Ask Navy
Conversational shopping, built around the customer’s task.

Ask Navy’s broader scope covered product discovery and shopping assistance. Jingle Jammies made the challenge concrete: a shopper could build a matching set for several people, then discover the final child’s size was unavailable. Choosing another print meant repeating much of the work.

The objective was to help customers assess family coverage earlier, understand available alternatives and revise a selection while retaining the rest of their request. Matching meant a common print, with possible flexibility in pieces, sizes and fits.

Consumer Insights’ May 2026 study included 3,004 U.S. apparel shoppers, including 1,136 Old Navy shoppers. Among Old Navy respondents, 36% were very likely to use a styling assistant for coordinated holiday or family-photo outfits. Open-ended responses described seeking family looks and shopping for children.

I used these findings alongside UX problem framing and prototypes to sharpen the pilot. The research established interest in the shopping mission; the exact interaction and availability behavior required further product definition.

The October 8–November 8, 2025 customer analysis recorded 1,604,716 Jingle Jammies customers. With $123 in Old Navy spend per customer and $76 in basket halo spend, the category represented a substantial existing shopping mission and basket-building opportunity.

I initially recommended Constructor’s assistant for the holiday pilot. My comparison emphasized in-chat cart actions, routing to matching-family collections, merchandising control and the dependencies in Google’s proposed MVP.

I used that assessment to push Google’s scope toward guided Jingle Jammies shopping and QuickShop. The assistant-vendor decision was distinct from the underlying catalog integration: Constructor also supplied product and availability responses within the Google-based solution.

A family of five could otherwise require five separate product-page handoffs. I made the commercial and experience case for QuickShop so customers could select product options and add items while keeping their conversation available.

The distinctive product work was bringing structured, visual elicitation and commerce actions into the same conversation. For family shopping, this required coordinating inputs across people and products while preserving context as customers moved toward a purchase.

The requirements separated the quiz and QuickShop into independent releases: guided discovery with PDP links first, then Old Navy-built QuickShop. This allowed the quiz and filter chain to progress while commerce integration continued.

I defined a print-first sequence, followed by family composition and sizes. Each step passed its selections to Constructor before the next rendered, narrowing eligible options as the customer progressed. The quiz remained optional; conversational inputs used the same filter logic, and partial changes retained prior selections.

I constrained the MVP to three linear steps and one partial-match fallback: matching bottoms. If no match remained, print choice reset while family and size inputs stayed intact. Split top/bottom sizing, piece-type questions, ranked fallback tiers, member weighting and adaptive branching stayed outside the MVP.

The mapping work connected internal print names, customer-facing “Match the Fam!” labels, seasonal flows, divisions and product-color identifiers. That relationship let one shared print resolve into products for different family members.

The quiz requirements described up to 22 prints across seven customer divisions, with approximately 188 core apparel SKUs in a 213-SKU snapshot. A separate planning workbook contained 301 style/color records. Those records described assortment relationships; current size and stock eligibility came from catalog responses.

Google owned quiz rendering, filter orchestration and Constructor API calls. Old Navy defined the questions, sequence and options. Constructor responses supplied available prints, divisions and product options, including out-of-stock suppression.

Print selection scoped divisions; family selection established whom to serve; sizes narrowed product results. The requirements specified full matches, a matching-bottoms fallback or a print reset. Progressive filtering supported the MVP without a separate family-set inventory engine. Constructor’s behavioral personalization was outside the Google API scope, so explicit quiz inputs supplied the family context.

A quiz size was a discovery filter. Whether QuickShop inherited it or required another size/color selection remained an integration decision. I made that boundary explicit rather than assuming a recommendation contained everything needed for a cart commitment.

Old Navy owned QuickShop behind a separate feature gate. Google’s onboarding contract documented merchant-side cart callbacks, including handleGetCart and handleCartUpdate. The interface and commerce teams therefore had distinct responsibilities for recommendations, option selection and cart updates.

I built two HTML prototypes on the same family-matching model. Prototype A organized the task as an on-site guided quiz; Prototype B placed starter prompts and a pinned question card inside an assistant-style conversation. Both made print, family and size selections concrete enough to review with partners.

A seasonal control simulated sizes selling through. With family inputs held in place, the guided prototype changed the recommended pieces and set totals, then offered contextual swaps. I used this behavior to make availability, matching flexibility and recovery decisions visible alongside the interaction design.

The guided exploration informed the on-site experience. The current-state storyboard and complexity synthesis helped frame the shopper problem; I translated it into triggers, dismissal rules, partial revisions and result-state requirements.

I organized a Consumer Insights and UX follow-up to request Old Navy-specific family and occasion findings, parent/non-parent cuts, gifting intent and the two or three customer journeys the study most strongly supported.

For example, 64% of Old Navy respondents were willing to share their own size with an AI shopping assistant. My follow-up questioned how that translated to a retailer-specific experience and information about other family members. That distinction helped define what the pilot still needed to learn.

I served as Old Navy’s main point of contact within a group of 13 named leads and contributors across Old Navy, Gap and Google. The group spanned product, architecture, UI and backend engineering, UX, analytics, AI strategy and technical program management.

The journey deck defined 15 review scenarios: ten general discovery journeys and five Jingle Jammies journeys. I coordinated product scope, requirements and readiness alongside quiz finalization, GCP and token-service readiness, UX handover, UI development and staging. Google and Gap’s technology teams owned production implementation.

The initial pilot discussion centered on discovery and a product-page handoff. The later public demonstration shows visual pattern choices, family selection, size and fit inputs, recommendations and in-chat QuickShop with Add to Bag.

My contribution covered the platform recommendation, business case, experience definition, requirements and prelaunch readiness. Gap announced the Old Navy experience on October 5, 2026, after my September departure.

The planning scenario used 15.5M weekly US visits and a $60.41 average order size. Over eight weeks, at 50% exposure, a 3-basis-point absolute conversion improvement implied 18,600 incremental orders and approximately $1.12M in demand. Each additional basis point added approximately $375K under the same assumptions.

I used the scenario to show how exposure and conversion lift changed the opportunity. It was an incremental-demand forecast; approved project costs and measured postlaunch uplift are not available in the reviewed sources.

The measurement framework covered adoption, discovery, family matching, basket quality, conversion and experience health. For the quiz, the requirements emphasized full-family-set match rate and add-to-cart behavior. Stock confidence and recovery mattered alongside commercial performance.

Quiz-completer comparisons would provide directional learning; a controlled evaluation would be needed to attribute incremental lift. The case’s measured postlaunch commercial results remain unavailable. The product lesson was to connect guided inputs, catalog constraints and commerce actions throughout the journey.