#AI4M: AI Shopping Agents Move From Apps to Storefronts

The Daydream app allows shoppers to use natural language to find fashion. You and your students can read about it in “AI shopping is coming to your favorite brand’s website” (Fast Company, July 29, 2026).

E-commerce search hasn’t really changed since Amazon started selling books online three decades ago: type keywords, scroll thumbnails, narrow by filter. Fast Company writer Elizabeth Segran tested an alternative on Alice + Olivia’s website — a “Shop with AI” button that let her describe what she needed (“chic, polished, and comfortable enough to sit on stage in”) and returned a curated set of dresses instead of a category grid. The tool behind it, Daydream, today launches Powered by Daydream, which embeds its natural-language fashion search directly into brand websites. Staud, Alice + Olivia, Couper, Cult Mia, and Hampden Clothing are live now, with more than 25 additional brands signed on, including Anine Bing, Mansur Gavriel, Sandro, and Maje.

Daydream started as its own consumer app, but founder Julie Bornstein says the bigger obstacle wasn’t the technology — it was thirty years of keyword habits. “People are trained to use basic keywords,” says co-founder Lisa Yamner. “Getting people accustomed to using natural language is a journey.” There’s also a trust problem: because large language models are non-deterministic, the same query can return slightly different results each time, which can read to a shopper as the tool being broken rather than working as designed. Embedding the AI directly into a site a shopper is already browsing — rather than asking them to download a new app — is Daydream’s answer to both problems, and the deployment is deliberately lightweight. Alice + Olivia’s Carrie McMahon says her tech team had it live within a day: “You can put this piece of code on the site and Daydream takes it from there.”

The other payoff is data. Because a chat query carries intent — not just “black dress” but “something to wear to a work event” — brands can see demand signals a click-based catalog never surfaces. One brand discovered shoe queries were disproportionate to the fact that shoes made up only 4% of its catalog. Alice + Olivia keeps getting requests for a discontinued line, which Yamner says might justify bringing it back. Bornstein, who worked at Nordstrom while it built its early e-commerce site, sees a parallel: slow, skeptical adoption followed by a tipping point after which “they won’t be able to imagine shopping without it.”

Claude was used to develop a first draft of this blog post.

Relevant Chapters in Essentials of Marketing

Chapter 5 (the consumer decision process, information search) is perhaps most relevant, and two named boxes there are close parallels: “Will AI shopping agents do all the work?” and the #AI4M box on AI, predictive analytics, and the consumer decision process, which uses a virtual assistant at Best Buy that “analyzes past purchases and browsing behavior to offer customized product explanations” — almost exactly what Daydream does. Daydream also maps onto the chapter’s four-step problem-solving model in Exhibit 5–10 — information search, identify alternatives, set criteria, evaluate alternatives — compressing all four into one conversational exchange. Two more connections are worth pulling in. Chapter 16 (AI-powered interactive content, personalized recommendation systems, conversational chatbots) is a near-direct match — Daydream is functionally the same kind of tool as the chatbot and recommendation-engine examples covered there, just purpose-built for one category. Chapter 7 (the data–information–knowledge–wisdom model in Exhibit 7–4, and the research methods in Exhibit 7–7) fits the second half of the article: the shoe-inventory mismatch and discontinued-line requests are knowledge, in the chapter’s specific sense — information that answers a “why” — generated by something close to what Exhibit 7–7 calls a chatbot survey. While not as relevant, the article shows how Daydream pivoted (Chapter 9) and how it works with retailers (Chapter 12).

Class Discussion Ideas

Two threads run through this article: how AI changes the shopper’s side of the buying process, and how it hands the brand a new kind of research data almost for free. The activities and questions below split roughly along those lines.

In-Class Activities

• Try It Yourself. Have students visit a live Daydream brand (Alice + Olivia or Staud) and run a natural-language query, then run the same request as a keyword search on a competitor’s site. Compare relevance and effort. Good for a short in-class demo or homework with a follow-up discussion. (Chapter 5)
• Search Query Data Mining. In groups, give students a hypothetical batch of “Shop with AI” queries for a fictional brand and have them identify at least two inventory or product decisions the data suggests, similar to the shoe and discontinued-line examples in the article. (Chapter 7)
• Redesign the Recommendation Engine. Teams pick a retailer without conversational search and pitch how it could deploy something like Daydream, including what new customer data it would start collecting once live. (Chapter 16)
• Adoption Curve Debate. Half the class argues AI shopping agents are near a tipping point, using Bornstein’s e-commerce comparison as evidence; the other half argues shoppers will keep defaulting to keyword search. (Chapter 5)

