Learnings from Consumer AI × Fashion (Part 1): Why Alle Didn’t Work

When product-market fit isn't enough: Lessons from building and shutting down an AI stylist

After I announced Alle’s closure a few weeks ago, I received a surprising number of messages from:

1. founders building in the consumer AI space

2. investors trying to refine their mental models

3. and the occasional journalist looking for a sensational angle

In the interest of the first two groups, and to gently discourage the third, I realised it might be useful to share from our experiences. My intention is to provide useful context and information for anyone considering something similar/adjacent - in the hope it makes them move faster faster and with greater clarity - while also protecting my time by minimising redundant conversations.

Since I do not have a formal structure for this, I will approach this by answering the questions I was asked most often.

---

Question 1: Consumer AI is transformative, and fashion is a massive market that seems primed for disruption by AI. Alle also had a first-mover advantage. So went wrong?

• Well yes, but those two factors alone are not sufficient enough to imply the existence of a large business opportunity for a new company at this time in the market. At Alle, while we found a product with product–market fit, we were unable to discover a supporting business model.

• The product that generated the strongest user love at Alle was the personal AI stylist. We knew it had PMF based on two classic signals:

  • A retention curve that flattened over the long term

  • Users consistently shared stories of real utility that could be reasoned from first principles as a clear 10x improvement over existing alternatives.

  As a personal stylist, Alle helped users navigate moments of fashion indecision, such as finding outfit inspiration for an occasion, getting styling advice, receiving feedback on a planned look, discovering similar products from inspiration they saw on Instagram, etc. This proved to be a 10× solution compared to incumbent alternatives like:

  • Asking friends, sisters, or mothers was less effective because Alle was always available, gave neutral answers laden in fashion theory, supplied visual ideas to reinforce its suggestions, and closed the loop by helping users shop directly from platforms of their choice (we had scraped inventory across top marketplaces (Myntra, Ajio, etc.), popular brands, and independent labels).

  • Hiring a human personal stylist, which is difficult to discover and economically inaccessible for most users.

• So we found a 10× product for a sharply focused and very common use case. That’s great, right? The catch, however, is that consumer usage frequency for this product is quite low (~3–4 times a month). And the reason is simple: people tend to seek a stylist only in moments of acute fashion confusion.

  It’s easy to assume (as we initially did) that deciding “what to wear” is a daily use case, and therefore that the underlying confusion exists every day as well.

  In reality though, this assumption overlooks an important principle of consumer product design. Consumers are constantly, and often unconsciously, weighing the ROI of using a product, which is a function of:

  • the acuteness of the problem

  • the physical, cognitive, and emotional effort required to use the solution

  • the effectiveness of the solution in solving the problem

  Viewed through this lens, “what should I wear today” is rarely a truly acute problem. People have been getting dressed every day of their lives, and over time they build strong mental models that work well for most days of the month. Compared to those ingrained habits, the current workflow of opening an app, framing a question, taking photos from one’s wardrobe, and parsing the response is simply not worth the trade off on a daily basis.

  (I suspect this calculus would change if AI-powered smart glasses became a prevalent form factor, as they would reduce physical and cognitive friction dramatically.)

  TL;DR: The current form factor of an AI personal stylist app that lives on the phone, while appealing to a large potential user base, naturally leads to low usage frequency (~3–4 times a month) - typically tied to specific occasions or new clothing purchases.

• Since a product (even a 10X one) can’t exist independently of revenue, we had to identify a viable business model for Alle. In the consumer space, the set of realistic business models is fairly limited:

  1. Commission-based revenue from the supply side (i.e., brands or sellers)

     1. Affiliate commissions

     2. Transaction commissions

  2. Advertising revenue from the supply side (i.e., brands, sellers, etc.)

  3. Subscription or microtransaction revenue from the demand side (i.e., consumers)

  We explored and experimented with each of these options. Here’s how it played out.

---

Business Model 1a: Commission from Affiliate Transactions (Question 2: Why couldn’t we monetise via affiliate commissions?)

• Because we had scraped inventory from most major marketplaces, brands, and labels, a meaningful share of user interactions naturally led to shopping scenarios. In some cases, users arrived with clear purchase intent. In others, the stylist’s visual outputs created that intent. This made it tempting to imagine a business that could avoid the operational complexity of a full-fledged commerce marketplace and instead function as a pure software layer, generating attractive revenue by driving affiliate traffic.

• However, while such an opportunity may exist, it comes with the following challenges that significantly limit its overall potential, and in extreme cases, make it infeasible:

  • First, the supply side of such an affiliate ecosystem is not fragmented; it is highly consolidated among a small number of large marketplaces and brands (for example, Myntra, Ajio, H&M, Zara, Savana, etc.). This concentration severely limits the leverage an affiliate can have. In practice, this shows up as low to modest commission rates (typically under 4 %) with downward pressure as volume scales. In a worst-case scenario, marketplaces could refuse to offer any affiliate commission at all, particularly if they perceive it as a growing traffic source that poses a potential threat to their own strategic position.

