Macy's AI Assistant Made Customers Spend 475% More. Here Is What Nobody Is Asking.

Macy's launched "Ask Macy's" two weeks ago — a Google Gemini-powered shopping assistant that lets customers describe what they want and get curated recommendations. During testing with half its website traffic, customers who used the chatbot spent 4.75 times more than those who did not.

Every headline is treating this as a success story. And it is — for Macy's. But I have been building a consumer AI platform for the past year, and the 4.75x number makes me think about a question that nobody in the coverage seems to be asking:

Whose side is the AI on?

The Upselling Machine

Let me explain what "Complete the Look" actually does. You search for a navy blazer. The AI does not just show you blazers. It shows you the blazer, then suggests a shirt that pairs with it, then a belt, then shoes, then a pocket square. Each suggestion is personalised based on your browsing history, purchase data, and stated preferences.

This is what a great store associate does — and nobody objects to a helpful salesperson. But there is a difference. A store associate works for the retailer, and you know it. When they suggest a pocket square, you understand that their incentive is to increase your basket size. That transparency is built into the human interaction.

An AI assistant feels different. It feels like it is working for you. The conversational interface — "Tell me what you are looking for, like you would a store colleague" — creates an impression of personalised service. But the AI's objective function is not "help this customer find the best product at the best price." It is "maximise basket size and conversion rate."

Those two objectives sometimes align. Often, they do not.

What 4.75x Actually Tells You

I want to be precise about what the data does and does not say.

It says customers who used the AI spent 4.75 times more. It does not say they were 4.75 times more satisfied. It does not say they got 4.75 times more value. It does not say they would not have found better products at lower prices elsewhere.

There is also a selection bias problem. During the test period, Ask Macy's was shown to approximately half of website visitors. The customers who chose to engage with a new AI chatbot are not a random sample — they are early adopters, likely younger, higher-income, and already high-intent shoppers. The 4.75x multiplier compares these self-selected power users against the general browsing population, which includes people who landed on the site by accident, were just checking prices, or had no purchase intent.

None of this means the AI is not genuinely useful. It probably is — for many customers, being shown coordinated outfits saves time and produces better results than browsing manually. But the headline number conflates "AI drove more spending" with "AI helped consumers make better decisions." Those are very different claims.

I Have Seen This Movie Before

I spent years in consumer markets before starting Mubboo, and this pattern is familiar. Every time a new technology is introduced into retail, it gets optimised for the retailer's metrics first and the consumer's interests second.

Email marketing started as "helpful product updates" and became inbox spam. Recommendation engines started as "customers who bought this also liked" and became algorithmic upselling. Loyalty programmes started as genuine rewards and became data harvesting operations with points that expire.

AI shopping assistants are following the same trajectory — and they are starting from a more powerful position because the conversational interface makes the commercial intent less visible. When Amazon's Rufus suggests a product, most consumers do not stop to ask whether that suggestion is optimised for their satisfaction or for Amazon's margin. When ChatGPT integrates direct checkout from Target, the line between recommendation and advertisement becomes invisible.

We are building Mubboo specifically because we think this trajectory is wrong.

The Independent Layer

Here is what I believe: consumers need AI that works for them, not for the retailer.

That sounds obvious, but the incentive structures in retail AI make it genuinely difficult. Macy's AI is trained on Macy's catalogue. It will never tell you that the blazer you are looking at is $80 cheaper at Nordstrom, or that a better-rated alternative exists at a brand Macy's does not carry. It cannot — that would work against the retailer's interests.

An independent comparison platform has a fundamentally different incentive structure. We get paid when you click through to any retailer, not when you buy from a specific one. That means our AI can genuinely say "skip this, the one at the other store is better for your needs" — because our business model does not depend on steering you toward a particular merchant.

I am not claiming we have solved this. We are early. Our AI features are not live yet. But the architectural decision — to build AI that sits between the consumer and the retailer, rather than inside the retailer — is a deliberate one. It is the only position from which AI can authentically represent the consumer's interest.

What I Would Watch For

If you are a consumer using any AI shopping assistant — Macy's, Amazon's Rufus, ChatGPT's shopping integration — here are three questions I would ask:

Does the AI ever tell you not to buy something? If it never says "this product has poor reviews" or "you can find this cheaper elsewhere," it is an upselling tool, not an advisor.

Does the AI show you options from competitors? If it only recommends products from one retailer's catalogue, it is not optimising for your interest. It is optimising for the retailer's.

Does the AI disclose its commercial relationship? Macy's AI is powered by Google Gemini, but it serves Macy's inventory. ChatGPT's shopping features involve commercial deals with specific retailers. Transparency about these relationships is the minimum standard consumers should expect.

The Bigger Picture

Forty percent of the top 20 US retailers have now deployed AI shopping assistants. Amazon has "Buy For Me." OpenAI has embedded checkout from Target, Instacart, and DoorDash directly into ChatGPT. Every major tech company and retailer is racing to become the AI layer that sits between consumers and their purchasing decisions.

The company that wins that race will have extraordinary influence over what people buy, from whom, and at what price. That is not inherently bad — if the AI is genuinely working in the consumer's interest, it could save people time, money, and effort at a scale no human advisor could match.

But 4.75x spending increases make shareholders happy, not necessarily consumers. And I think we should be honest about that before we celebrate.

I am building Mubboo because I believe the consumer AI space needs at least one platform whose primary loyalty is to the person shopping — not the company selling. Whether we succeed is an open question. That the attempt is necessary seems increasingly obvious.