Shopping recommendations
2020 (Accenture)
The client
One of the UK's largest retailers with both a physical and online prescence in the marketplace.
The goal
Accenture undertook an independent piece of work, as part of a procurement process, investigating possibilities to improve the retailer's online product recommendations offerings. I led a design team and we worked with a technical team to create a proof of concept on Google's search recommendation technology.
The challenges
- Product recommendations are seen by users as 'adverts' and actively ignored
- Users do not understand why specific product recommendations are shown to them
- No access to client or users due to procurement process
The work
Guerrilla research
Despite having no formal access to users through the client, we were able to go out and ask friends, family and colleagues, in a 'guerrilla' fashion, to enquire about their shopping habits and attitudes to product recommendations.
By its nature this type of research will not reach a full spectrum of target users, but even the completion of limited interviews manages to provide a range of views and opinions to begin building personas and make design decisions.
As part of the pitch to the client, we made sure to propose that research across a wider spectrum of users would be required, to validate, update and better understand the users' needs.
Personas
From the guerrilla research we identified several key user dimensions, including:
- Low <-> high budget
- Browsing <-> shopping list only
- Foodie <-> disinterested in food
- Restricted <-> unrestricted (diet, health, allergies etc)
- Habitual <-> experimental shopping choices
- Low <-> high environmental impact
From which we produced three personas:
- Family shopper on a budget
- Health conscious shopper
- Experimental foodie
Re-design
While mapping out any existing journey is always a good step, here it also helped, in lieu of client input, to uncover additional features beyond the basic; browse/search for product, review, add to basket and checkout. Some of these included:
- Logged in vs. a guest user
- Delivery vs. click and collect
- Different types of offers, promotions and product recommendations
- Product categorisation
Product recommendations
Product recommendations are the core underpinning of what is shown to the user on any specific page, based on information the retailer knows about the user. The recommended algorithm would use data science to sort the products on display to bring those most likely to be bought to the top.
On top of the core product recommendations the client could then overlay different categories of special items, like offers and promotions, which they want to be brought to the users' attention. In this way, the client could offer a more 'Google Ad' style offering to manufacturers and show products to specific market segments.
Visually, we changed promotions and offers to look the same as a regular product, but highlighting why this product is displayed to the user (see green border around strawberries in Figure 2).
Previous order
Another suggestion was to allow users to quickly and easily review their previous order and add those items to their basket. While it was already possible for a user to see their previous order it was hidden away. We proposed making this part of the core journey, along with being able to easily amend product quantities before adding to basket, or even skipping this step entirely.
Regular, infrequently bought
We also created the idea of 'regular, infrequently bought' items. These would be suggested before checkout, based on the frequency of purchase. For example, if you buy washing powder every five weeks, and it's been five or more weeks since you last bought it, we could remind you, as it would not be in your previous order.
Budgeting help
By giving the users an option to set a budget for their regular shopping, the system can alert them when they approach this amount, and recommend alternative products at a lower price. The identified difficulty of this approach was how to ensure the alternative product is comparable. For example, tomato ketchup and cherry tomatoes are not the same, despite both having tomato in them.
Additional ideas
Because of the short-term, proposal nature of this piece of work, there were ideas we were not able to develop and/or ratify their feasibility. For example:
- Product recommendations to include recipe ideas based on data like the user's basket, seasonal ingredients and offers
- Meal prep functionality to support the core retail offering
The outcome
The completed proposal was presented in mid-2020 and was well received by the client, both for the work done around product recommendations, but also for suggesting ideas on the wider user experience.