Inspire Cockpit | A recommendations tool
A self service tool to allow Farfetch Partners and Internal Users to create recommendation strategies using machine learning algorithms.
My role
Senior Product Designer
My responsibilities
Product Discovery, Information Architecture; User Research; Visual Design
Client
Farfetch
Year
2022-2023
What's a recommendation?
Suggestions recommended by algorithms based on different parameters to deliver appealing products with the purpose to amaze customers and increase conversion.
Context and business opportunity
“I want to show on the Product Detail Page similar products to the one user is seeing”
Partners and Internal users (PMs) with business needs...
...but need to ask for a PM to create a use case...
...but only Engineers were able to create strategies using algorithms.
PAIN POINTS
🔴 External partners were migrating to third-party solutions
🔴 Users had to wait a long time to get their cases done.
🔴 Internal team spent too much time on support work
Red flags
🚩 It was not clear how the success of the tool would be measured.
At first it should be related with sales conversion.
🚩 Due to a tight deadline I focused on the desk research to learn about the user’s pain points and get feedback later on with usability tests.
🚩 I faced some resistance from the GPM with design process, because the tool would be too technical to identify "user needs".
Discovery
USER ROLES
Engineers and Data Scientists | Access to create, manage and test algorithms under development
PMs | Create and manage strategies for their business needs
CRM | View only, with the purpose to know the company’s algorithms and portfolio
MAIN JOURNEYS: CREATE AND MANAGE STRATEGIES
Business case triggers the need for a new strategy
User defines basic information (name, channel, touchpoint, results quantity
User choose algorithms + details to enhance the strategy’s results
Manage settings
User tasks and behavior
At this point it wasn’t very clear how different these 2 users behavior would be. I agreed to move forward based on assumptions, to meet the deadline. However, I proposed to run an usability test at the end.
Scope definition
I moderated a workshop session with the PM to help us prioritise the scope of the MPV.
Used the Confluence page with the requirements and we brainstormed over the pre-defined solutions. Also, created the product pillars aligned to the main Jobs to Be Done:
-
Access (onboard user)
-
Explore (find strategies)
-
Create (a new strategy)
-
Manage (an existing strategy)
Workshop outputs
-
Brainstormed a few "how might we's"
-
Prioritised 3 Jobs to Be Done to start:
-
Find strategies
-
Create strategies
-
define basic information
-
define logic
-
-
Manage strategies
-
activate/deactivate
-
change settings
-
MVP + Research insights
Jobs to be done: find strategy and create strategy
Jobs to be done: create strategy (define basic information)
🚩 At this point of the project, there was the understanding that selections on the Modules section would impact the Parameters information. The hypothesis was that users would switch attention from left to right sides while filling the form.
User testing learnings
⛔ PMs insecure to fill the form, missed instructions or explanations.
✅ Data Scientists recognized the fields and didn’t missed anything.
Autonomy: users wants to create strategies and be autonomous about that
Guidance: users wants their doubts answered inside the tool
Performance: Track results is mandatory, there’s no use without dashboards
Jobs to be done: create strategy (define logic)
Continuous discovery and MVP iterations
Beta testers Group
While I was working on the design improvements I created a slack channel to keep in touch with the users who tested the POC. The PM also used to group to provide support to their questions.
Experience Iterations
Simplify all the mandatory rules to create a strategy wouldn’t be possible via back-end to the project’s deadline.
All changes must be made in the front-end, so experience design would be the key to improve the tool.
Less steps + sections with instructions
"Light version": less fields for entry level editor users
The second round of research was key to provide to stakeholders enough evidences that a review of the tool’s architecture would be required.
I proposed an Entry Level role for users without experience on recommendations but needed to create their own strategies
New additions to the Design System and more iterations
Onboarding tour on first access.
Side panel with additional content, giving users more guidance throughout experience.
Learnings
5 key product teams from Marketplace are currently creating strategies every month as part of their conversion goals.
🚩 How to measure Success?
- Customer Satisfaction Score
- Engagement with key features (clicks on preview algorithm, page views...)
- Flow dropout
Clear design process
To be able to execute a good design process I had to speech my decisions all the time to the cross functional team.
I had to be open for feedbacks and quickly adapt to constraints.
User feedback is a must
The recommendations subject was all new to me, along with the business and user needs.
I had to find ways to include tests as much as I could to fundament my design decisions.