Everyone wants to ship hyper-personalized experiences today.

Glenn Fogel CEO at Bookings shared with Barrons in January

You want something personalized so much so that it's just like it was in the old days when it was a human being travel agent that knew you so well that knew what you liked and what you could afford.

Walmart’s CEO Doug McMillon said the same in August last year

There’s a great opportunity for us to be more anticipatory, and to be more relevant to [customers] and communicate in a way that shows that we know who they are, in a healthy way, while protecting privacy.

Hyper-personalization – the anticipation of customer needs given complete user context – feels inevitable.

We really want a hyper personalized experience that understands our needs one step ahead of our own brains.

said Micah Rosenbloom a seed stage investor at Founder Collective.

Hyper-personalization’s benefits to businesses appear clear – especially among businesses with large inventories, personalized recommendations of relevant products shift up consumer demand and increase revenue.

But hyper-personalization – its operation on vast consumer context – could also come with a cost.

‍

You like that car, huh?

On a path to hyper-personalization, consumers are poised to have a choice. Turn on hyper-personalization in an app – bring some or all of your context with you to unlock hyper-personalization – or not.

The benefits of turning it on is that the internet becomes more like you. Everything you do in your life can be reflected back by any app, service, piece of hardware, or ecommerce store you want. No need to reintroduce yourself to each domain – just turn on hyper-personalization with a tap.

But what about hyper-personalized prices?

For instance, suppose you’re at a car dealership looking for a new car. The salesperson asks to learn a bit about you

• What sort of car are you in the market for
• Any particular reason you’re looking for a new car (new job? promotion?)
• Fast or fuel efficient?
• Are you a fan of the brand?

Even what you’re wearing could give the salesperson an impression of what sort of customer you are.

Based on all of this information, the salesperson can give you a recommendation. You benefit from this because you might not have known that much about the available options, and your additional context is helpful for the salesperson to recommend the car that best fits your needs.

On the other hand, when it comes to buying the car, all this shared information could be used against you. For instance, the salesperson knowing that you love a particular car may make the salesperson less willing to compromise on price. They are personalizing their price according to their estimation of your willingness to pay.

Many car dealers have committed to resolving this by setting a no-haggle policies wherein all prices are fixed up-front.

On the other hand, e-commerce uses dynamic pricing broadly, using demand signals and other inputs to adjust prices throughout the day.

This is related to but empirically distinct from microeconomics’ price discrimination. Here’s how a Boston Consulting Group brochure described it in 1986:

A key step is to avoid average pricing. Pricing to specific customer groups should reflect the true competitive value of what is being provided. When this is achieved, no money is left on the table unnecessarily on the one hand, while no opportunities are opened for competitors though inadvertent overpricing on the other. Pricing is an accurate and confident action that takes full advantage of the combination of customers' price sensitivity and alternative suppliers they have or could have.”

With hyper-personalization, first-degree price discrimination – where firms charge each of us exactly our willingness to pay – seems entirely possible.

Under hyper-personalization, firms know our past actions and preferences – they could theoretically figure out exactly how much we value everything and charge us exactly that.  

Though personalized pricing could lead to greater efficiency like making products available to more people, it also appears to lead to less consumer surplus, taken as the difference between our willingness to pay and the prices we actually pay.

Ideally consumers could get personalization without affecting consumer surplus.

‍

I know you know I know

It seems though that we shouldn’t worry too much about this.

This is for two reasons.

First, even though many businesses already have access to rich data about us, empirically they don’t seem to use it to personalize prices as much as they could.

For instance, Uber adjusts ride prices based on where you are and how many drivers are around. Of course, Uber knows other personal signals that it could activate to further adjust its prices. For instance, suppose it's cold and rainy and your phone is at 3 percent battery. If you're looking for a ride, that would be pretty rich signal that your inclination to pay a higher price is probably quite a bit higher than usual. That said, Uber reports not using signals like battery life for its dynamic pricing.  

An older 2017 study of 2000 e-commerce sites similarly found that personal-data-induced price discrimination is rare.

Second, it turns out the possibility of personalized pricing under selective context sharing in hyper-personalization might make consumers better off.

Microeconomics professor Shota Ichihashi studied this stylistically in the prestigious American Economic Review.

In deciding to share context, consumers face a trade off: share more context in exchange for more relevant experiences but at the possible cost of price discrimination.  In this framework, a seller can

• commit to not personalizing prices to induce more sharing, setting prices in advance
• not commit and leave a consumer to decide how much to share given the possibility of personalized prices, setting prices once context is observed

No surprise, the model reveals the seller is better off committing to not personalizing prices. The intuition is what you’d expect. When a seller commits to not personalizing prices, consumers tend to share much more context that sellers can use to provide more relevant experiences that shifts up customer spending.

Interestingly, however, this seemingly consumer-friendly policy – committing to not personalize pricing – may actually make the consumer worse off.

Why?

Ichihashi explains

To see this, imagine that a seller commits to not personalize prices, and all consumers except you provide their data. The seller can use data to learn consumers’ tastes, and show each consumer the products he or she likes. Because the seller can present each product to consumers who value it highly, the seller can set a relatively high price for each product. Since prices are not personalised, you will also have to pay these high prices, regardless of whether you provide data.

That is, other customers’ context sharing has the negative externality of higher prices, that you, someone who decided not to share, has to pay. In the car dealership case, although a no-haggle policy allows customers to be fully open about their needs, it could also lead to higer prices for someone more privacy sensitive.

‍

Will AI protect us from price personalization?

Ichihashi focuses on the case where a buyer discloses information to a seller in exchange for a more relevant recommendations and experiences. It’s understood that the seller has richer information about available inventory than the buyer, and that’s why a buyer must disclose information to the seller.

Many are excited about the possibility of user-owned AI agents protecting us from these dynamics.

We’re not so sure.

Suppose you’re thinking of buying products from a seller, and you’ve dispatched your AI agent to learn their inventory to recommend a product to you. In this case, the seller cannot personalize prices because it never has an opportunity to – your AI agent scraped all the seller's inventory and your agent does its analysis in an environment you own.

But if the seller can infer that your AI agent is accessing its inventory on your behalf, it can revert to its strategy of setting relatively high prices for its products, as it knows that its products are distributed – via buyer-owned AI – to those who value its products highly, even though it’s not the one doing the personalization.

So while this scraping pattern keeps your data private (though possibly violates seller Terms of Service), it may not protect against price personalization.

‍

Turning on hyper-personalization

In turning on a hyper-personalized internet, we’ve worried about implications of discrimination. If a third party now knows something about us, what can they do with that information, even if it’s shared under an explicit agreed-upon context?

We're excited to see more study here. For instance, extending Ichihashi's framework to competitive markets with differentiated products and repeated seller-buyer interactions could give new context into incentives to personalize prices as a function of how seller reputation affects buying decisions.

Aside from strategic implications, it seems like many businesses do not use personalized prices because it just doesn’t feel right. Turning on hyper-personalization should create great internet experiences while leaving consumers feeling like they’re getting mutual value.

But we’re excited to see theory showing how, per Ichihashi

“it could be optimal for an individual consumer to [turn on hyper-personalization], as long as sellers do not personalise prices.”