6 Pricing Assumptions That Cost Commercial Leaders Millions

Magdalena Okrzeja
Magdalena Okrzeja
November 17, 2026
•
8 min read
Loading the Elevenlabs Text to Speech AudioNative Player...

This article is for commercial leaders in retail and CPG who own pricing decisions across large assortments. It tests six common pricing assumptions against published research, explains why each one quietly costs margin, and ends with a short demo of how DS Stream sets up a SKU-level pricing campaign with guardrails. Leading retailers are rethinking these assumptions, and the numbers below show why.

Assumption 1: Lower prices always drive growth

Not necessarily. A lower price can lift volume while it erodes margin, and the extra volume needed to compensate is usually larger than teams expect. In the 1992 Harvard Business Review article Managing Price, Gaining Profit, Marn and Rosiello showed that a 1% improvement in price, with no loss of volume, increased operating profit by 11.1%, while improving unit volume by 1% yielded a 3.3% increase. The arithmetic runs in reverse too: in their example, a 1% price decrease would destroy 11.1% of operating profit dollars.

Promotions follow the same pattern at scale. In 2014, FoodNavigator-USA reported on a Nielsen review of 39 million US promotional events that found almost three quarters of promotions run by CPG companies failing to break even.

So the useful question is narrower: where are discounts actually worth it? Some SKUs respond to a price cut with enough extra volume to pay for it. Many do not, and the only reliable way to tell them apart is to look at each product on its own. Our article on AI markdown optimization in retail applies the same question to end-of-season discounts.

Assumption 2: Products in the same category should follow the same pricing strategy

Customers buy products. The category is a planning unit that lives on the retailer's side of the shelf. Two SKUs in the same category can have very different demand, margin and inventory dynamics, and one shared set of pricing rules tends to hide those differences.

Key value items make this concrete. McKinsey defines KVIs as "those product categories and SKUs whose prices consumers tend to notice and remember," and wrote in 2016 that, ideally, KVIs will account for 15 to 25 percent of sales in the category. The remaining products can play a different role. In 2017, McKinsey added that key-value categories can account for up to 80 percent of an average retailer's revenue but only half of its profit.

A shopper who checks the price of milk every week reacts differently from one who buys a specialty cheese twice a year. Price both with the same rule and at least one of them ends up mispriced. Grouping products by the job they do, for example separating traffic drivers from margin drivers and flagging items with inventory risk, is the first step. Our product segmentation demo shows how that grouping works in practice.

Assumption 3: More data means better decisions

Most retailers already have enough data. Transaction history, stock positions, promotion calendars and loyalty records are sitting in their systems today. The harder problem is understanding which signals matter for each SKU.

McKinsey estimated in 2014 that up to 30 percent of the thousands of pricing decisions companies make every year fail to deliver the best price. In a 2022 article on grocery analytics, McKinsey also noted that out of more than 100 use cases for advanced analytics, just ten account for 80 percent of the value at stake. Focus pays better than volume.

For pricing, focus means a model that can tell you, product by product, whether elasticity, competitor moves, stock age or seasonality is driving the result. A new data source earns its place only if it changes one of those answers. More on that shift in our guide to moving from pricing rules to SKU-level AI.

Assumption 4: You can't optimize new products without sales history

A new SKU has no sales history, so many teams price it by analogy and instinct, then wait for data to arrive. The wait often rests on a gloomy belief that most launches fail anyway, a claim usually put at 80% or higher. Castellion and Markham called that figure a persistent myth in the Journal of Product Innovation Management (2013), pointing to empirical studies since 1977 that put the new product failure rate at 40% or less. Launches survive more often than the folklore suggests, which makes their first price worth getting right.

AI can support that first price before sales data exists by drawing on product similarity and on how the category behaves. The clearest published example comes from online retailer Rue La La. In Manufacturing & Service Operations Management (2016), Ferreira, Lee and Simchi-Levi describe a business pricing and predicting demand for "products that it has never sold before, which account for the majority of sales and revenue." Using machine learning for demand forecasting combined with price optimization, they estimated a revenue increase of approximately 9.7% in the test group, with a 90% confidence interval of 2.3% to 17.8%.

Our demo on pricing new products without sales history shows how product attributes and comparable items feed a starting price that the team can then refine.

