Favourites required manual curation
Customers were using Favourites as a workaround for rebuilding regular shops, but they still had to maintain the list themselves and decide what was relevant each time.
01 Case Study
QuickShop.
Reducing weekly grocery friction through personalised basket building
A behaviour-led basket-building experience designed to help customers rebuild their regular shop faster using purchase history and recommendation confidence.
- Role
- Senior Product Designer
- Company
- Waitrose & Partners
- Timeline
- 8 months
- Focus
- Predictive personalisation / Behavioural UX / Commerce
My role
I led the translation of Data Science’s Basket Building Model into a customer-facing proposition. I defined how recommendation confidence influenced hierarchy, grouping and progression; designed and tested three interaction models; created a data-connected prototype capable of representing different customer behaviours; and worked with Product, Data Science and Engineering to narrow the initial release to Top Regulars.
- Translated behavioural model outputs into interface hierarchy, grouping and progression
- Designed and compared three personalised basket-building interaction models with UX research
- Built a data-connected prototype to test recommendation behaviour across customer datasets
- Worked with product, data science and engineering to shape and ship a focused Top Regulars MVP
02 The challenge
Repeat shopping still felt like starting again.
Most online grocery experiences treated every shopping session as a new browsing journey. Regular customers still searched, navigated categories and rebuilt similar baskets week after week.
Favourites preserved historical purchases, but historical behaviour did not always reflect current intent. As saved products accumulated, finding what mattered now became slower — especially when diets, households or routines changed.
QuickShop explored whether predictive personalisation could make routine shopping faster, clearer and more controlled without reducing customer confidence.
03 Existing behaviour
Why regular shopping still took too much effort.
Customers already had ways to revisit familiar products, but each route placed work back onto them. The opportunity was not simply to create another list. It was to make rebuilding a routine shop faster and more relevant to the customer’s current needs.
Previous Orders required searching backwards
Only 2.4% of total add-to-trolley actions came through Previous Orders. Customers had to open historic orders and identify what they wanted to buy again.
QuickShop could predict what was relevant now
Rather than asking customers to maintain a list or browse previous baskets, QuickShop used recurring purchase behaviour to prioritise products they were most likely to need.
QuickShop shifted repeat shopping from remembering and searching towards recognising and confirming.
04 Research
Understanding when personalisation stopped feeling personal.
Working closely with the UX researcher, we combined behavioural evidence, customer conversations and prototype testing to understand where the existing experience broke down and which interaction model customers trusted most.
05 Participants
Testing across different shopping routines.
We compared the concepts with six experienced Waitrose online customers whose basket sizes, shopping responsibilities and routines varied substantially. This helped us test whether the interaction model worked beyond one narrow type of repeat shopper.
Participant 01
- Segment
- Platinum Loyals
- Indicative basket
- £200+ indicative recent basket
- Shopping responsibility
- Primary household shopper
- Online grocery experience
- Experienced online grocery customer
Participant 02
- Segment
- Loyals
- Indicative basket
- £60–£80 indicative recent basket
- Shopping responsibility
- Shares household shopping
- Online grocery experience
- Regular Waitrose online shopper
Participant 03
- Segment
- Loyals
- Indicative basket
- £160–£180 indicative recent basket
- Shopping responsibility
- Primary household shopper
- Online grocery experience
- Experienced online grocery customer
Participant 04
- Segment
- Loyals
- Indicative basket
- £100–£120 indicative recent basket
- Shopping responsibility
- Primary household shopper
- Online grocery experience
- Regular Waitrose online shopper
Participant 05
- Segment
- Loyals
- Indicative basket
- £100–£120 indicative recent basket
- Shopping responsibility
- Shares household shopping
- Online grocery experience
- Experienced online grocery customer
Participant 06
- Segment
- Loyals
- Indicative basket
- £120–£140 indicative recent basket
- Shopping responsibility
- Primary household shopper
- Online grocery experience
- Regular Waitrose online shopper
06 Insights
Customers wanted speed, but not at the cost of control.
01
Historical behaviour could become stale
Products previously bought often remained visible after diets, preferences or household routines had changed.
02
Relevance determined trust
A small number of visibly unsuitable recommendations could undermine confidence in the entire experience.
03
Structure reduced effort
Participants found a staged journey easier to understand than one long, continuous product feed.
04
Related products needed to remain comparable
Customers evaluated variants, pack sizes, offers and substitutes together, even when the recommendation model scored them differently.
07 Behavioural model
Turning repeat behaviour into recommendation confidence.
I worked with Peter in Data Science to translate the existing Basket Building Model into a customer-facing experience. The model examined both how frequently and how regularly an individual customer purchased each product.
Together, these signals produced a predictability score for each customer-product relationship. A product bought frequently and at a consistent interval could be treated as a high-confidence recommendation. Occasional or irregular purchases required more cautious placement.
