In June 2026, Shipt launched Shipt Kits, a one-click shopping feature that lets customers add complete themed grocery bundles to their carts, after building the entire platform in just three weeks. The company timed the rollout around a global soccer event through a national brand campaign featuring celebrity partnerships, according to a technical case study published on InfoQ. The system replaced a manual workflow that previously required merchandisers to spend days querying databases, translating product codes across retailers, and building bundles with no guarantee they'd display correctly at individual stores.
Before Shipt Kits existed, assembling a themed shopping experience meant merchandisers had to query the data warehouse for products, manually select relevant product codes, translate each code to retailer-specific IDs for Target, Meijer, Publix and other chains, build a category from those IDs, construct a product list from that category, and finally integrate the list wherever the bundle needed to appear. The API was completed in ten days, while the remaining eleven days went to the admin interface, customer-facing surfaces, and operational hardening needed to let merchandising teams operate the system without engineering supervision. The platform now materializes store-aware kits automatically—customers see only items actually stocked at their local store, and kits disappear entirely from a customer's view if essential "anchor items" like meat for a grilling kit or pasta for an Italian dinner kit aren't available at that location.
The architecture rests on three composable entities that share data by reference rather than duplication: items (atomic catalog units tagged with metadata like dietary restrictions and seasonal themes), shopping lists (curated sequences of items used for both merchandiser-assembled kits and customer-built grocery lists), and kits (the customer-facing bundles carrying editorial content, images, and references to items and lists). According to author Khurshid Hameed, the platform's most important property is that items are shared by reference across all surfaces—the same Heinz ketchup item powers multiple bundles and customer lists simultaneously, so updating one item's metadata or adding a new retailer mapping lights it up everywhere it's referenced. The system determines which kits to show a customer by calculating the intersection of three independently configured sets: retailers serving the customer's zip code, retailers the kit is enabled for, and stores meeting any special eligibility requirements like active alcohol licenses for alcohol kits.
The team modeled regulatory and compliance requirements as distinct kit types rather than flags scattered through shared code paths. Alcohol kits carry their own eligibility predicates that automatically check store licensing, filter to alcoholic products only, and hide kits when customers fall outside a retailer's licensed delivery area—merchandisers don't need to remember compliance rules because the kit type enforces them automatically. The case study notes this approach makes new regulated categories like pharmacy items or age-restricted content additive rather than compounding the complexity of existing kit types, since each new category becomes its own type with isolated rules. For brand-sponsored kits where consumer packaged goods companies pay for prominent placement, the platform includes a configurable setting that lets merchandisers choose whether to hide just the sponsored item or the entire kit when that product isn't available at a customer's store, matching whatever the brand's contract requires.
Because the three-week deadline didn't allow time for a full draft-publish workflow with version control and diff displays, the team shipped four layered guards instead: a persistent warning banner on published kits, a first-edit confirmation modal that reprompts in each new browser tab, explicit Save and Discard buttons per section with unsaved-changes indicators, and cross-section locks that prevent instant actions when other sections have unsaved edits. Hameed writes these four layers deliver roughly 70% of what a true draft-publish system would provide at about 10% of the engineering cost—enough to ship safely under real time pressure—and the team has since added a Testing status that makes kits live in production but visible only to internal reviewers before customers see them. The next planned investments include completing the draft-publish lifecycle with per-field drafts and audit trails, closed-loop revenue attribution for brand-sponsored kits, and per-customer filtering so shoppers with dietary preferences see bundles tailored to their restrictions. For teams building similar bundled-product platforms, the report emphasizes picking a small set of composable entities and resisting pressure to add more, modeling real-world constraints as types rather than boolean flags, and shipping enough to learn from actual usage patterns rather than designing against assumptions. Platforms that let users combine multiple items into purchasable units—whether grocery kits, travel packages, or subscription boxes—face pressure to proliferate entities and special-case logic, but composition through shared references tends to age better than duplication across disconnected systems.