Same groceries. Same store. Same time.

Prices up to 23% apart.

That's what Consumer Reports and Groundwork Collaborative found when 437 volunteers ran a coordinated shopping test on Instacart (Groundwork). AI dynamic pricing in retail went from a niche topic to a political one in about four months.

And last week a US state banned part of it.

What's actually happening with AI pricing in retail

First, three terms that get mixed up all the time:

→ dynamic pricing: the price changes over time (demand, stock, time of day) but everyone sees the same price at that moment → price testing: a retailer shows different shoppers different prices to learn what sells → personalised or "surveillance" pricing: your price is set using data about YOU

Now the timeline:

  1. December: the Instacart study. Across four US cities, 74% of grocery items in the test showed up at more than one price. The average gap between lowest and highest was 13%. The researchers estimated a family of four in the pricier group could pay about $1,200 more a year (Groundwork).
  2. Instacart pulled the plug. It ended all item price tests, saying they never used customers' personal, demographic or behavioural data (CBS News). So by its own account this was price testing, not surveillance pricing. Shoppers didn't really care about the difference.
  3. New York made it visible. Since November, businesses using personal data in algorithmic pricing must show "THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA", with fines of up to $1,000 per violation (Jones Day).
  4. Maryland banned it for food. On 28 April the governor signed HB 895. Large grocers and delivery apps can't use personal data to RAISE food prices for an individual, from 1 October. Discounts, loyalty deals and normal supply and demand changes are still allowed (MultiState).

Meanwhile Walmart is rolling digital shelf labels out across its US stores. It says they won't be used for dynamic pricing and that prices are the same for everyone in a given store (IBTimes). (Retailers clearly know how this looks.)

What uni won't teach you about dynamic pricing

Business degrees teach price elasticity and revenue management, usually with airlines and hotels as the friendly example.

What gets skipped:

  1. Fairness is a pricing input now. Maryland's law protects discounts but bans personalised increases. That line is basically a fairness rule written into code.
  2. Disclosure changes behaviour. A label saying your data set the price is a design problem, not just a legal one.
  3. Trust is fragile and fast. Instacart's tests were reported in December and stopped within weeks. Same lesson as media trust, which we covered in our piece on AI slop.

The critical-thinking move: 5 questions for any AI pricing system

Use these on placement, in a case study or in an interview:

  1. Which kind is it? Dynamic, testing or personalised? Name it precisely before judging it.
  2. What data goes in? Stock and weather are one thing. Browsing history and location are another, and now legally riskier.
  3. Can it only go down? Personalised discounts are widely accepted. Personalised increases are what laws are targeting.
  4. Would you put it on the label? If the honest description would scare customers off, that's your answer.
  5. Which rules apply where? New York requires disclosure, Maryland bans some uses. One national algorithm may need several settings.

What to do this month

  1. Read the Groundwork findings. Note the method: how many shoppers, which stores, what counted as a price gap. Method questions impress interviewers.
  2. Run your own mini test. With two friends, check the same product on the same app at the same time. Record what you see.
  3. Map the laws. Make a one-page table of New York and Maryland's rules. Add any state that passes something next.
  4. Connect it to personalisation in marketing. AI is reshaping how shoppers find brands too, as we covered in how AI search is changing marketing.

More on the field in our retail and ecommerce hub.

so TLDR: AI pricing isn't banned, but personalised price RISES are becoming toxic. Learn the difference between dynamic, testing and surveillance pricing.

Would you accept a personalised price if it could only ever be lower than the shelf price?

Sources

Get the next issue by email

theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.