Loblaw Digital: Talk, Shop, Discover - AI Voice Assistant for Grocery Navigation

Voice-assisted way-finding for the PC Express app and in-store kiosks. Final project for the MDEI program at the University of Waterloo.

Role: Designer / Prototyper
Timeline: 10 days

Deliverables: Figma prototype
20-minute pitch
Program: Master of Digital Experience Innovation
University of Waterloo

Summary

Mobile app prototype

In-store kiosk app prototype

Example of Lola’s in-store navigation and recommendations using the app

Loblaw Digital’s brief for our MDEI final project: how can AI help shoppers find what matters to them beyond static signage? We designed Lola, a voice-assisted guide inside the PC Express app, plus an in-store kiosk for shoppers without it.

With two teammates, we mapped out the flows for Scenario A (sign-up, preferences, lists), then I built the prototype for Scenario B, AI navigation and product discovery, including the app’s voice assistant feature. After our mentor suggested making voice the primary feature, I incorporated it into the prototype for Scenario A. I also researched competitors and ran a usability test each round.

Team: MDEI students Arwa Haggag, Evangelyn Alexander, Francesca Ferrario, Karen SueAnn Yao, Patrick Duran Yutiga, and Wahba Alsawaf, along with our mentors Markus Grupp and Kael Karlo Cruz from Loblaw Digital
Scope: Research, prototyping, two test rounds, final pitch
Platform: Mobile app (extending PC Express) and in-store kiosk app

See our full pitch deck here.

Disclaimer: Loblaw Digital and Loblaws brand elements, as well as other company images, were used for academic purposes only

The Problem

Interview findings conducted by team members

The brief: How might Loblaw Digital create a seamless and engaging way-finding system that empowers customers to efficiently locate and discover products in-store that align with their personal preferences, such as Canadian-made goods or diabetes-friendly options?

Shoppers with specific needs hunt aisle by aisle and often leave without the item they came for: it isn’t where they expected, and staff can’t always help since products move for promotions.

Our market scan, including my audit below, found apps lack real-time aisle-level navigation and most way-finding is costly to maintain.

“Customers often ask me where certain items are, and sometimes they are not in their usual spots because of promos. An app that would tell us where things are, or even if they’re in stock would be super useful”

A Loblaws store employee

“Not in a grocery store but choosing an item in the app/ in-store screen and it telling me where I can find the item exactly on the floor of the store made my shopping experience x10 better.”

A Canadian Shopper

My competitor audit

Instacart Smart Shop

Dietary preferences shape search and recommendations; tagging is automated.

Pro: Personalizes results without manual tagging.
Con: Can’t tell shoppers where an item is in the store.

Image: Instacart

ALDIgo

Checkout-free, run by cameras and scanners across the ceiling.

Pro: No app, cart or checkout line needed.
Con: Needs store-wide cameras and scanners; solves checkout, not finding products.

Image: VMSD, ALDI and Grabango release

Cust2Mate

An attachable cart panel with sensors, a scale and a store map.

Pro: Retrofits existing carts.
Con: Adds hardware to every cart to manage.

Image: Cust2Mate

Constraints

Ten days, seven team members. We ran it like a real work project with daily stand-ups around our jobs, and delivered a prototype and a 20-minute pitch with the process documented.
Different shoppers, tailored solutions. Two personas: Sara, 30, phone-fluent, and Frank, 65, avoids apps. The app works for Sara and the in-store kiosk makes sense for Frank.
A live app to extend. Our screens had to fit PC Express and Loblaw Digital’s brand while showing what was new.
Personal data and trust. Lola uses personal preferences and location, so permission had to be granted before touching any data.

The Decisions

1 — Give Lola a distinct icon

My decision, made after test round one.

After: an “L” with an audio mark and distinct colors

Before: the mic icon blended into the UI and users weren’t sure it was the AI assistant. Five of six round-one testers missed it or found it unclear.

After: an ‘L’ for Lola with an audio mark, in colors chosen to stand out and give Lola an identity. It also had to work as a floating button, since mentors pushed voice AI as the solution.

