Personalize the pitch. Keep the price public

Here is a discount flow plenty of brands will run this month without a second thought. A subscriber who browsed boots twice, opened three emails and sits in a high predicted lifetime value segment gets a 10% code. Another subscriber, who browsed once, lives in a cheaper ZIP code and has not bought in a year, gets 25%. Nobody set those numbers by hand. A model picked the smallest discount it thought each person needed, and everybody on the marketing team called it personalization.

On November 12 the New York City Council voted 44 to 5 to call it something else. Introduction 891-A, sponsored by Speaker Julie Menin, bans what the bill calls surveillance pricing: pricing set completely or in part by an algorithm that uses personal data to offer different prices to different customers for the same goods or services. Mayor Zohran Mamdani said publicly that his administration supports it. The Council’s legislation page lists it as waiting on his desk for signature, and the text takes effect one year after it becomes law.

So nothing changes at checkout this Black Friday. What changes is the question your offer engine has to be able to answer. And the cheapest week to find out whether it can is the week every offer you run is sitting in front of you.

What the bill actually says

Most of the coverage stopped at “New York bans personalized prices.” The text is more useful than that, because the Council rewrote it after pushback. At a June hearing the Retail Council of New York State and the Hotel Association of New York City warned the first draft could interfere with loyalty rewards, and DoorDash submitted testimony that it would block targeted restaurant promotions. According to amNewYork, the original two-page bill grew to ten pages, mostly spelling out which discounts survive.

Here is the shape of it.

It reaches far. The ban covers any entity domiciled in the city or transacting with consumers there. Read plainly, that is an online store in Ohio shipping a jacket to Brooklyn, not just shops on Broadway. Small businesses with fewer than 15 locations that mostly trade from physical premises in the city get a warning on a first violation and 30 days to fix it, restitution included.

The core rule is short. Nobody may set or adjust the regular price, or the price after ordinary sales and coupons, using surveillance pricing. Nobody may advertise an offer built that way. And nobody may collect, use, sell or share personal data to make it possible.

Personal data is defined broadly: any data that identifies or could reasonably be linked, directly or indirectly, with a specific consumer or device. That last word matters. Device level identifiers are what most of our targeting runs on.

Ordinary promotions are fine. The bill names flash sales, end of season sales, early bird offers, price matching and rebates, as long as the discount is generally available without using personal data.

Then come the custom discounts, the part every CRM team should read twice. There are four lanes. A loyalty or rewards discount, where signing up for a mailing list counts as joining, offered to all members under publicly disclosed terms, with tiers based on purchase history allowed if the tiers are disclosed. A discount for a group the customer tells you they belong to, such as veterans, seniors, teachers or birthdays, based on information the customer gave knowingly and “not derived or inferred” from anything else. A discount based solely on the customer’s purchase history with you: what they bought, when, how often, first order, returns. And a set of narrower rules for online marketplaces.

Two conditions sit on top of those lanes. A custom discount has to be offered “uniformly to any consumer who meets the disclosed eligibility conditions or criteria.” And anyone offering one has to keep procedures that let a consumer ask what personal data was considered and how, and hand the records to the city’s Department of Consumer and Worker Protection when it asks.

Penalties run up to $1,000 per violation, with each consumer or transaction counted separately, plus restitution. There is no private right to sue. The department enforces, and the Council’s fiscal estimate, as amNewYork reports it, pays for four extra staff, two of them investigators.

Where the line falls for marketers

Strip away the legal drafting and the bill separates two things marketing has spent a decade blending together: who hears about an offer, and what the offer is.

The first stays yours. The text says nothing prevents a business from marketing a bona fide discount to win a new customer or bring a lapsed one back, unless that person opted out of the messages. You can still use everything you know to decide who gets the email, the push or the ad.

The second is where the law bites. The safe lanes run on narrow data, and the tripwire is a single phrase. Purchase history qualifies only if it is “not paired, combined, or cross-referenced” with any other personal data except the fact that someone joined your program. That is a problem, because combining is the whole point of a modern customer data stack. Browse behavior, email engagement, predicted lifetime value, third party enrichment and ad platform signals all go into one score. Use that score to set the depth of a discount, and as I read the text, you have left the lane and walked into the definition of surveillance pricing.

“15% off your first order” for anyone who has never bought: fine, with disclosure. “The model chooses 10, 15 or 25 based on everything we know about you”: that is the practice the bill was written to stop.

Charcoal and red chalk sketch of a shopkeeper measuring a customer's coat sleeve with a tailor's tape while an assistant behind them leans in to look at the customer's open wallet.

The uniformity clause catches the cleverest margin trick in the playbook too, which is sending the deep code only to people the model thinks will not buy without it. If your disclosed rule is “lapsed for a year gets 20%,” every lapsed customer gets 20%, not just the ones a propensity score flags as needing a push. Offer depth testing becomes an open question. The bill does not address randomized tests, where two eligible customers see different codes, so that is one to take to counsel rather than guess at.