  • Second, low commission rates combined with low usage frequency of the stylist product, constraint the customer lifetime value (LTV) too. This, in turn, would require the business to acquire users at close to zero customer acquisition cost (CAC) in order to remain profitable over time. In the consumer space, zero-CAC opportunities are extremely rare (historically limited to social and messaging networks, or to step-function technology breakthroughs such as Google or ChatGPT).

---

Business Model 2: Advertising

• While consumer advertising represents an enormous revenue pool, a product needs to meet a few strict criteria for an advertising business model to be viable.

  • Scale: At least a few million monthly users, which is difficult to achieve unless the company has either a near-zero CAC acquisition engine or an interim business model that can support/absorb acquisition and technology costs on the path to scale.

  • Engagement frequency / time spent: Even scale alone is insufficient if users do not engage extremely frequently (~15-30 times per month). If you look at companies where advertising is the primary business model, this pattern holds consistently.

  Because the consumer advertising funnel starts in moments of low intent, conversion rates are very low. As a result, the product needs a very large volume of consumer touchpoints to have any chance of delivering a decent ROI to advertisers.

• The real kicker, however, is that even if a product satisfies both of these requirements, it may still fail as an advertising business unless it can provide better - or at least comparable - ROI to alternative channels where advertisers already deploy budget (e.g., Meta, Google). It takes significant economic and time investment for advertisers to seriously evaluate a new channel, and that effort is only justified if the expected returns are close to what they already achieve elsewhere.

Since Alle didn’t have the level of engagement frequency that advertising demands, this business model was not a viable fit either.

---

Business Model 3: Consumer Subscription

Consumer subscriptions have begun to show renewed promise in the AI era (helped, in India, by mandates around recurring UPI payments). Since the stylist product delivered clear, ongoing value to users, it felt natural to test a subscription-based model. And because the U.S. represents a significantly larger and more mature market for consumer subscriptions, launching there felt inevitable

A Short Detour to US (Question 3: Why not build for US as the primary market?)

• The primary motivation for launching in the U.S. was that it represented a much larger and more mature market for consumer subscriptions. In addition, the market structure - greater fragmentation on the supply side and no few dominant marketplaces - made it more conducive to affiliate transactions as a secondary revenue stream.

• We were confident the product would achieve a similar level of PMF as it had in India, if not better, and early metrics validated that belief. However, the real test was whether consumer subscriptions could work as a business model. On that front, the results were deeply disappointing. Our subscription conversion rates were roughly one-tenth of the benchmark for comparable subscription products (and would declined further with scale). While incremental improvements to it were possible, the gap was so large that it pointed to a fundamental, structural issue rather than something that could be fixed through better acquisition or product iteration alone.

• This led us to our core insight: consumer subscriptions work as a business model only when product usage frequency is very high (yes, we’re talking about frequency again). Consider products where subscriptions are the primary revenue model:

  • Entertainment apps like Netflix and Spotify (watching TV and listening to music are extremely high-frequency behaviours).

  • Language-learning apps like Duolingo (they primarily monetise English, since learning it enables upward economic mobility; and paying users are those who spend time practicing daily).

  • Even apps like ChatGPT and Gemini - which arguably deliver enormous value - have only ~5% conversion, and the primary use case is productivity, something their paying users rely on daily.

  Since Alle was not a high frequency product, the willingness to pay within any acquired cohort will be correspondingly low.

• At the same time, another foundational assumption behind the business began to break down. We had believed that horizontal consumer applications (ChatGPT, Gemini etc.) would not prioritise building a truly 10X product for fashion use cases. However, both Google and OpenAI publicly signaled their intent to move into shopping, and Google went as far as making fashion central to its demos at Google I/O. Fashion, it turned out, is a large enough revenue category for these companies to justify focused & sustained investment in building high-quality software. Once this became clear, two implications followed:

  • There was no longer any reason to believe we could continue building a 10× product that outperformed these horizontal applications, especially given that we lacked proprietary technology, data, or distribution.

  • Even if a narrow 10X opportunity for a new company were to exist, we did not have a clear right to win against the local teams solving the same problems.

• Altogether, our detour to the U.S. lasted a couple of months, after which we decided to pivot back to India and try the last remaining business model.

---

Business Model 1b: Commission from Commerce Transactions (Question 3: It feels Myntra is outdated and the timing seems right for a Myntra 2.0. Why don’t we pursue that?)

• We consistently heard from a meaningful number of our users (primarily urban, mass-premium to premium Gen Z demographic), that they faced the following frustrations with Myntra:

  • Selection fatigue: Users felt they were repeatedly shown the same brands and products. The sense of novelty and trend discovery they experienced on platforms like Instagram was largely absent.

