Magic Bag: Rebuilding the Weedmaps Experience Around the Product, Not the Retailer

Weedmaps' cart forces customers to gamble: pick a dispensary before knowing what it stocks, then hope it happens to carry everything else on their list. I quantified how often that gamble fails — as astronomically low as 1-in-2,100 odds to find any several items a customer may want, and of the 90% of customers that shop for a 2nd item, 75% are abandoning the site.

Noticing an existing feature on certain product detail pages that reveals what retailers carry the item, I came to the idea that this awareness could be aggregated across several product selections and synthesized to propose several good retailer options to the customer. This enabled them to make an informed choice as to where to continue with a functional cart (bag), rather than guess at what retailer may have what they need. Having designed a draft experience, I pitched the “Magic Bag,” to the VP of Design and Marketplace Pillar Leads, and it was received with glowing appraisal and excitement. Unfortunately, we were laid off the following week.

The problem

Weedmaps' marketplace is built retailer-first: a shopper picks a dispensary, then hopes it happens to carry everything on their list. Finding the first product is easy — customers pick a the product from Search or the Catalog, and start a cart with the retailer where it’s in stock. Finding the second and third product is a different story, rarely would a retailer have the second or third item or anything comparable, and nothing in the marketplace experience could tell customers that going in to creating that first cart. So, customers were forced to abandon that cart and start over with a new retailer, hoping they’d have a combination of products that worked for them.

I wanted to know exactly how bad it was, so I pulled our own Snowflake session data (1.2M sessions, 30-day window) and a live snapshot of all 3,233 active California retailer menus, and ran the numbers myself rather than relying on anecdote:

• A typical big-brand product is carried by only 2.5% of CA retailers; a niche product, just 0.25%.

• Once a shopper finds a store with product #1, the odds it also has #2 and #3 range from 1-in-57 (mainstream products) to 1-in-2,100 (niche products).

• I ran a real case study for a 10-mile radius around a point in Irvine, CA with three real products: of 209 nearby retailers, only 3 carried all three — a 1.4% blind-pick success rate.

• 90% of shoppers who add one product keep browsing for a second instead of checking out immediately — and 75% of those give up and leave without ordering anything at all.

This wasn't a UX polish problem, it was a structural one, and it was costing us tens of thousands of sessions a month in California alone.

The decision

Before proposing a fix, I brought in another designer from the team to go along with me in the journey of exploring the problem, to be an interlocutor, and work on the design with me. Mari Pearson worked in the “Order Success” pillar, so we got the nod of approval from her partner PM and the pillar leaders to investigate and come back with recommendations. We deliberately weighed what appeared at first to be cheaper options (more on that below). Ultimately, Mari and I concluded they weren't worth building: they required roughly the same engineering lift as a real fix, while only patching the symptom (conflict modals, repeated searches, restarted carts) instead of removing the retailer-first bottleneck causing them. If we were going to spend the effort, it needed to buy a structural improvement, not a cosmetic one.

That shined a spotlight on the Magic Bag: flip the flow so customers build a bag of products first, with no retailer attached, and then see which retailers can fulfill the whole bag — sorted by price, speed, or rating. The cart isn't created until the customer picks a retailer to check out with, so the sequence becomes add → compare → choose, instead of choose → hope → restart.

I designed ways to account for the flow's edge cases up front rather than leaving them for later:

• Loyal shoppers who want a specific dispensary get a half-sheet confirming their bag matches that store, so we don't force a comparison nobody asked for.

• Low-inventory areas get an adjustable search radius, since in less dense regions the fix isn't "find more retailers," it's "widen the net."

I also made a point of framing this to leadership as a surface migration, not a new build — the same in-stock-retailer logic already existed on individual product pages; Magic Bag just extends it across multiple products at once. That framing mattered for a VP-level pitch: it's a much easier "yes" when the ask is "extend logic we already have" rather than "build something new."