Frequency
How often the customer purchased the product across their order history.
Regularity
How consistently the purchase appeared within the customer’s routine.
Predictability
The combined confidence that the product was relevant to the customer’s next shop.
- Purchase history
- Frequency and regularity
- Predictability score
- Recommendation priority
- Customer interface
The challenge was not only producing recommendations. It was deciding how much confidence was required before a recommendation deserved the customer’s attention.
08 Constraints and trade-offs
Prioritising what we could prove in the MVP.
QuickShop needed to validate predictive basket building without blocking roadmap delivery. Real constraints around engineering effort, category scale and customer trust shaped what shipped first.
Constraints
Fast validation required
We needed to prove the predictive shopping experience quickly without delaying broader roadmap delivery.
Engineering cost of the full vision
A fully guided multi-step experience demanded significant build effort before value was proven.
Relevance at category scale
Maintaining useful recommendations across large product ranges added complexity to logic and presentation.
Risk of recommendation overload
Too many layers could overwhelm customers and weaken confidence in what was being suggested.
Speed, familiarity and discovery
The journey had to balance routine efficiency with enough discovery without slowing repeat shops.
Evidence before scale
Behavioural and commercial metrics needed to justify investment before expanding the experience.
Trade-offs
Top Regulars as MVP
We launched a focused entry point first rather than the full multi-step basket-building experience.
Less exploration upfront
Category breadth was reduced initially to improve delivery speed and learning clarity.
Structured steps over Joy Scrolling
Testing showed continuous feeds caused cognitive fatigue and loss of orientation, so we prioritised step-based navigation.
Inspiration deferred
Recipe-led and inspirational experiences waited until core behavioural assumptions were validated.
Customer control over automation
Selections stayed reviewable rather than relying too heavily on automated basket creation.
09 Design principles
Reduce effort, not control.
Prioritise confidence over quantity
A smaller set of relevant recommendations was more valuable than a large set of uncertain suggestions.
Reduce decision fatigue
The experience should minimise unnecessary browsing, scanning and repeated searching.
Reflect natural shopping behaviour
Recommendations should follow routines, replenishment patterns and recognisable categories.
Keep the experience lightweight
Customers needed to progress quickly while retaining the ability to review and refine their basket.
10 Concepts
Three ways to make repeat shopping faster.
A major part of the project focused on testing different interaction models for personalised basket building.
We put three clickable prototypes in front of six experienced Waitrose online customers — comparing how each model handled speed, orientation, trust and basket completion across different shopping routines.
How we tested
Participants worked through realistic repeat-shop tasks. We observed completion time, confidence in recommendations, ability to review selections and willingness to continue — then compared patterns across the three models.
01
Joy Scrolling
An exploratory continuous feed designed to make browsing feel engaging.
Benefit: Continuous browsing created a fast, lightweight entry into recommendations.
Risk: It created scanning fatigue and weakened customers’ sense of progression.
Finding: Customers struggled to maintain context. Recommendations felt overwhelming, category switching created friction and lower-confidence products reduced trust.
01Rejected
Joy Scrolling
02
Step-by-Step
A structured flow that grouped recommendations into manageable stages.
Benefit: Clear stages reduced cognitive load and gave customers a stronger sense of progression.
Finding: It provided the clearest orientation, strongest progression and best fit with routine shopping behaviour.
02Preferred direction
Step-by-Step
Top RegularsFood & DrinkHouseholdInspiration
03
Netflix-style navigation
A collection-led model designed to encourage discovery across recommendation groups.
Benefit: Horizontal groups created clear category separation.
Risk: It felt visually familiar but required too much browsing for a task primarily driven by speed.
Finding: It improved category separation but created excessive lateral scanning, fragmented focus and weaker progression through basket building.
03Useful learning
Netflix-style navigation
Direction chosen
Why Step-by-Step became the direction
Step-by-Step Shopping became the preferred direction because it matched how customers already thought about a weekly shop — handling regulars first, then moving through familiar categories with a clear sense of progress.
Carousel-style grouping still informed how related products were presented within each stage, but the overall journey was structured rather than feed-driven.
11 Prototyping the system
Static screens could not test a dynamic product.
The proposed experience depended on customer-specific data, recommendation confidence, product categories, quantities, offers and grouping rules. Conventional Figma prototypes could show individual journeys, but they could not represent the behaviour of the wider system efficiently.
The limitation
The spaghetti trap
Manually connecting every possible state created prototypes that were difficult to edit, debug and maintain.
Rigid customer journeys
Static screens could not respond credibly to different customer histories and product recommendations.
Slow iteration
Building a realistic journey could take up to two weeks, limiting the number of ideas we could test.
Scaling failure
As personalisation increased, the number of screens and connections expanded faster than the prototype could support.