  • Discovery inconvenience: Finding relevant products required excessive scrolling and filtering. The interface felt cluttered with discounts and deals, feeds lacked personalisation, and the overall experience made discovery feel more laborious than delightful.

• Our experience at Meesho had taught us that the core pillars of commerce are:

  • Trust

  • Selection

  • Price

  • Convenience

    • Pre-order (primarily discovery experience)

    • Post-order (fulfilment time, returns, customer support etc.)

• By this point, we had already built and validated that our AI-first discovery capabilities, such as conversational multimodal search, taste-based personalization, and content-led feeds, significantly improved the discovery experience (by ~10×). At the same time, a large portion of high-quality fashion supply, especially new-age D2C brands and independent labels, were either absent of difficult to discover on Myntra.

  This led us to believe there could be an opportunity to build a new marketplace. We articulated a thesis around an AI-first fashion shopping platform for Gen Z, with differentiated selection and AI-led discovery as the primary GTM drivers. Over time, we hoped this would allow us to evolve toward a zero-inventory, zero-commission marketplace similar to Meesho, and eventually build defensibility through lower prices driven by economies of scale. On paper, the thesis was strong, and early conversations with users and brands further reinforced our confidence in the opportunity.

• To test this thesis, we onboarded ~100 sellers, primarily popular D2C and design-first indie brands, and scaled to ~100 orders per day.

• However, it quickly became clear that the data and user behavior did not support the size of opportunity we had expected. Discovery retention (and, consequently, order retention) were abysmally low. In user conversations, a consistent theme emerged: most people used Alle only for special-occasions (eg: birthdays, date nights, etc.) rather than for everyday wear, which accounts for the majority of fashion market. While users appreciated the freshness and trendiness of the designs, they also felt that pricing across much of the catalog was relatively high.

• As we dug deeper into the pricing issue, we arrived at a more structural insight. India still lacks sufficient supply of Gen Z–focused brands that struggle with, a) discovery and b) strike the right balance between design, quality and price. The few brands that do get this balance right (eg: NewMe, Savana, The Souled Store, etc.) are already discoverable through their own direct channels, while established global brands (such as H&M and Uniqlo) are easily accessible on existing marketplaces. Without enough liquidity of brands that genuinely need discovery and offer the right selection–price fit, a new marketplace would be unable to deliver the 10× user value required to compete meaningfully with incumbents.

• Another common pattern we observed among new-age D2C brands that have broken out (NewMe, Savana, Urbanic, Outzidr, etc.) due to strong selection–price fit is that all of them source their a bulk of their supply from China (and most of them are the same designs manufactured for U.S./European markets). In contrast, indie or smaller brands with strong design sensibilities tend to manufacture in India, which results in significantly higher price points.

  This led us to a core insight behind the scarcity of brands (that offers both selection and price fit) which makes a new marketplace unviable today:

  • India’s western clothing manufacturing ecosystem is not yet mature enough, in terms of scale and efficiency, to deliver prices comparable to China’s.

  • The market already has several brands that effectively exploit the arbitrage of importing China-manufactured designs intended for Western markets, while executing well on supply chain and marketing.

  • As a result, for any new brand to break through, they must do one of the following:

    • Identify & trade new designs already being manufactured in China for other markets (an opportunity that is largely already capitalised, as noted above).

    • Create new designs and manufacture them in China to hit price points where most of the demand sits. But there is no strong “why now” that would meaningfully accelerate the creation of such brands.

    • Wait for India’s manufacturing ecosystem to mature enough to deliver lower price points than China’s. While this is a positive long-term trend, it is still several years away from becoming a reality.

• The conclusion we arrived at was that the timing is not right for a new marketplace that relies on selection as its primary GTM lever. In terms of timing, we expect to first see a growing number of independent brands over the next 4–6 years. If, by then, Myntra, Ajio, and Nykaa Fashion have still not addressed consumer pain points around selection fatigue and discovery convenience, an opportunity for a new fashion marketplace may open up once again.

---

TL;DR: Alle built a 10X AI personal stylist with clear product-market fit, but the core challenge was not product - it was business model (or lack of). Usage frequency was low (3–4 times/month), because users sought help only in moments of acute fashion confusion, which capped lifetime value. This broke every major consumer business model. Affiliate commissions were structurally small and required near-zero CAC. Advertising demanded far higher engagement and scale. Subscriptions failed because users wouldn’t pay for an infrequent utility. A pivot to an AI-first fashion marketplace also failed due to a lack of differentiated supply liquidity at the right price points, which rendered the marketplace unviable.

---

So that’s the story behind Alle. I hope you found it useful. Like all knowledge, these learnings are fallible, and I do not assume our conclusions are definitive. They are shaped by the context we were operating in, some of which we may have misunderstood or missed entirely. It is also possible that we simply were not creative enough to solve certain problems. If you see things differently, I would genuinely love to hear from you and learn.

In addition to these, I also received a few questions around agentic shopping and quick commerce in fashion, which I will cover in a separate post.