Evaluating the options

Rather than presenting Magic Bag as the only viable answer, I benchmarked it against three other cart models against the jobs customers actually needed done — finding a specific product across the market, browsing broadly, and comparing prices across retailers:

• Singular Bag (Weedmaps today) — retailer-first, one store's menu at a time. Also add better signaling — warnings when when the user began to shop other stores to avoid abandoning the original cart, clearer prompts to search again within the menu of the retailer in which they’d started a cart, or modify search to work only within a store you've already committed to, etc.

• Multi-Bag (Instacart's model) — build separate bags per store and compare them side by side, which still requires visiting each retailer's bag independently to check availability. Doesn’t solve the low-odds issue.

• Mixed Bag (DoorDash's DoubleDash) — a single cross-retailer order, but built for market-wide fulfillment rather than surfacing which retailers can complete a specific list. Not possible given the current structure and licensing of delivery services. Doesn’t solve for the low-odds issue with pickup retail locations.

• Magic Bag (the proposal) — product-first, using the same in-stock-retailer data our product pages already show, extended across a whole bag at once.

Scoring all four against discovery and comparison-shopping jobs made the gap explicit: Singular Bag and Multi-Bag only partially serve product discovery and price comparison, and Mixed Bag serves discovery well but isn't built to help a customer compare prices for the same item across retailers. None of the three alternatives were wrong approaches — they're legitimate patterns other marketplaces use successfully — but each left a real gap for our customers' specific job (finding and comparing a multi-item basket across nearby dispensaries). Laying them out side by side in the pitch let me show leadership that Magic Bag wasn't the only option on the table, just the one that didn't leave a job underserved.

The pitch and projected impact

I set the primary target as a 20%+ reduction in abandoned carts — the metric that actually reflects what Magic Bag fixes, versus a GMV or take-rate number that doesn't map to how Weedmaps makes money.

Weedmaps doesn't earn a fee on completed orders; retailers and brands pay for attention — sponsored placements and sponsored product listings. That means Weedmaps' business doesn't grow by capturing a cut of each cart, it grows by keeping more customers actively shopping the marketplace: every session that survives past product #1 instead of abandoning is a session where a sponsored placement or listing can still do its job. Retention and new-customer acquisition are the real growth levers, because a bigger, more engaged audience is what Weedmaps actually sells to retailers and brands.

Modeled against the abandoning cohort, cutting abandonment by 20%+ translates to:

While the metric the project would be accountable to is the 20%+ reduction in cart abandonment and the resulting Daily Unique Shopper retention gain, the fix carries real material upside for our retail clients too.

Given AOV ranges from $75 (conservative) to $95 (optimistic) — multi-product baskets run larger than the current platform average — the increase in order success translates to retailers and brands seeing $12M–23M in incremental annualized GMV directly. That's revenue landing in retail partners' pockets, not Weedmaps'; it's the case for why they keep investing in sponsored placements and listings on a marketplace that's actively growing their sales.

I pitched this directly to Weedmaps' VP of Design and the Marketplace Pillar Leads, positioning Magic Bag as the direct answer to a problem we could now put a number on instead of describing in vague terms — both what it recovers for Weedmaps' audience and what it puts back in retail clients' hands. The pitch was received with positive acclaim, but we were laid off the following week.

Why this matters

This project is less about the Magic Bag interaction pattern itself and more about how I got there: turning an anecdotal complaint into a quantified, data-backed business case using data I pulled myself (then confirmed with my Data team partners); explicitly weighing and rejecting alternatives with a clear rationale instead of defaulting to the path of least resistance; and tying the design case to the metric that actually maps to how the business makes money. Weedmaps doesn't monetize completed carts — it monetizes attention, sold to retailers and brands as sponsored placements and listings. So the design case had to argue in terms of retention and acquisition, not GMV, because a bigger and more engaged customer base is the real asset being grown. That's the muscle a Head of Product Design role needs — not just "here's a flow I designed," but "here's how I diagnosed the real problem, ruled out the wrong fix, and tied the fix to the metric leadership actually cares about."