Discussion Questions

1. Where in the consumer decision process does “Shop with AI” intervene, and what does it change about that stage? (Chapter 5)
   • Answer: Exhibit 5–10 breaks problem solving into four steps: information search, identify alternatives, set criteria, and evaluate alternatives. Daydream compresses all four into a single conversational exchange — the shopper states her criteria in her own words, and the system searches, filters, and ranks the catalog in one pass instead of leaving each step to her. That’s a bigger change than it first looks: it’s not just a faster search box, it’s collapsing four separate steps of the model into one.
2. Julie Bornstein says shoppers have to “unlearn” thirty years of keyword search habits. Explain this using a concept from Chapter 5. (Chapter 5)
   • Answer: This describes routinized response behavior — Chapter 5’s term for what happens when a consumer has enough experience meeting a need that no new information is required to do it. Decades of keyword search have become exactly that kind of low-effort, automatic routine, so a genuinely different interaction, like typing a full sentence, requires more conscious effort even when it produces a better result. That extra effort is a real barrier to adoption regardless of how good the underlying tool is.
3. Bornstein notes that LLM results are non-deterministic — the same query can return different results each time. Why might that frustrate shoppers, and how would you address it in the design of the shopping experience? (Chapter 5)
   • Answer: Shoppers expect a search tool to behave predictably, so variation reads as a malfunction rather than a feature, which undermines trust. Options include setting expectations upfront that results may shift, showing why an item was recommended, or letting shoppers refine and lock in criteria rather than blindly rerunning the same query.
4. Alice + Olivia discovered that shoe queries far outpaced shoe inventory, which made up only 4% of the catalog. What kind of marketing information just got generated for the brand, and what should management do with it? (Chapter 7)
   • Answer: This is close to what Chapter 7’s Exhibit 7–7 calls a chatbot survey — a quantitative questioning method — except it’s continuous and customer-initiated rather than a one-off study a researcher had to design. It’s exactly the kind of external, always-on data an MIS is built to capture. Management should validate the signal against traffic and sales data, then likely expand the shoe assortment or promote the category more aggressively if the underlying opportunity holds up.
5. Compare Daydream’s data — “why someone is looking for something” — to a standard retail analytics dashboard that only tracks clicks and purchases. What does the qualitative “why” add that click data can’t? (Chapter 7)
   • Answer: Chapter 7’s data–information–knowledge–wisdom model in Exhibit 7–4 draws exactly this line: data becomes information when it answers who, what, or how much, but it only becomes knowledge when it answers how or why. A click log tells a brand what a customer bought; Daydream’s query — “something to wear to a work event” — tells them why, which is knowledge in the chapter’s specific sense and a far richer signal for merchandising and messaging decisions.
6. How is Daydream’s shopping agent similar to — and different from — the chatbots and personalized recommendation systems described in Chapter 16? (Chapter 16)
   • Answer: Like the examples in Chapter 16, Daydream uses natural language and behavioral signals to personalize recommendations and cut down the customer’s search effort. It differs in scope: it’s narrowly focused on one category and sits earlier in the funnel, driving discovery and purchase rather than post-purchase support, and its value depends on catalog-wide product knowledge rather than an individual account’s service history.
7. Bornstein compares today’s AI shopping agents to early e-commerce adoption at Nordstrom, predicting a similar “tipping point.” Evaluate this comparison — is it a fair analogy? (Chapter 5)
   • Answer: Chapter 5’s adoption process — awareness, interest, evaluation, trial, decision, confirmation — is built for exactly this situation: a “really new concept” where past experience doesn’t transfer. Early e-commerce and AI shopping agents both had to move real customers through all six steps before a habit could form, which supports Bornstein’s comparison. Where the analogy gets weaker: non-deterministic results introduce a kind of friction — the same query behaving differently each time — that early e-commerce never had to design around, so the evaluation and confirmation steps may take longer to complete this time. Strong answers should land on a specific judgment rather than just listing similarities and differences.
8. If you were advising a small independent boutique, rather than a brand like Alice + Olivia, on whether to adopt a tool like Daydream, what would you want to know first? (Chapters 5 and 7)
   • Answer: Key considerations include catalog size — is there enough inventory variety for natural-language search to beat simple browsing — integration effort relative to McMahon’s “one day” claim, and whether the boutique has enough traffic to generate useful query data in the first place. A small, tightly curated catalog may get less benefit than a brand with thousands of SKUs.