The response
I used Figma Make with a structured Supabase dataset to create a more realistic prototype. This allowed different customer datasets and product rules to populate the same interface rather than manually rebuilding every possible journey.
Structured product data
Products included recommendation order, quantity, category, grouping, product type and offer information.
Reusable interface logic
The same components could respond to different datasets and behavioural conditions.
Faster learning
We could test the behaviour of the recommendation system, not only the appearance of individual screens.
12 Interaction logic
How behavioural signals shaped the interface.
The model produced scores and product attributes, but customers needed a clear, manageable shopping experience. I translated those outputs into a set of interface rules governing what appeared, where it appeared and how products were grouped.
- Behavioural data
- Confidence threshold
- Product grouping
- Journey position
- Quick basket action
01
High-confidence products became Top Regulars
Products above the agreed confidence threshold could be surfaced prominently as likely repeat purchases.
The prototype initially used a confidence threshold of approximately 60% to explore how many products could be shown before relevance and trust began to fall.
02
Related products were grouped together
Alternative or related products could be assigned a shared grouping identifier and displayed together, reducing repetitive scrolling and supporting quicker comparison.
03
Categories shaped progression
Product categories helped determine where recommendations appeared within the Step-by-Step journey, aligning the flow with recognisable shopping behaviour rather than a continuous recommendation feed.
04
Offers supported value without disrupting the journey
Offer data could influence product presentation and quantity recommendations while keeping customers within the main flow.
Product groups could use different presentation patterns such as standard, single and double depending on the recommendation context.
13 Validation and iteration
Testing changed the density, confidence and scope.
Testing focused on more than whether customers could complete the flow. We needed to understand how recommendation confidence, information density and progression affected trust.
Before testing
- Broader recommendation sets
- More categories shown upfront
- Lower-confidence products appeared too prominently
- Continuous browsing increased scanning
- The full multi-step concept carried significant delivery complexity
After testing
- Higher-confidence products were prioritised
- Fewer recommendations appeared at each stage
- Related products were grouped more clearly
- Step-based progression improved orientation
- Top Regulars became the focused initial release
Structured flows increased confidence
Customers responded positively to Step-by-Step because it reduced overwhelm, improved clarity and created stronger progression.
Recommendation transparency mattered
Trust improved when recommendations felt relevant and behaviourally logical.
Simplicity outperformed novelty
Customers prioritised speed, predictability and efficiency over visually exploratory browsing.
Poor matches reduced trust quickly
Lower-confidence recommendations made the experience feel less reliable when they appeared too prominently.
The winning direction was not the most visually novel. It was the one that best matched routine shopping behaviour.
14 Solution
A faster way to rebuild regular shops.
The broader Step-by-Step model established the long-term direction, but the first release focused on Top Regulars: the highest-confidence repeat purchases within each customer’s history.
Recommendations were prioritised using behavioural confidence and presented in a lightweight, reviewable flow. Customers could quickly select familiar products, adjust quantities and continue building their basket without handing complete control to automation.
Top Regulars
High-confidence repeat purchases formed the foundation of the initial basket-building experience.
Structured recommendation groups
The broader design direction organised products around shopping behaviour and recognisable categories.
Lightweight interaction design
The experience prioritised rapid selection, clear progression and easy basket refinement.
- Purchase behaviour
- Predictability score
- Confidence threshold
- Product grouping
- Quick basket action
15 Results
The impact of reducing repeat shopping friction.
QuickShop delivered measurable gains in speed, basket value, engagement and repeat behaviour — showing the value of behaviour-led personalisation when it is designed around customer confidence and control.
Easier to find regular productsFaster to complete the shopMore products added with confidenceHigher commercial value
25%
faster completion
26 mins → 21 mins
5 minutes saved per shop
Completion time dropped from 26 minutes to 21 minutes, saving customers around 5 minutes per shop.
45% reduction from Favourites
Navigational add-to-basket actions from Favourites dropped by 45%.
10% reduction from Search
Add-to-basket actions from Search dropped by 10%, showing customers relied less on manual searching.
Browse and discovery remained stable, showing QuickShop reduced repeat-shopping friction without stopping spontaneous shopping behaviour.
16 Reflection
Personalisation was only valuable when it made the shop feel simpler.
QuickShop showed that predictive technology alone does not create a useful experience. Customers judged the system through the quality, order and density of what appeared in front of them.
The most important design work was translating invisible behavioural signals into clear interface rules, then reducing the first release to the smallest proposition that could prove customer and commercial value.
Design the confidence, not only the recommendation
Where and how a suggestion appeared mattered as much as the product itself.
Prototype the system behaviour
Realistic data and rules exposed issues that static happy-path screens could not.
Prove the habit before expanding the vision
Top Regulars allowed the team to validate repeat-shopping value before investing in the broader Step-by-Step journey.