Bing Offers optimization

Design for search engine  •  Information design  •  Data-driven design  •  Research & testing  •  Motion Design
Introduction

Bing Shopping launched the Offers segment in early 2018 to help users get the best coupons and deals from their favorite brands.

The goal was to help users save their time by giving them instant results on SERP (search engine results page) without having them to browse different websites for appropriate coupons. I took upon the responsibilities of the Offers segment in 2019 and this was the first project I worked on.

Role & duration

I worked as the sole-designer on this project with 1 PM, 1 front-end developer & 2 backend developers in 2019. This case study showcases the work of around 2 months.

TASK OVERVIEW

My task was to identify the short-comings with the then existing answer (features on Bing search results page are called answers) and redesign it to get better engagement from the users.

Solution preview
PROCESS: WHERE DO WE START?
It all started with an extensive experience audit and a few stakeholder discussions

I analyzed the existing experience at that time and did a thorough analysis of that answer based on factors such as scan pattern, information heirarchy, affordance, etc.
I also talked to the PM & design leadership to understand their hypotheses and why they had certain opinions.
Then I turned to the usage data and looked at how our users were interacting with the answer. Talked to a few users as well. Found some interesting insights there.

Business Goal
The team had an APSAT goal of 50% or more

APSAT is an internal Bing metric, which in simpler terms, was used to calculate the user satisfaction for a particular Bing experience.

Goal guidance: stakeholders

"Communicate the value of this answer better to the users."

User Study findings

Users want to see the details of the offer before applying/using it.

Answer usage metrics

68% of the total clicks came from the carousel arrow buttons.

Design Analysis

Readability, white-space usage & information design could be improved while looking at SERP whole page.

Some interesting actionable insights
LAYOUT optimization
Scan pattern analysis 101

Bing search-engine result page (SERP) and other similar text-heavy web-pages (unbounded containers) have a typical F-shape scan pattern. On the other hands, UI cards (bounded containers) typically have a Z-shape scanning pattern. That’s the general paradigm but there could be exceptions.

My assumption

Information consumption is quicker and easier during linear scanning versus in zig-zag scanning. Hence, I wanted to explore an answer layout that sits coherently within SERP’s layout.

Rough representation of serp scan pattern  →
Success Criteria
What does success look like for this answer?
Positive user feedback

Different users seek out different things when the target audience is broad enough. Getting to know that users are achieving their own goals (could be comparing offers, copying faster, getting right info, etc.) successfully with this answer is the primary success criteria.

Increased APSAT (50%+)

APSAT is the measure of user satisfaction and is largely defined by the % of users who interact with the top answer after coming to a search results page. People interacting with the links below the answer without interacting with it contribute to negative APSAT. A 50%+ APSAT is generally considered fit for shipping.

Higher success rate

Obviously, the coupon code success rate is paramount but there are other micro tasks like finding the relevant coupon, consuming required info, etc. which aid to an overall high success rate.

Reduced task completion time

When dealing with a lot of data, a reduced task completion becomes an important success criteria because it shows the overall product usability and comprehensiveness.

DESIGN PRINCIPLES
How should the design be experienced by the user?
Time saving/efficient

Different users seek out different things when the target audience is broad enough. Getting to know that users are achieving their own goals (could be comparing offers, copying faster, getting right info, etc.) successfully with this answer is the primary success criteria.

Relevant

APSAT is the measure of user satisfaction and is largely defined by the % of users who interact with the top answer after coming to a search results page. People interacting with the links below the answer without interacting with it contribute to negative APSAT. A 50%+ APSAT is generally considered fit for shipping.

DESIGN direction
How should the proposed design be structured?
Incorporate more offers in
a single view.
INSIGHTS USED:

From stakeholders: the value proposition should be communicated.
From the business ask: Bring in more engagement.Success criteria: Get higher APSAT and success rate.

Assumption:

Showing more offers in one view would lead to more engagements and would communicate the value proposition of the answer.

Include offer information in
the answer itself.
INSIGHTS USED:

User interviews: Users were going to the provider website to read the details and came back after knowing the offers was not applicable for them. Success criteria: Reduce task completion time.

Assumption:

Including coupon details in the answer itself would save time by providing all the necessary info at one place.

Use the most efficient and
natural layout.
INSIGHTS USED:

Design analysis: A quick design analysis of the existing answer showed a good scope in the improvement of information design of the answer. Success criteria: Reduce task completion time.

Assumption:

Using a better and more natural answer layout than the current answer would result in easy information consumption and hence would save time.

DIVERGING FOR IDEATION
Increasing the number of offers in the view-port

I started the ideation process with exploring options to increase the no. of offers because the other 2 design directions needed some form of fidelity in the designs.

Based on the offer data available and the Bing search results page framework, I created some wireframe layouts to show more than 3 (in the old design) offers. Wireframe for the old design is shown on the left.

