# Fabular vs. Alfies vs. Gurkerl — storefront & AI-commerce mechanics benchmark **Date:** 2026-06-06 · **Genre:** competitive mechanics audit (not a scrape-data audit) · **Method:** live inspection of all three storefronts via Playwright (DOM extraction + assistant interaction), 2026-06-06. **Subjects** - `fabular.pages.dev` — "**Fresh Haven**", the fab4minds **fabular eCommerce platform** reference storefront ("powered by fabular platform · fab4minds fabular eCommerce API"). - `alfies.shop/de/at/shop` — **Alfies**, Austrian full-range supermarket delivery (60-min). - `gurkerl.at` — **Gurkerl**, Austrian online supermarket (Rohlik Group, 3-h delivery). --- ## 0. Read this first — two are businesses, one is a platform demo The footer says it plainly: Fresh Haven is a **storefront demo for a platform**, not an operating grocer. Alfies and Gurkerl are funded businesses with real supply chains (17 000+ SKUs, 500 000+ customers, sub-hour logistics, loyalty infrastructure). A flat "who wins" table would conclude *fabular loses on assortment / speed / loyalty* — true and useless: that scores a tech demo's 40-product catalog against two grocers' warehouses. So the benchmark splits in two layers, and only the second is fabular's actual arena: | Layer | What it measures | Is it fabular's fight? | |---|---|---| | **A — Operational scale** | catalog depth, delivery speed, loyalty infra, fulfilment | **No** — business maturity, not platform capability | | **B — Storefront & AI-commerce mechanics** | UX patterns, conversational/agentic commerce, personalization, merchandising | **Yes** — this is what the platform ships to every tenant | The blueprint (§4) lives entirely in Layer B. --- ## 1. Layer A — operational scale (context, not scoreboard) | | Fresh Haven (fabular) | Alfies | Gurkerl | |---|---|---|---| | Positioning | Regional/Bio, direct-from-producer | Full supermarket + Italian/specialty | Full supermarket, "best of the city" | | Catalog | ~40 demo products | "thousands" | **17 000+** | | Delivery | **48 hours** | **60 minutes**, 6 days/wk | **3 hours**, 15-min slot window | | Free-delivery threshold | ab 49 € | min. order 25 € | first order > 39 € free | | Social proof | 4,8/5 (3 563) | — | **4,9/5 (12 471+)**, 500 000+ customers | | Loyalty / subscription | **none** | minimal | **Gürkchen-Club** + **Xtra** sub (free delivery + 10 % bio) + referral | | Food-waste play | none | Mehrweg/Pfand options | **"Rette Lebensmittel"** discount aisle + own-brand discount line | **Takeaway:** fabular's 48-h cadence is a *deliberate model* (direct-from-farm box), not a loss to Alfies' 60-min q-commerce — different business. The real gaps that a *real* fabular tenant would inherit and must close are **loyalty/subscription** and **food-rescue merchandising** — both are pure software mechanics, both ship as nothing today (§4). --- ## 2. Layer B — the AI-commerce battleground (headline finding) All three now ship an AI shopping assistant. This is where the benchmark is decided — and the naive read ("fabular has Jule, it's ahead") **inverts on inspection**. | | **Jule** (Fresh Haven) | **Maia** (Gurkerl) | **Alfies AI** | |---|---|---|---| | Surface | on-site floating panel | on-site panel | dedicated page `/alfies-ai` | | Catalog-grounded | ✅ **verified** — returned real SKUs w/ price + "Angebot/Beliebt" tags | ✅ tool-calls live ("Suche nach den besten Rezepten…"); advertises product search | gated behind cookie/address wall (not verified) | | Recipe / meal ideas | ✅ 3 modes (Rezepte / Produkte / Budget) | ✅ recipes + **personalized meal plans** | "Meine Produkte" personalization | | Manage cart / orders | ❌ recommends only | ✅ **cart, lists, orders, favourites** | ❌ (browse) | | Image / photo input | app-exclusive ("Foto-Scan") | ✅ **on web** — upload photo of ingredients/list | — | | **Off-site agentic (MCP)** | ❌ | ✅ **public MCP server** | ❌ | | Disclaimer hygiene | ✅ allergy/medical | ✅ allergy/medical | — | ### 2.1 The decisive mechanic — Gurkerl's MCP server Gurkerl publishes a public **Model Context Protocol** endpoint: `https://mcp.gurkerl.at/mcp/` (footer link → `/de-AT/seite/mcp-server`). It lets external assistants — **ChatGPT, Claude Desktop** — shop Gurkerl programmatically. Advertised tools, verbatim: - *Produktsuche* — find products by name/category/dietary preference - *Warenkorbverwaltung* — add/remove/update cart items - *Bestellinfos* — order status & history - *Angebote & Aktionen* — discover current deals - personalized recommendations Setup is literally "edit `claude_desktop_config.json`, add credentials, restart." This is **off-site agentic commerce** — a stratum *above* an on-site chat widget. The store becomes a tool any agent can call; the buyer never visits the site. Jule is a good on-site conversational layer; it does not exist where the customer's agent lives. ### 2.2 Corrected headline > On the AI axis, **Gurkerl is ahead of fabular, not behind.** Maia matches Jule's > grounding and exceeds it on cart/order actions and on-web image input; the MCP server > opens a channel fabular has no answer to. Fabular's *verifiable* edge is narrower: > Jule is cleanly catalog-grounded and the only assistant wired natively into a > **platform-generated** storefront (every tenant gets it for free). That is the real, > defensible fabular advantage — productized AI commerce as a platform primitive, not a > per-merchant integration project. *Caveat logged:* Jule overshot its own budget constraint — asked for "vegan dinner for 2 **under 15 €**" it led with a 24,99 € veggie box. The "Smarter Einkauf / budgetoptimiert" claim is aspirational in the demo, not enforced. Alfies' assistant depth is **unconfirmed** (cookie/address gated) — scope any Alfies-AI claim accordingly. --- ## 3. Mechanics catalog — who does what **Where fabular's storefront already matches or leads (Layer B):** - **Conversational commerce native to the platform** — Jule ships with the storefront, no integration. Neither competitor's assistant is a reusable platform product. - **Producer/regional narrative as merchandising** — "Kulinarisches Italien/Wien/Bayern", "Unsere Produzenten", per-brand panels (Voelkel). Richer storytelling than either grocer. - **One-click recipe→cart bundles** ("Mit einem Klick ins Sortiment") — parity with Alfies' *Themenwelt* ("Ein Klick bis zum BBQ"). - **Onboarding personalization** — "Dein persönliches Sortiment in 2 Minuten." - **App differentiation story** — push, app-exclusive assortment, Foto-Scan ordering. - **Polish** — clean Next.js build, DE/EN, trust badges, testimonials, PLZ gate. **Where the incumbents have mechanics fabular's storefront lacks:** - **MCP / agentic channel** (Gurkerl) — §2.1. - **Cart/order-managing assistant + on-web image upload** (Maia) — Jule recommends only. - **Loyalty + paid subscription + referral** (Gurkerl: Gürkchen-Club, Xtra, Empfehlung). - **Food-rescue & discount-tier merchandising** (Gurkerl: Rette Lebensmittel, Gurkerl-Günstig own-brand line) — margin + sustainability + traffic in one mechanic. - **Dark mode** (Gurkerl) — Inter, system/light/dark toggle. - **Pure transactional storefront mode** (Alfies) — lands the user *in the product grid*, no marketing scroll; multipack/carton deals (− 6 € je Karton), price-per-unit, Pfand. - **Live trust telemetry** (Gurkerl) — "12 471+ ratings", "500 000+ customers" surfaced. --- ## 4. Blueprint — what the fabular platform should ship Ordered by leverage. Each is a *platform* feature (ships to every tenant storefront), not a one-off. "Edge kept" = defends fabular's existing advantage; "Gap closed" = neutralizes an incumbent mechanic. ### Tier 1 — frontier (close the channel gap before it's table stakes) 1. **Ship a fabular MCP server as a platform primitive.