Allbirds teardown: an AI shopping assistant on a store that keeps saying no

Allbirds is a direct-to-consumer brand that now has to win almost entirely online. Its FY2025 10-K reports net revenue of $152.5 million, down 19.7% from $189.8 million, and a net loss of $77.3 million. The same filing says the company closed its remaining US full-price stores in the first quarter of 2026, says most of its sales come through direct channels, and mentions a discontinued mobile app. So the website is carrying more of the business than it ever has.

I spent some time on allbirds.com on October 28, 2026 and took screenshots as I went. This is what I saw, what I would change, and why.

What Allbirds does well

The AI is visible, and it is in the right places. There is a "Shop with AI" button pinned to the bottom of the homepage, an "Explore this product" chip on the product page, and an "I can help narrow things down" chip on search results (screenshots taken Oct 28, 2026). A lot of brands bury their assistant in a chat bubble nobody opens. Allbirds puts a different prompt on each page type, which tells me someone thought about intent per page.

The product page answers the fit question in one line. Under the size grid on the Men's Tree Runner NZ it says the shoe "fits true-to-size for most customers," with a link to a fit guide (screenshot taken Oct 28, 2026). Fit is the main reason people hesitate on shoes online, and this is the cheapest possible fix for it.

Search tags tell you what you are looking at. A search for "waterproof" returned 75 results, and each card carries badges like WATERPROOF, WATER-REPELLENT, FINAL FEW, and SALE (screenshot taken Oct 28, 2026). That is good product data showing up where it helps.

What Allbirds does badly

The store says no a lot, and the AI does not help with that. On the Men's Tree Runner NZ in Medium Grey, only 3 of 14 men's sizes were available: 8.5, 9 and 11 (screenshot taken Oct 28, 2026). The page has a FINAL FEW badge and the button reads "Select a size." If your size is crossed out, you are done. The AI chip is sitting right there, but nothing connects it to the moment you hit a sold-out size.

The shipping warning is in the wrong place. The rotating banner at the top of the homepage said "Due to increased demand, orders may take up to 30 days to ship" (screenshot taken Oct 28, 2026). On the other pages I captured, the banner showed "Free ground shipping on orders over $100" instead. A 30-day wait is the kind of thing a shopper needs to know at the size picker and in the cart, not in a banner slot that rotates away.

The first thing a new visitor gets is a popup. On a fresh visit, a modal offered 10% off in exchange for email and SMS signup, on top of the hero (screenshot taken Oct 28, 2026). After you close it, a "GET 10% OFF" tab stays pinned to the bottom left of every page I visited. I understand why. Email is cheap revenue. But it competes with the AI button for the same corner of the screen and the same few seconds of attention.

Search is broad when the shopper is specific. Someone typing "waterproof" usually means it. The results included WATER-REPELLENT items and SALE items in the second row (screenshot taken Oct 28, 2026). The badges are honest, which I like, but the shopper still has to do the sorting. The AI chip offers to narrow things down, but only if you click it.

Help is an email address. The help page header says "Email us at help@allbirds.com," and below it is a returns FAQ (screenshot taken Oct 28, 2026). I did not see live chat or a phone number on that page. For a brand that has closed its US full-price stores and is asking the site to do more, the help page is where a working assistant would pay for itself.

How I'd improve it

My opinion: Allbirds has already paid for the hard part, which is an assistant that can see the catalog. The work now is pointing it at the three places where the store currently says no.

1. Turn a sold-out size into a choice, not a dead end. When a shopper taps a crossed-out size, the assistant should open with options it can actually back up from inventory: the same size in colors that are in stock, a back-in-stock alert, or help choosing between two sizes using the fit note the page already has. This is a small build. The data is already on the page.

2. Put the ship date where the decision happens. If orders can take up to 30 days, say so next to the add to cart button, as a date, from the same inventory data. Surprises after checkout turn into support email, and support is email only.

3. Make the AI ask the one question that splits the results. For "waterproof," that question is "fully waterproof, or is water-repellent fine?" Asking it up front, inline above the grid, is a better use of the assistant than a chip that waits to be clicked.

4. Put the assistant on the help page. Order status, returns, and "where is my order" are the questions an assistant handles best, and the help page is where people bring them. Email can stay as the fallback.

5. Give the signup offer its own moment. Show it after someone has looked at a product or two, not as the first thing on screen, and keep it out of the corner the AI button lives in.

None of this needs a new platform. The 10-K names Shopify among the third-party partners Allbirds relies on, and in my experience everything above is the kind of change you can ship on that kind of stack in weeks, not quarters.

Technology detected

I fetched allbirds.com on October 28, 2026 and listed only what I could verify from the HTML, response headers, script hosts, cookies, and /cart.js. If a vendor was only named in a disabled config flag, I left it out.

Engineer's take

Shopify is doing the job a theme is supposed to do. Catalog, cart, Shop Pay, and a pile of app embeds all sit on one storefront. That is why a merchandiser can ship Attentive, Clarity, and a custom DNAM theme without standing up a new app.

The friction is that the useful data is already on the page and the AI chips are not using it. The size grid knows which sizes are out. The banner knows a 30-day ship warning exists. Those are theme and inventory facts, not a missing platform. Wiring the chips to a crossed-out size is theme work.

What I would watch: the custom theme plus GTM, Amplitude, Attentive, and Clarity on every page. That is a lot of JavaScript in the same viewport as the signup tab and the AI button, which already compete for the bottom of the screen in the screenshots. I would keep new embeds off the product page until the sold-out path is wired.

Architect's take

I would not replatform this. The structural move is to treat inventory, ship date, and help as one system the assistant can read. Shopify already holds the stock and the cart. The theme already renders the size grid. The missing piece is a single path from "this size is gone" or "where is my order" into that data, instead of a chip that waits and a help page that is only email.

That is an AI layer on top of a store that already knows how to say no. The question is not whether Allbirds has an assistant. It is whether the assistant can see the same facts the theme already has.

If you run a commerce brand and want this kind of review pointed at your own store, it is what my AI Opportunity Teardown is for. For $2,000 I spend a week on your site and data, and you get a written one-pager and a 15-minute Loom walking through where AI would actually move revenue and where it would not.