Auto Completion for Ask
Timeline:
Nov 2021 - Dec 2021
sector:
Business Intelligence, Search, Natural Language, Artificial Intelligence.
Team:
Design team: Avi Agarwaal, Sanjana Mohan
Product Management: Ganesh S, Vivek Natarajan
Engineers: Prasad Bari, Nehal Gala
My role:
End-to-end project delivery, Product thinking and Solutioning, User Experience, Visual Design, Web, Android and iOS designs, and Working closely with the tech team.
Auto Completion for Ask
Timeline:
Nov 2021 - Dec 2021
sector:
Business Intelligence, Search, Natural Language, Artificial Intelligence.
Team:
Design team: Avi Agarwaal, Sanjana Mohan
Product Management: Ganesh S, Vivek Natarajan
Engineers: Prasad Bari, Nehal Gala
My role:
End-to-end project delivery, Product thinking and Solutioning, User Experience, Visual Design, Web, Android and iOS designs, and Working closely with the tech team.
Introduction
The feature of Search at the Crux is called Ask. The goal was to make Ask the most powerful tool for the users and auto complete was the first step towards this goal. Auto-complete transforms a recall problem into that of recognition. When the user is asks a question, the system should suggest all possible terms that aid the user to complete that question.
Problem
What is working for us?
Ask as a feature for our product is excellent! It's very unique and makes the life of a user super simple.
Ask is a very go-to feature for people (Data from clients Shalini Mookarjee had interviewed for nudges for the future)
Users are aware of how to use it.
The recently asked is super helpful. (frequent and recent)
'See what you can ask' is very interesting.
NLP has made it easier for people to type in simple words and get the answers.
Showing answers for - is a good thing to have and the drop-down to select other answers around similar data.
What is not working for us?
A new user might get confused trying framing a question and how to frame a question as there is no support to help them guide through.
We currently have recent searches but it might be frustrating if the question they are looking for is not there.
The 'See what you can ask' page is crowded and overwhelming.
Because we allow users to type in any open-ended question, it might be very confusing for the user when the system changes the answer title and they miss out on seeing 'showing answers for.
Users are not aware of what is there in the system - entities in the system.
Process:
Defining the Principles
I learned that it’s great to base the foundation of any project on a set of principles that can be used as a guide and can be referred to time and again.
✏️ Efficiency
Making ask more efficient
🎯 Accuracy
Accuracy in understanding the input, providing the output, understanding and conveying errors
💁♀️ Natural
To make the system be more adaptable to natural language even while mapping the asked question to the answer
🧠 Smart
The system provides suggestions that are the most relevant and eventually often asked by the user
Creating a Basic Framework
The first plan of action was to divide the types of entities that a user can ask. This not only gave a broad understanding of the way the backend processes the data but we were able to cumulate all the data into bite size information that is also easily comprehensible by the user. This also allowed us to make sure that the user can easily input the question while they are aware of what entity they are using.
Mobile Explorations
This project had a mobile first approach, we first solved and explored for mobile screen and then the same was translated into web.
MVP Design
mobile Design
Web Design
The same designs were translated into web.
www.cruxintelligence.com
Solving for Parent Suggestions
If the suggestion has two different parents that it is associated to this must be displayed in order to direct the user to the right result.
Solving for Synonyms
Sometimes the user tends to type in the synonym of a search word instead of the main entity itself, like revenue instead of dollar sales. In this case earlier the backend used to change the search result to “dollar sales” even if the user had input “revenue”. This might confuse the user. This is where synonyms come into place. While asking the question itself, the user will now input the actual entity instead of the synonym with the help of suggestions.
Replacing
In the scenario where the user edits a search and replaces on entity, for example in dollar sales the user wants to replace dollar with unit, in this case the whole unit “dollar sales” will be replaced with “unit sales”.
Tag Expansion and Collapse
Its understandable that since the entity tags on the right hand side of the suggestion is abbreviated, it might take a while for the user to get used to the abbreviation or they might not be a frequent user, or might be a new user. For better user experience, this tag can be expanded and collapsed.
Business Object Level Analysis Suggestions
There are three main business objects, Performance for region, Performance for all brands and Performance for all regions.
In case of Performance of region, the user must specify which region after selecting this option in order to get an answer. Hence the system will first inform the user incase they proceed to do so. But if they choose to search without specifying the region, the system will not be able to interpret and give an error.
Future Implementations
To make this more user-friendly, designs were also made for the smart suggestion which was taken up and implemented after the MVP was launched.
