I Told a Client AI Was Pricing His Customers. Then I Went and Checked.

He asked a simple question. I gave him a confident answer. It took me a week and about forty sources to find out I was wrong, and the real answer is better than the one I made up.

A guy who runs a home services company asked me a question on a call a few weeks ago, and I answered it before he finished asking.

He had been hearing about AI pricing. Airlines using it. Grocery apps using it. Somebody had told him the software could figure out what a customer was willing to pay and charge them exactly that. And he wanted to know the thing every business owner wants to know when a new capability shows up.

Should we be doing this?

And I said something like, yeah, this is real, it's happening, let me pull the research and send it over.

I want you to notice how fast that happened. He asked. I answered. There was no gap. No "let me look into it." I had a position ready to go, fully formed, and I delivered it with the confidence of a person who had done the work.

I had done no work. I had a feeling and some headlines.

So I went to get him his proof. And I could not find it.

What I Was Actually Sure About

Before I show you what happened, I have to be honest about what I believed walking in, because the belief is the whole story.

I thought the machine was reading all of us. Not in a tinfoil way. In a boring, obvious, of-course-it-is way. I assumed my flight price moved because I had looked at it twice. I assumed the grocery app knew my household and priced accordingly. I had cleared my cookies before booking a trip more than once, the way you knock on wood.

I run a marketing agency, a wedding venue, and a small AI consulting practice, and I have nine kids, so I buy a lot of everything. I was not neutral on this. I was already annoyed.

That is not analysis. That is a mood. And I handed it to a guy who was going to make a business decision with it.

The First Thing I Lost

I told AI to go get me the three best documented cases of companies charging different people different prices. Give me the proof.

It gave me the three everybody knows. I already had them in my head. That should have been the first warning.

Orbitz charged Mac users more. Except it did not. In 2012, Orbitz changed the default sort order of hotel results for Mac users, putting nicer hotels higher on the page. The same room cost the same for everybody. You could sort by price and see identical inventory. Orbitz's CEO, asked directly whether Mac users paid more for the same room, said it would be "absolute nonsense." His reason for the sorting was that Mac users were 40% more likely to book four and five star hotels.

The Wall Street Journal headline said Mac users were steered to pricier hotels. Steered. The right word was in the headline. It got lost somewhere between the article and the group chat, and I was carrying the lost version.

Amazon got caught pricing by demographics in 2000. Also no. Amazon's statement at the time is that prices varied "on a totally random basis, not with respect to customer demographic information." They refunded 6,896 customers an average of $3.10 each.

Uber charges you more when your phone battery is low. This one is the cleanest, and it is the one I would have bet money on. The source is a 2016 NPR interview with the head of economic research at Uber. He said battery level is one of the strongest predictors of whether somebody will accept surge pricing. And then, in the very same breath, he said "we absolutely don't use that to kind of like push you a higher surge price, but it's an interesting kind of psychological fact of human behavior."

That is a denial. It has been circulating as a confession for ten years. I have repeated it out loud, to people, as a fact.

Okay. Fine. Three anecdotes are anecdotes. I told myself I needed real research anyway.

The Second Thing I Lost

There is real research. It is just not on my side.

In 2014, a team from KU Leuven and Stony Brook ran the experiment the way you would run it if you actually wanted to know. Twenty-five airlines. More than 130,000 queries. Sixty-six different user profiles over three weeks. They handed the airlines every opportunity: manipulated cookies, built an affluent shopper persona and a budget shopper persona, tested five browser and operating system combinations, sent Do Not Track headers, ran it from two countries.

Their finding, in their words: "Despite presenting the companies with multiple opportunities for discriminating us, and contrary to our expectations, we do not find any evidence for systematic price discrimination."

The one anomaly they saw reversed itself before the study ended. A clean, reproducible price gap in Argentina turned out to be a 20% government tax on foreign credit card purchases.

The paper is called "Crying Wolf?"

I did not like that. So I went looking for a study that disagreed, which is not research, that is shopping.

Consumer Reports ran their own in 2016. Simultaneous searches, one browser with a full history, one scrubbed clean, 372 queries. Prices matched on 330. Of the 42 pairs that differed, the scrubbed browser showed the higher fare 25 times and the lower fare 17 times.

The clean browser was more often the expensive one. My ritual was not just useless. If anything it was backwards.

And Consumer Reports was more careful than almost anyone who cites them. They wrote: "We can't say whether the fare differences we found are due to pricing based on browser histories." A Kayak spokesman told them a few seconds of timing drift between two supposedly simultaneous searches could explain the entire result.

I had built a belief out of three anecdotes I had never checked and a habit that ran the wrong direction. And I had already given a client an answer based on it.

The Part That Explained Why I Believed It

Here is the piece that actually made me feel better, because it explains the feeling instead of just dismissing it.