And look at who carries the risk. A service provider acting for a business is exempt unless it shows reckless disregard of the business’s conduct. That means your email platform, your offer optimization vendor and probably your agency are mostly outside the blast radius. You are not. “The AI picked the number” moves no liability anywhere.

It is not one city

If this were a single council bill, you could file it under 2027 and move on. It is not.

In Seattle, Mayor Katie Wilson signed an ordinance on October 5 banning large grocers and delivery services from algorithmic price discrimination, effective September 1, 2027, with the city attorney enforcing and a private right of action for shoppers, according to The Seattle Times.

New York State already has a disclosure law. Its Algorithmic Pricing Disclosure Act took effect on November 10, 2025, and requires a clear notice near the price saying it was set by an algorithm using your personal data, with penalties of $1,000 per violation. When the law kicked in, the Attorney General’s consumer alert said this kind of pricing shows up most often in apps and loyalty programs, as individualized discounts. In January she demanded answers from Instacart after a Groundwork Collaborative and Consumer Reports study found some shoppers seeing prices up to 23% higher for the same item, in the same store, at the same time. Instacart ended its item price tests, but suggested its retail and brand partners could keep testing promotions and discounts. Notice where the activity went. The discount is the escape hatch, and the New York City text is aimed squarely at it.

The state legislature also passed a full ban, the One Fair Price Act, in June. It is still unsigned on Governor Kathy Hochul’s desk, according to amNewYork and The Record. George Slover of the Center for Democracy and Technology told The Record that New Jersey, Connecticut and Maryland already have similar bans and that more states will see bills when sessions open next year.

Federally, the FTC on October 28 ordered Delta, United, JetBlue, Lyft, Uber, DoorDash, Instacart, Target and Walmart to explain whether their prices vary person to person based on loyalty program participation or personal data. They have 45 days to answer.

Grace Gedye of Consumer Reports, who worked on the city bill, made the point that matters most for anyone running a national program: businesses that have to change for New York City “may decide to do it for the rest of the country at the same time.” Nobody builds a separate discount model for five boroughs. The unit of compliance is not the city. It is your offer engine.

Why it belongs on this week’s list

The city ban will not bite for at least a year. Three things make it a Black Friday item anyway.

The state disclosure law applies now. If any price a New York shopper sees was set by an algorithm using their personal data, the label requirement is live today, in the middle of your biggest sales week.

This is also the one week of the year when your offer logic is fully visible. Every code, every flow, every tier is configured and running. In March nobody will remember why the winback flow sends 30% to some people and 15% to others.

And January is when next year’s CRM, loyalty and vendor contracts get signed. Building a 2027 discount architecture you will have to tear out before it pays back is the expensive path.

Charcoal and red chalk sketch of a shop owner pressing a large blank sheet against the inside of a shop window while people on the sidewalk outside stop to look at it.

What to do before the sale

Run an offer census. List every holiday discount you will run and put each in one of three columns: public, meaning the same for everyone who sees it; custom in a lane, meaning loyalty, a self-identified group or purchase history alone; or model chosen. The third column is your problem list. Most teams have never written it down.

Check the New York label today. If anything in the third column touches New York shoppers, ask whether the state disclosure applies and whether it is showing near the price. That question has a live answer this week.

Split the audience decision from the amount decision. Keep every signal you have for choosing who gets the message. For the discount itself, make the rule read purchase history and program membership only. Pull browse data, engagement, enrichment and predicted scores out of the number.

Write the rules where customers can read them. First order offers, winback offers and loyalty tiers should each have a one-line rule in your program terms. “Spend over $500 a year and get 20% off” is a rule. “Our VIPs get surprise offers” does not tell anyone how to qualify.

Find the “only if needed” logic. Any flow that withholds a discount from eligible customers because a model predicts they would buy anyway is the flow the uniformity clause is about. Flag it now.

Ask your vendors three questions in writing. Which inputs set the discount amount, not the audience? Can you restrict that decision to purchase history? Can you produce, for one named customer, a record of what data was considered for their offer? If the answer to the third is no, the bill’s request procedure will be hard to meet.

Agencies, separate the briefs. Put the audience brief and the offer brief on different pages, and treat the discount rule as a fixed input the client signs, not something the campaign optimizes.

Park offer depth tests for legal review. Do not kill them in a panic. Do not scale them into 2027 either until someone who reads statutes for a living has looked at the uniformity clause.

Personalization has always been sold as relevance, and almost nobody objects to relevance in the message. The fight was always going to be over the number. New York just put that line in writing: aim the pitch as precisely as you like, but the price has to be one the next person could get too.