Metrics
Response time <1s for suggestions
Google-like experience in suggesting and selections
Learnings
This was the first project I had worked on where I was parallely working with the product management team 60% of the time because they had to make sure that the functionality of these designs were aligned with how the backend intended to develop it. It was one of the smoothest projects because of great management from the product manager Vivek Natarajan. It showed me how important the plan before work is and how a good planning makes it effortless for the design team
The more iterations the better. Giving myself a set number of days in a sprint to just iterate depending on the timeline of the project allowed me to really weed out all possibilities and come up with the best design for MVP.
Introduction
The feature of Search at the Crux is called Ask. The goal was to make Ask the most powerful tool for the users and auto complete was the first step towards this goal. Auto-complete transforms a recall problem into that of recognition. When the user is asks a question, the system should suggest all possible terms that aid the user to complete that question.
Problem
What is working for us?
Ask as a feature for our product is excellent! It's very unique and makes the life of a user super simple.
Ask is a very go-to feature for people (Data from clients Shalini Mookarjee had interviewed for nudges for the future)
Users are aware of how to use it.
The recently asked is super helpful. (frequent and recent)
'See what you can ask' is very interesting.
NLP has made it easier for people to type in simple words and get the answers.
Showing answers for - is a good thing to have and the drop-down to select other answers around similar data.
What is not working for us?
A new user might get confused trying framing a question and how to frame a question as there is no support to help them guide through.
We currently have recent searches but it might be frustrating if the question they are looking for is not there.
The 'See what you can ask' page is crowded and overwhelming.
Because we allow users to type in any open-ended question, it might be very confusing for the user when the system changes the answer title and they miss out on seeing 'showing answers for.
Users are not aware of what is there in the system - entities in the system.
Process:
Defining the Principles
I learned that it’s great to base the foundation of any project on a set of principles that can be used as a guide and can be referred to time and again.
✏️ Efficiency
Making ask more efficient
🎯 Accuracy
Accuracy in understanding the input, providing the output, understanding and conveying errors
💁♀️ Natural
To make the system be more adaptable to natural language even while mapping the asked question to the answer
🧠 Smart
The system provides suggestions that are the most relevant and eventually often asked by the user
Creating a Basic Framework
The first plan of action was to divide the types of entities that a user can ask. This not only gave a broad understanding of the way the backend processes the data but we were able to cumulate all the data into bite size information that is also easily comprehensible by the user. This also allowed us to make sure that the user can easily input the question while they are aware of what entity they are using.
Mobile Explorations
This project had a mobile first approach, we first solved and explored for mobile screen and then the same was translated into web.
MVP Design
mobile Design
Web Design
The same designs were translated into web.
www.cruxintelligence.com
Solving for Parent Suggestions
If the suggestion has two different parents that it is associated to this must be displayed in order to direct the user to the right result.
Solving for Synonyms
Sometimes the user tends to type in the synonym of a search word instead of the main entity itself, like revenue instead of dollar sales. In this case earlier the backend used to change the search result to “dollar sales” even if the user had input “revenue”. This might confuse the user. This is where synonyms come into place. While asking the question itself, the user will now input the actual entity instead of the synonym with the help of suggestions.
Replacing
In the scenario where the user edits a search and replaces on entity, for example in dollar sales the user wants to replace dollar with unit, in this case the whole unit “dollar sales” will be replaced with “unit sales”.
Tag Expansion and Collapse
Its understandable that since the entity tags on the right hand side of the suggestion is abbreviated, it might take a while for the user to get used to the abbreviation or they might not be a frequent user, or might be a new user. For better user experience, this tag can be expanded and collapsed.
Business Object Level Analysis Suggestions
There are three main business objects, Performance for region, Performance for all brands and Performance for all regions.
In case of Performance of region, the user must specify which region after selecting this option in order to get an answer. Hence the system will first inform the user incase they proceed to do so. But if they choose to search without specifying the region, the system will not be able to interpret and give an error.
Future Implementations
To make this more user-friendly, designs were also made for the smart suggestion which was taken up and implemented after the MVP was launched.
Metrics
Response time <1s for suggestions
Google-like experience in suggesting and selections
Learnings
This was the first project I had worked on where I was parallely working with the product management team 60% of the time because they had to make sure that the functionality of these designs were aligned with how the backend intended to develop it. It was one of the smoothest projects because of great management from the product manager Vivek Natarajan. It showed me how important the plan before work is and how a good planning makes it effortless for the design team
The more iterations the better. Giving myself a set number of days in a sprint to just iterate depending on the timeline of the project allowed me to really weed out all possibilities and come up with the best design for MVP.