Airfares genuinely move while you are looking at them. Fares get filed as tariff data, something like 306 million active fares with more than 12 million changes a day, sold out of a limited set of booking classes. When the cheap bucket empties, the next price shows up. That is real, it is fast, and it has nothing to do with you. You just watched a seat get sold to somebody else.

But the better answer is this one. Researchers at Northeastern found that Expedia and Hotels.com randomly assign shoppers to A/B test buckets, and they do it through cookies.

Which means clearing your cookies genuinely does change what you see. Randomly. In both directions. With no relationship at all to anything the site knows about you.

The trick "works" often enough to feel real because it is a coin flip, and we only remember the flips we won.

That is not a company manipulating you. That is a slot machine, and I had been pulling the lever and building a theory out of my wins.

At this point I was ready to call the guy back, tell him I had been wrong, and kill the article. It had become a piece about how nothing is happening, which is not a piece.

Then I looked at why Congress is currently making phone calls.

The Thing That Is Actually Real

As I write this, eight US airlines are sitting on a deadline. On August 11, the ranking member of the House Energy and Commerce Committee sent letters to American, Delta, United, Alaska, JetBlue, Southwest, Frontier and Hawaiian demanding answers to nineteen questions about whether they use personal data to set prices. Responses are due August 25. Twenty-five grocers and retailers got the same letter in May.

That does not happen over nothing. So what is it happening over?

Not the airlines' behavior. The vendors' sales pitches.

Read what the pricing software companies say about themselves. Marketing copy is written to impress buyers, not regulators, so it is the least guarded language in this entire story.

PROS Holdings, one of eight companies the FTC subpoenaed, describes a system in its own case study that uses "15+ dimensions" to "match customer willingness-to-pay." It runs something it calls "price exploration," which is a polite name for live price experiments on real customers, as a designed feature. Mastercard's personalization product opens its page with "Capture person-level data, no matter the source."

And then there is the one that stopped me cold.

In a letter to Costco, quoted in the congressional inquiry, Instacart wrote this to a client:

"by thoughtfully pricing certain items with high or low price sensitivity, retailers have seen customers' overall price perception improve and their engagement increase as a result. For some of our major grocery partners, this has led to millions of dollars in annual incremental sales."

That is not a leak or an allegation. That is a company explaining to a customer, in writing, that pricing by price sensitivity produces millions in extra revenue.

And That Is When It Clicked

Here is the sentence I wish somebody had handed me on day one, before I opened my mouth on that call.

Nobody is watching you. Somebody is selling the watching.

That single reframe made every contradiction in my research line up.

It explains why the proof keeps evaporating. There is very little proof that companies charged you a personal price, because the thing being bought and sold is the capability, and capability leaves sales decks and case studies, not receipts.

It explains why the FTC study says what it says. Everybody cites it, I cited it, and its own text says it "only includes hypothetical examples of surveillance pricing." The famous new-parent-and-the-baby-thermometer case is an illustration pulled from vendor materials. It is not an observed transaction. The flagship federal investigation into surveillance pricing did not document one consumer paying more. It documented that this is a product line.

It explains the strangest fact I found all week: the airline lobby publicly supports banning surveillance pricing while insisting no airline does it. That is not hypocrisy. That is an industry that would like to stop being sold something.

And it explains the best evidence on the other side. When Consumer Reports tested Instacart last December with 437 shoppers in simultaneous sessions, they found about 74% of items showed more than one price. Real, measured, not folklore. But they also collected shopper demographics, ran the regressions, and reported that shopper characteristics did not explain the differences. They had every reason to claim targeting and they published the weaker, truer finding. What they found was price experimentation, not personalization. Somebody was testing. Not reading.

The machine everybody is afraid of exists. It is just still mostly in the showroom.

The Number That Was Never There

One more thing, and this is the part that changed how I work.

The congressional letter to the airlines contains this sentence: "One study found that an airline boosted its own revenue by as much as six percent by leveraging artificial intelligence pricing based on consumer information." That figure is now in national press as "one study cited by Pallone."

I opened the footnote. It points to a Forbes Advisor consumer advice column. I opened that. The sentence reads: "In one trial, a large unnamed network airline saw a 6% revenue boost by switching to AI-powered pricing."

No citation. No link. No institution. No researcher. No airline.

There is no study.

An unsourced line in a consumer column became "one study found" in a letter from a ranking member of Congress, and then became national news. That took me about ten minutes to unwind, and I only did it because by then I had stopped trusting anything that agreed with me.

I am not telling you Congress lied. I am telling you something less dramatic and more useful. Everybody in that chain did exactly what I did on that phone call. They passed along a thing that felt obviously true without stopping to check, because checking a claim you already believe feels like a waste of time.

The claim I never questioned was the one that agreed with me. That is not a media problem or an AI problem. It is what certainty does, and it is fast.

I should say the research itself was excellent. AI found the primary sources, the actual transcripts, the peer-reviewed studies, the statutes. It did the work of a research team in a week. The failure in this story is not the machine's. It is that I had already decided before I asked.