** One endpoint spec, auto-provisioned per tenant, exposing `product.search / cart.* / order.status / offers.list / recommendations`. This is the Gurkerl-parity move *and* the productization fabular is uniquely positioned for: a grocer builds an MCP server once as a project; **fabular ships it to N tenants as a checkbox.** That asymmetry is the platform's whole thesis. *(Gap closed: Gurkerl MCP. Edge kept: platform-native.)* 2. **Promote Jule from recommender to agent.** Add cart/list/order tools so it *acts*, not just suggests. Reuse the same tool layer the MCP server exposes — one tool contract, two front doors (on-site chat + off-site MCP). *(Gap closed: Maia's action depth.)* 3. **Bring Foto-Scan to the web** and actually enforce the budget mode. Image-to-cart on web (Maia already does it), and make "Smarter Einkauf" honor the stated € ceiling — today it doesn't, which undercuts the headline feature. *(Edge kept, made real.)* ### Tier 2 — commerce mechanics every tenant needs 4. **Loyalty + paid-subscription module** (points/club + a Xtra-style "free delivery + X % off own-range" tier + referral). Pure software; fabular ships none. Biggest Layer-A gap a real tenant inherits. 5. **Food-rescue / discount-tier merchandising** — a "Rette Lebensmittel" surface + an own-brand/discount price tier. Margin, sustainability story, and repeat traffic in one. 6. **Storefront mode switch: narrative ↔ transactional.** Fresh Haven is all marketing-scroll; Alfies is all grid. A platform should let a tenant pick (a Bio box-scheme wants narrative; a q-commerce tenant wants the grid). Ship both as a layout toggle. ### Tier 3 — polish & parity 7. **Dark mode** as a theme token (Gurkerl has it; cheap, expected). 8. **Live trust telemetry** — surface real rating counts / customer counts, not static badges. 9. **Per-unit price + multipack/carton deal rendering** (Alfies) as a product-card variant. ### What NOT to copy - Don't chase 60-min q-commerce — fabular tenants like Fresh Haven are a *different model* (direct-from-producer, planned delivery). Speed isn't the axis; **agentic reach and platform-productized AI** are. --- ## 5. The one-line thesis > Three storefronts, three AI assistants — but only fabular's is a **platform primitive**. > Gurkerl proved the frontier (assistant that *acts* + an MCP server agents can call); > fabular's win condition is to ship that frontier **as a product every tenant gets for > free**, not as a feature one merchant built once. Match the agentic depth, expose the MCP > channel, keep the platform asymmetry. Everything else (loyalty, food-rescue, dark mode) is > catch-up software that a real tenant needs regardless. --- ### Sources - Live inspection (Playwright DOM + assistant interaction), 2026-06-06: `fabular.pages.dev`, `alfies.shop/de/at/shop` + `/de/at/alfies-ai`, `gurkerl.at`. - Gurkerl MCP server page: `https://www.gurkerl.at/de-AT/seite/mcp-server` (endpoint `https://mcp.gurkerl.at/mcp/`), fetched 2026-06-06. - Jule grounding probe: query *"Schnelles veganes Abendessen für 2 unter 15 Euro"* → returned Gemüsekiste Bunt 24,99 €, Avocado 2,99 €, Kirschtomaten 2,49 €, Olivenöl 6,99 €. - Maia capability list + live tool-call observed (recipe search) on same query. - Screenshot: [`assets/2026-06-06-fresh-haven-home.jpeg`](assets/2026-06-06-fresh-haven-home.jpeg) (full-page Fresh Haven capture). **Related:** [`tasks/0002-fabular-benchmark.md`](../../tasks/0002-fabular-benchmark.md) (the `_ref-fabular` digital-health benchmark — *this* doc is its UX/AI-mechanics companion).