Skip to main content
Grocery store aisle with price tags

January 6, 2026

Your Grocery Store Knows
What You'll Pay

How algorithmic surveillance pricing is charging different customers different amounts for the same products.

The Experiment

Investigators put a room full of people together. Everyone ordered the same 20 items to the same address. Each from their personal phone. Instacart gave everyone different prices.

3different prices for the same cereal
20items, all priced differently per person

After five months of investigation, More Perfect Union uncovered something bigger than one app. They found a system. One that grocery stores and tech companies built together. It learns from every purchase. And its primary goal is to maximize profit.

After being caught, Instacart denied using dynamic pricing. Then quietly announced they'd stop adjusting prices based on data.

The System Behind the Prices

80%of retailers now use some form of dynamic pricing
$1.3Tin retail revenue influenced by algorithmic pricing
15-25%price variance on identical items
Real timeadjustments based on your behavior

This isn't about supply and demand. It's about data extraction. Your purchase history. Your location. Your device. Your browsing behavior. All of it feeds an algorithm designed to find the maximum price you'll pay.

The FTC has launched an investigation into "surveillance pricing," the practice of using personal data to set individualized prices. But while regulators study the problem, the systems are already deployed.

It's Not Just Online

In-Store Surveillance

Some grocery stores have cameras everywhere. They track you through the store using facial recognition. They know your past purchases. They watch what you look at and how long. They assess how you're dressed, your physique, your apparent state of health.

AI instantly assesses all of it. In real time.

Electronic shelf labels can change prices as you approach them. The same item can cost different amounts depending on when you shop, how you look, and what the system predicts about your willingness to pay.

Security camera in retail store

The Pattern Is Everywhere

Airlines pioneered it. Hotels refined it. Concert tickets normalized it. Now it's coming for everything else.

Airlines

Prices change by the minute based on search history, device type, and perceived urgency.

Hotels

Room rates adjust based on local events, your loyalty status, and how many times you've viewed the page.

Concerts & Sports

"I have my kids buy tickets. They try to charge me double for the same seats."

E-commerce

Add items to your cart but don't buy? Watch the price drop in your email. Or rise if you seem committed.

If people want to stop this, we need to return to a primarily cash-based society. Sadly, too many people survive on credit.

Person paying at checkout

Once everything has to be purchased electronically, they have complete control over what we buy and how much we pay for it.

The problem isn't the algorithm.
The problem is the intent.

The Trust Problem

Personalization and dynamic pricing aren't inherently wrong. Airlines selling last-minute seats cheaper than they'd otherwise go empty? That's efficient. Offering loyal customers better rates? That's fair.

But using data to find the maximum each person will pay? To exploit information asymmetry for extraction? That's different. That breaks trust. And trust, once broken, doesn't come back easily.

People are already adapting. They're using VPNs. Clearing cookies. Having their kids buy tickets. Shopping in incognito mode. Every workaround is a signal that the relationship between business and customer is broken.

Corporations will stop at nothing to get a profit. This isn't a failure of regulation. It's a failure of design ethics.

Person reviewing receipts

Every workaround customers develop is a signal that the system was designed against them.

What This Means for Product Design

The same data that enables surveillance pricing could enable something else: transparent, fair, trust-building commerce. The choice is in the design.

Companies that use data to serve customers (not extract from them) will have a structural advantage as awareness grows. Trust is becoming a differentiator.

We believe:

  • 1Data should serve users, not exploit them. Personalization is powerful. Extraction is corrosive.
  • 2Pricing should be explainable. If you can't justify a price to a customer's face, you shouldn't charge it.
  • 3Trust compounds. Short-term extraction destroys long term relationships. Design for the long game.
  • 4Transparency is a feature. The companies that show their work will win as customers become more aware.

What We Build

Transparent Commerce

E-commerce and retail platforms where pricing logic is clear and fair.

Loyalty Systems

Programs that reward customers without exploiting their data.

Subscription Platforms

Billing systems designed for clarity, not confusion.

Data-Ethical Products

Applications that use personalization to serve, not extract.

Are you designing for trust?

Or optimizing for extraction?

Share this perspective

Donny Smith

Written by · January 6, 2026

· Founder & Executive Creative Director, Bttr.

Over the past 15 years, he has led creative teams and contributed to products used by millions of people worldwide, working with companies including Apple, Opendoor, JP Morgan, GE Aerospace, Pepsi, and Alterra Mountain Company.

LinkedIn

Investigation and reporting from More Perfect Union via Wall Street Apes, January 2026. The FTC is actively investigating surveillance pricing practices including Instacart's use of the Eversight pricing tool.

More Insights

Perspectives on product design, technology, and the forces shaping how people interact with the things we build.

What Is Temperature in AI
7 min read

What Is Temperature in AI

Temperature is a setting that controls how random a language model’s output is. At each step the model scores every word it knows, turns those scores into probabilities with a softmax, and temperature is applied inside that step: it divides each score before the probabilities are worked out, which sharpens or flattens the gap between the leading word and the rest. A Hugging Face guide to text generation describes the move as making the distribution sharper by increasing the likelihood of high probability words and decreasing the likelihood of low probability ones, and notes that as temperature approaches zero the sampling becomes equal to greedy decoding, where the model takes its single highest scoring word every time. OpenAI’s API documentation puts the range between 0 and 2, says higher values like 0.8 make the output more random while lower values like 0.2 make it more focused and deterministic, and recommends altering temperature or top_p but not both. The catch most teams miss is that a low number is not a lock. Anthropic states in its glossary that even with temperature set to 0 the results will not be fully deterministic, and identical inputs might produce different outputs across API calls. The operator rule follows: run cold when the work has one right answer, run warm when you want range, and never read a warmer setting as a smarter model.

What Is a Multimodal Model
7 min read

What Is a Multimodal Model

A multimodal model is an AI system that takes in and reasons across more than one kind of data, such as text, images, audio, and video, inside a single context. It maps each input into one shared representation, so it can relate a word, a picture, and a sound as parts of the same thing. The distinction that matters is how it is built. When Google launched Gemini in December 2023, it noted that the standard approach trained separate components for each modality and then stitched them together, and said Gemini was instead pre-trained from the start on different modalities. OpenAI made the same move with GPT-4o in May 2024, a single model trained end to end across text, vision, and audio so that every input and output runs through one neural network, which is why it answers audio in about 320 milliseconds rather than the seconds of lag in a stitched voice pipeline. The operator rule follows: a native multimodal model keeps every sense in the same context, while a vision tool bolted onto a text model reasons only over what it already turned into words.