Assumption 5: Pricing recommendations should be fully automated

The best pricing systems help teams make better decisions within business rules and guardrails. Full automation sounds efficient, but it cuts out the people who know about a supplier change, a competitor launch or a brand rule that never made it into the data.

The research supports a middle path, and it cuts both ways. In a field experiment published in Management Science (2020), Kesavan and Kushwaha found that merchants' overrides of a data-driven tool reduced profitability by 5.77%. The same study found that merchants increased profitability for growth-stage products and decreased it for mature and decline-stage products. Human judgment added value in specific situations and cost money in others.

That is an argument for guardrails and selective overrides, with people stepping in where they know something the model cannot see. BCG reported in 2024 that retailers that made the transition to AI-powered pricing increased gross profit by 5% to 10%. In the setup BCG describes, merchants still give final review and approval before prices are executed.

Assumption 6: One pricing strategy fits 50,000 SKUs

Picture a multi-category retailer with 50,000 products. That catalog holds 50,000 demand patterns, and each one needs its own pricing decision. The number is an illustration, but even a single grocery store is large: according to FMI, the average US supermarket carried 33,248 items in 2025.

Uniform rules leave money on the table. In an NBER working paper first released in 2017 and later published in the Quarterly Journal of Economics, DellaVigna and Gentzkow found that most US food, drug and mass-merchandise chains charge nearly uniform prices across stores, even though demand estimates reveal substantial within-chain variation in price elasticities. By their estimate, the median chain sacrifices $16 million of annual profit relative to a benchmark of optimal prices.

No category team can review 50,000 prices by hand every season. Pricing at that scale needs a system that proposes a price per SKU inside the limits the business sets, and a team that spends its time on the exceptions. We covered this approach in the Pricing at Scale webinar on protecting margin across 50,000+ SKUs.

See it in practice: setting up a pricing campaign

In the demo, Magdalena Okrzeja, AI Transformation Advisor at DS Stream, walks through how a multi-category retailer sets up a pricing campaign in the DS Stream price optimization solution. The 9-minute video starts from a familiar situation. Category managers adjust prices by hand across 15 spreadsheets, each chasing a different goal, while leadership hesitates to automate out of fear of a runaway model that might discount flagship styles by 40%.

Watch on YouTube: Price Optimization for Smarter Revenue Growth

The setup takes seven steps. The team names the campaign and sets its scope, then picks an optimization objective from five goals that compete with each other, such as inventory reduction or revenue growth. Numeric expectations come next, for instance a minimum margin. Global overrides define what the system must never cross; the demo uses a minimum profit margin of 5%. The team then chooses the range of historical data the prices are based on, such as the last six months. A/B testing lets the team exclude groups of products, price them manually and compare the results with the system's recommendations over time. A final summary shows the campaign and its safeguards before anything runs.

Settings sit on three levels. The strategy sets the direction, numeric expectations narrow it down, and global overrides hold whatever the other two say. That structure is the practical answer to assumption 5: the model proposes prices, and people decide the limits it works within.

Your pricing problem may not be price

All six assumptions treat pricing as one lever, pulled the same way everywhere. In most portfolios the biggest commercial opportunity is more specific. It might be margin given away through discounts that never paid back, or a segmentation that prices traffic drivers and margin drivers by the same rule. Where is it in yours?

To see how this works on a real assortment, explore DS Stream's AI pricing optimization solution, or talk to a DS Stream AI Advisory Strategist.

FAQ

Do price cuts ever pay off?

Yes, for some SKUs. The bar is high, though. In the 1992 Harvard Business Review analysis, a 1% price cut destroyed 11.1% of operating profit, while a 1% volume gain added only 3.3%. A SKU-level model shows which products respond strongly enough to justify a discount.

How can AI price a product with no sales history?

It estimates demand from similar products and from how the category behaves. In the Rue La La field study, machine learning applied to products the retailer had never sold before produced an estimated 9.7% revenue increase in the test group.

Should pricing decisions be fully automated?

Usually not. Merchant overrides cut profitability by 5.77% in the Kesavan and Kushwaha field experiment, yet improved it for growth-stage products. Guardrails combined with selective human review keep control with the team without throwing away the model's gains.

Share this post
AI in Retail
Magdalena Okrzeja
MORE POSTS BY THIS AUTHOR
Magdalena Okrzeja

Curious how we can support your business?

TALK TO US

More insights

More news

No items found.