The cost: an off-brand mark. One round-two tester said its color didn’t match the brand, and another expected an assistant-like face, which we left out so no one would mistake Lola for a real person.

2 — Introduce Lola with a full page, not a tooltip

My decision, made after test round two.

Before: Lola is only introduced with a tooltip and in the chat

Before: only subtle tooltips pointed to Lola, with no context on what it is. In round two, testers still asked what it was, and one questioned whether people would want AI reading their data.

After: an onboarding intro to Lola that gives initial guidance and clarifies its role, followed by a permission request before any data is accessed.

I chose a full page over a larger tooltip or pop-up because shoppers are already setting up their profile. The cost: an extra onboarding step that could lead to drop-off.

After: Lola is introduced during onboarding with permissions

3 — A hands-free guide, with a kiosk for non-smartphone users

A team decision.

Version 1 of the three directions we took to mentors: mobile app, cart clip-on, and kiosk

We started with smart glasses, then a clip-on, and landed on the phone shoppers already carry. Glasses raise hygiene concerns, don’t suit prescription wearers, and don’t exist off the shelf. A clip-on duplicated the phone. My competitor research raised concerns for hardware costs: Cust2Mate bolts sensors onto carts, ALDIgo rigs cameras across the ceiling.

With our mentors’ guidance, we settled on Lola in the PC Express app, which shoppers can talk to hands-free, plus a kiosk for those without a smartphone. The cost: hands-free needs earbuds and a smartphone, and the kiosk is a second surface to maintain.

Craft & System

Lola: chat with voice built in

Lola started voice-first, with the chat transcript as an overlay. I changed it to a traditional chat with voice built in, so shoppers can choose chat or voice and see the transcript from the chat.

A flat route map

We replaced the isometric map with a flat one after mentors said isometric maps are typically harder to understand. It also had unnecessary detail that distracted from the pins marking where items were.

Two surfaces with their own use case

The mobile home page is tailored to each shopper’s preferences. The kiosk works with or without logging in and should encourage exploration, so it leads with discounted and trending items.

Promoting the app from the kiosk

Mentors asked how we could encourage more app downloads, so I added a pop-up with a QR code after shoppers create their list. It’s large enough to draw attention, but small enough to leave their next actions uncovered.

Ship & Collaboration

Figma screenshot of our final mobile prototype

At the end of the ten days, we delivered a Figma prototype, demo video and 20-minute pitch to the rest of our cohort and our two Loblaw Digital mentors.

A daily stand-up kept us aligned and one FigJam hub held research, flows, test notes and decisions.

The Outcome

KPIs we would measure:

Kiosk QR code scans and downloads per week, plus conversion rate
App downloads in the eight weeks before and after the kiosk launches
How often staff are asked where an item is, per shift, before and after the kiosk
Share of shoppers who find and use Lola unprompted in their first session

“This is a tool that we need in the store - both for customers and employees.”

A Canadian Superstore employee, Round-two tester

Limitations:

Accurate aisle and inventory data: a route to an empty shelf is worse than no route
Training for staff and shoppers: there’s a learning curve and not everyone knows how to use voice assistance
Clear privacy communication for data collection: one round-two tester questioned whether people would want an AI reading their data
Scaling across store layouts and languages, plus kiosk maintenance costs: stores differ, and our mentors noted kiosks require investment

Next steps:

Run pilot tests in select stores to measure the KPIs and interview shoppers about the impact on their shopping time, product discovery and satisfaction
Refine AI capabilities to balance helpfulness with non-intrusiveness, since testers missed or misread Lola and some questioned AI using their data
Gather feedback from frontline stakeholders to improve onboarding and in-store support

Reflection

Our prototype used preloaded conversations, so testers went through the motions instead of actually chatting with the AI. We could test whether shoppers found and understood Lola, but not whether its answers were useful, and we had no real store data to power a live version. Next time I’d have an AI role play as Lola against a made-up store layout, so shoppers could ask their own questions and we could test the conversation and usefulness, though not its accuracy for a real store.