DESIGN DECISIONS
Shortlisted card designs

I wire-framed/ sketched various card designs for each layout and shortlisted one card. Factors considered while deciding the card:
• Scan pattern of the individual card
• Information hierarchy & grouping

Also, I tested each card for the edge cases and stress-tested all the individual cards before shortlisting the final one.

CONVERGING
Narrowing down options

After some discussions & reviews of all the explored visual design options, I narrowed down to 2 final options (Layout #1 & Layout #4), eventually leading to the selction of Layout #4 as the final direction.

Evolution of the final design
Visual iteration journey of the shortlisted design

Once the layout direction was finalized, that’s when the iteration on the design started. At this stage, I was giving a lot of attention to detail to bullet-proof the answer and optimizing the hell out of it. 🌚

Each design during this stage did not go through an extensive feedback cycle and the decisions were based mostly on informal team-feedback and my own design intuition.

92%

users (46/50) preferred the new experience over the old one in basic A/B testing
(focused on ease of use, usability, info. consumption, etc).

FINAL DESIGN
Final iteration used for development and flighting
Positive user comments:
  • Easy to scan and compare multiple offers.
  • Earlier design didn’t show me the breadth of what this answer offers.
  • Less user interaction and hence faster task completion.
  • Provider logos instill trust and confidence and make it kind of ‘official’.
  • It takes effort to look for relevant info in earlier design.
  • Makes It easy to find what I’m looking for.
Notable wins
PUTTING IT ALL TOGETHER
Mobile optimization & looking at the whole page
Further iterations
Quick iterations after launch based on usage data
from this ↑ 
to this ↑ (de-duplication & see more prominence)   
Bringing in motion
Exploring interactivity through micro-interactions

Micro-interactions & animations on SERP is tricky. The PLT metric (Page Load time - one of the primary Bing performance metrics) greatly narrows down the options one could explore. I have always looked at motion on SERP more as a problem solving tool rather than a delight creator. (Although solving problems could bring in the desired delight).

MOTION STUDY FOR THE INFo-PANE INTERACTION
SOLVING THE COPy-CLICK FIASCO
⛔️ Lack of association between the action and the confirmation message – hence leading to multiple redundant copy clicks.

This interstitial message state (Copied to clipboard) serves 3 important purposes:

✅  Preventing unwanted continuous copy clicks while the code is being copied (due to the overlay).

✅  Showing the action confirmation in the same proximity providing good association between action & confirmation.

✅  Less implementation time.

LET’S talk business
Final business impact

The new answer exceeded the team’s expectations by a huge margin. We were aiming for an APSAT of around 50% (previous answer’s APSAT was around 46% - 47%).

68%

approximate APSAT was observed consistently.

24%

drop in Offers page abandonment.

Reflecting on the journey
What were some of the key challenges I faced as a designer?
Trimming the answer
Problem

I had to let go many elements from the answer which was frowned upon by the product team. It’s easy to lose sight of what’s the core value of this answer when dealing with so much data. It was hard to convince the stakeholders to eliminate some of  the intelligence built by the team.

Solution

I immediately turned to users and focused on what’s the key task. User feedback helped me convince the product team to further explore this direction.

Decision making with data
Problem

The hard part of information design is to make it work for every case. Edge cases must be clearly defined. Some data points could really be absurd. The blue link length in an offer was one such data point. It was impractical to design for huge string length.

Solution

In this case, I turned to Bing usage data and got the numbers extracted regarding string length. A quick analysis revealed that using around 46 with characters would cover around 78% of all offers. I went ahead  that and decided to truncate strings greater than that length. Had I delved into designing accurately for all lengths, we wouldn’t have gotten such great results.

Reflecting on the journey
What could I have done better?
Increase user involvement in the design process

Looking back, I realize I could've involved users more in the process (like for example: I could’ve tested all the different card explorations with the users for detailed feedback). I relied on both expert reviews and user reviews, but I wanted to have more user feedback.

More interaction with the previous designer

I did not conceptualize this answer. I improved upon it. In hindsight, I could’ve interacted more deeply with the initial designer of this answer that probably would have resulted in great insights eventually end up saving the team’s time.

Delve into eye-tracking

Scan-pattern was one of the key aspect of this overall design project. Verbal feedback isn’t very useful in such cases. I could’ve utilized eye tracking teams’ help to have an eye tracking study for my answer and explorations.

Validate all my assumptions

I couldn’t concretely validate the reduce in task completion time assumption in this project. Although I could strongly say that it happens with the new design, I could've done a study to calculate the same in quantitative terms.

Learn from other Bing answers

All the answers on Bing SERP have gone through a stringent design/development process. I could’ve gathered more insights to inform my designs and learn from what works or doesn’t work for those answers.