So What Do I Tell the Guy Who Asked

He wanted to know if he should be doing this. Two answers, and the second one is the one that actually worries me.

First answer: no, and you almost certainly cannot anyway. The economics do not work the way people assume. The best research on personalized pricing, a 2023 study in the Journal of Political Economy, found that more than 60% of customers ended up paying a lower price under personalization. In a market with a lot of competitors, which is exactly the home services market, personalized pricing tends to be a race that moves money toward customers, not away from them. The scary scenarios all require somebody to have real market power. A contractor with six trucks does not have it.

Second answer, and this is the one I did not see coming. He may already be exposed, without doing any of this.

New York passed a law that has been enforceable since November. If a price is set by an algorithm using personal data, it has to carry this, clearly, in the same medium as the price:

THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA

Then read the definitions, which I did twice because I did not believe them.

"Algorithm" means "a computational automated process that uses a set of rules to define a sequence of operations." Not AI. Not machine learning. A rules-based process qualifies.

"Personal data" means anything that "could reasonably be linked, directly or indirectly, with a specific consumer or device." A service address qualifies.

"Consumer" covers somebody buying a good or a service. And there is no small business exemption. Connecticut's version starts October 1 with its own required label, and its disclosure rule applies to "any person" doing business in the state. New York has a full ban sitting one signature away, passed by both chambers in June, unsigned, with no size threshold at all.

Put it together and a contractor whose quoting software adjusts a price based on the customer's address is arguably inside the literal text.

Meanwhile I went and checked what the trades are actually being sold, and this is the part I want to grab people by the shoulders about. Nobody is selling surveillance pricing to contractors. I looked at ServiceTitan, Housecall Pro, Jobber and Workiz. ServiceTitan ships a feature literally named "Dynamic Pricing," and it is not dynamic pricing in any regulatory sense. It recalculates flat rate prices when material or labor costs change and applies your own markups. The inputs are your costs. Not your customer's data.

But a contractor whose website says "we use dynamic pricing" has created a problem for something he is not doing. Wendy's proved that in 2024. National reputational beating and a Senate letter over a feature it never built, never launched and arguably never proposed. The whole liability was two words on an earnings call.

The Tuesday Afternoon Version

For the record, I could not find one enforcement action, warning letter or lawsuit against a home services business in any state. Not one. The exposure lives in the text of the statutes, not in anybody's docket. So this is a thing to fix on a Tuesday afternoon, not a thing to lose sleep over.

Publish your price book. Uniform published pricing is close to a complete defense under every one of these laws, because a business that charges everybody the posted price is not doing personalized pricing by definition.

If you charge more in a zip code, put the reason in a file and make the reason a cost. Drive time, fuel, permits, insurance. Every one of these laws exempts geographic differences that are cost justified, and the documentation is what turns it into an exemption instead of a problem.

Publish your discount criteria. "10% for veterans with ID" is exempt in every statute I read. "We take care of veterans" is not.

And take the phrase "dynamic pricing" off your website. Say flat rate. Say upfront pricing. Those are accurate, they are what customers actually want to hear, and they trigger nothing.

The real risk was never software anyway. It is the undocumented judgment call. The bid that came in higher because there was a nice truck in the driveway. That was always shaky under laws that have existed for decades. These new statutes did not create that risk. They raised the odds somebody writes it down.

And if you are on the buying side of all this: stop clearing cookies before you book a flight. Sort by price. Turn off location before you open a retailer's app in their parking lot. Turn on Global Privacy Control, which twelve states now legally require companies to honor, and which California regulators have fined companies six and seven figures for ignoring. Those four things do more than every browser trick you have ever been told about.

What I Owe Him

I called him back.

Not to correct a fact. To correct the way I answered. Because the fact was almost incidental. What actually happened on that call is that a guy asked me a real question about his business, and I gave him the version of the answer that was already sitting in my head, and I said it in a tone that told him he did not need to check.

That is the failure in this article, and it is mine. The research was not wrong. The research was outstanding. I was wrong before the research started, and then I nearly published it anyway, because being angry at big companies is comfortable and being wrong out loud is not.

Here is the part I am still sitting with. If I had not gone looking for proof to send him, I would still believe all of it. I would still be clearing cookies. I would still be repeating the Uber battery thing at dinner. The only reason I found out is that somebody made me put it in writing.

I did not get corrected by better information. I got corrected by having to show my work to somebody who trusted me.

So here is the question I would put to you, and it is the one I had to answer about myself.

What do you currently believe about your own business that you have never once checked, because it feels too obviously true to be worth the ten minutes?

I write about building with AI in the real world at BenSAIBrain.com. If you want a second set of eyes on whether your pricing and quoting setup creates exposure you did not know about, that is exactly the kind of question we work on at Digital Ignitor.