the decision, from what customers say about the item and what the customers who buy it did when something similar changed, and both can be checked after, by comparing the result with the range you would have expected if nothing had changed. Dimension Labs gives a multi-location or multi-product business the data to do both: every review, ticket and survey comment labelled by product, joined to who bought what, with the effect measured against a comparison.
This page answers the launch-or-cut decision itself: what will happen to sales, how to read demand from customer conversations, and whether a cut cost you customers. It draws on a real backtest of a hardware company's product launch (the company is unnamed) and on public reviews around Taco Bell's 2020 menu cuts. The general method of grading a past decision is in how to backtest a business decision you already made, and how to read reviews around one release is in how to tell whether a product release changed your reviews.
What will happen to sales if we discontinue a product?
It depends on three groups of the product's buyers, and you can size each one before you decide. Some switch to another item you sell, so their spend stays. Some buy the item as part of a larger order and keep coming for the rest. And some come mainly for that item, so when it goes, their visits go too. Total sales of the item overstate the loss; the third group is the real cost.
Your data already holds the split. Basket data shows what else the item's buyers order and how often the item is their only purchase. Past substitutions, such as a time the item was out of stock at some locations, show what those customers bought instead. And their comments show how attached they are: a guest who writes that an item is "the only reason I come here" is telling you which group they are in. Labelled across every review, survey and ticket, and joined to those customers' orders, those comments become a count of how many customers are in the third group and how much they spend.
Did removing a menu item cost us guests?
Taco Bell is a public case with dated cuts. It began trimming its menu on August 13, 2020, including its potato items, removed the Mexican Pizza from November 5, 2020, announced the potatoes' return on January 14, 2021 for March 11, and brought the Mexican Pizza back on May 19, 2022, after a petition gathered nearly 200,000 signatures. We counted the public Google Play reviews of the Taco Bell app that mention either item in the windows around each date.
Public Google Play reviews, Taco Bell app
Guests named the cut items far more often once they were gone, and the Mexican Pizza's return drew the biggest wave of all
Share of app reviews that mention potatoes or the Mexican Pizza, in windows around each dated cut and return, with 90% ranges.
Before the first cut, 7 of 4,313 reviews named either item (0.16%). In the weeks after the potatoes went and the Mexican Pizza's removal was announced, it was 45 of 3,044 (1.48%), about nine times as often, and the reviews were requests and protests: bring back the potatoes, do not get rid of the Mexican Pizza. While the pizza stayed off the menu, from July 2021 to April 2022, guests still named the items in 0.57% of reviews. When it came back, 97 of 4,128 reviews mentioned it or the potatoes (2.35%), the highest share in the whole period.
Reviews tell you who cared and how much; they do not on their own tell you how many guests stopped coming. Several reviewers wrote that they would not return, and some wrote that they came only for that item, which is the third group above. To turn that into a number, join the comments to each guest's visits: did the guests who wrote about the item visit less after the cut than similar guests who did not, at the same locations, over the same weeks? That comparison is how you answer the question with a range instead of a headline, and it is the same comparison comparing restaurant locations with guest reviews uses across stores.
How do I predict demand for a new product from customer conversations?
Count the requests, then check who is making them. Customers ask for products in reviews, chats, tickets and survey comments long before you build them, and they ask for discontinued ones for years: Taco Bell guests named the missing Mexican Pizza in app reviews through the 18 months it was gone, and its return drew the largest share of mentions in the period we read. That signal becomes a demand estimate when three things are added:
- Who is asking. Join each request to the customer's purchase history. A request from customers who visit weekly is worth more than the same count from lapsed ones.
- What it replaces. Read what those customers buy now. A new item that the same customers would order instead of an existing one adds less than its own sales.
- A past launch to calibrate on. Measure how request volume before your last launch compared with what it sold, and use that ratio, with its range, for the next one.
The third step is what makes the estimate trustworthy, because it is a prediction you can grade.
How do I read the effect of a launch before and after?
Before the launch, set the range you would expect if the launch changed nothing, using only data from before it. After, compare what happened with that range. A real example: for a hardware company's February 2026 product launch, the method predicted 1,591 to 2,410 orders and $3.16M to $4.63M in revenue over the 13 weeks after, if nothing changed. The company booked 2,092 orders and $4.38M, inside both ranges.
A real backtest, customer unnamed
Predicted from data before the launch only, both totals landed inside the no-change range
A hardware company's February 2026 product launch: the range the business would expect over the next 13 weeks if the launch changed nothing (90%), and what happened, as a percentage of the middle of each range.
Both totals held, which tells the company that the launch did not move total orders or revenue by more than normal variation in those 13 weeks. For a launch, that is a precise and useful answer: whatever the new product sold, the business as a whole sold within the range it would have expected without the launch, so the next question is which products it took sales from. The method itself was chosen by a contest: four methods each forecast 11 placebo launch dates on which nothing launched, and carrying the recent past forward had the lowest error.
A real backtest, customer unnamed
The scorecard counts the wide ranges and the misses, not only the hits
All 14 ranges in the launch backtest: two totals and twelve customer segments.
A scorecard should show its weak spots as plainly as its hits. Of the 14 ranges in this backtest, 11 held, but 7 of those held only because they were wide; 4 held with ranges tight enough to test, and 3 missed, all on the high side, in segments that were still growing when the baseline learned. That is what makes the 4 tight hits worth trusting, and it tells the company exactly where to add data before its next launch. The full backtest write-up goes through every segment and the method contest.
The same before-and-after logic works for a cut. Set the no-change range for the locations, customers or product line the cut touches, using a comparison group that kept the item, and read the cut against it. If the result sits inside the range, the cut cost less than normal variation; if it falls below, you have a size for the loss and, from the comments, a reason.
Measured launches and cuts are what a customer world model is built from: what customers said about a product, what they bought next, what the business changed and what else was going on, kept together so the next product decision starts from a range. How to test a business decision before you make it shows that forward step on sample data.
How does Dimension Labs help with a launch or a cut?
Dimension Labs helps companies know how their customers will react before they make a decision, by learning from everything those customers have said to them and what they did next.
For a COO or product leader deciding what to add or drop, the platform does four things today. It is fully omni-channel, reading reviews for every location, app reviews, surveys, tickets and chats together, so every request for a product and every complaint about its absence is counted. It integrates with operational data, joining those comments to orders, baskets and locations by customer; the join to CRM and revenue data and effects with confidence intervals are on Prove ($2,999 a month) and up. It enriches every interaction with Dimensions written for your catalog, such as "asks for a discontinued item", "only comes for this item", "switched to a substitute" or "new item sold out", usable by product, operations and finance; custom Dimensions start on Discover ($999 a month). And its visual analytics layer delivers the read as a deck, dashboard, PDF or executive report, or an interactive data story the team can open product by product.
The platform measures what your past launches and cuts did, product by product, with a range on every effect. That is what the next product decision starts from: if the relationship measured on your last launch holds, it shows which customers a cut would lose and where a launch would add sales rather than move them, and a scorecard like the one above shows how far to trust each part.
Speed matters around a launch, when the first weeks of feedback decide what to fix. Covered California uses Dimension Labs to launch and measure new self-service use cases with structured evidence, and launched a new one in 2 weeks, with a 25% reduction in the resources required to implement it (Covered California story).
When something simpler is the better choice: if you sell a handful of products in one channel, a sales report by item and a read of recent reviews may answer a cut. Dimension Labs earns its place when products sell across many locations or channels, comments arrive in several places, and you want the cost of a cut or the lift of a launch measured against what would have happened anyway.
FAQ
How do I know whether discontinuing a product will hurt sales?
Size the customers who buy mainly that item: check how often it is their only purchase, what they bought when it was out of stock, and what they say about it. Then compare their visits after a cut with similar customers at locations that kept it.
Can customer reviews predict demand for a new product?
They show where demand is and how strong it feels, and they become a forecast when joined to who is asking and calibrated on a past launch. Taco Bell's Mexican Pizza was named in app reviews throughout its 18-month absence, and its return drew the highest share of mentions in the period.
How do I tell whether a launch worked?
Compare the result with a no-change range built only from data before the launch. In a real backtest, a hardware company's orders (2,092) and revenue ($4.38M) both fell inside the ranges predicted beforehand.
Is Dimension Labs suited to a single product decision?
Yes, when the product sells across many customers, locations or channels and the feedback is spread across reviews, surveys and tickets. One past launch or cut, with its date and the orders around it, is enough to start, and every later decision is measured the same way.
Bring the reviews or tickets from around your last launch or cut, with its date, and start with one file to see what customers said and did. Every plan from Explore to Scale starts with a 14-day free trial.