AI-Driven Search For Turbotax
$40 Million saved by designing search improvements in the Intuit-wide help platform that leveraged design solutions using AI combined with user research and a comprehensive design process.
Problem
Intuit customers use TurboTax and QuickBooks without inherent expertise and often become challenged by questions or problems that they don’t know how to resolve. To find these solutions, they turn to searching or calling customer support for answers that use up their time while costing time and revenue for Intuit.
Millions of people every year get answers to their tax questions using the Intuit help system.
Help system accessed from within TurboTax
Help system accessed from the TurboTax support site
“Challenges in forming a good query were seen across all products and experience levels. This is the most important problem to address.”
Query formation improvements can have a major impact on customer outcomes.
Solution
By understanding the entire journey of the customer as well as the customer support team, I was able to design personalized AI-driven solutions that helped people find answers faster and easier. At the same time, these changes enabled product teams to more easily surface the right content while also freeing up engineering and support agent resources.
I led the design on the following solutions in this multi-year journey:
Autosuggest in search. Predictive search powered by machine-learning (ML) and personalization makes entering queries faster and more likely to lead to the correct answer.
Updated search results page elements including summary extraction, content ranking, related searches, and metadata.
Added customer feedback to articles enabling a feedback loop for content designers and our Machine-Learning (ML) training model to improve search relevance.
Getting feedback on why a customers found an answer helpful or not was one of many inputs that informed our ML model
Our ML models provided improved results for auto-suggest, as seen in TurboTax.
AI/ML models were trained using a variety of inputs including user interactions, voice of the customer, context and success outcomes.
Our models powered auto-suggest, search results candidate ranking, answer summaries, and related searches.
Improvement in the search results page led to 5% reduction in customer care contacts with summary extraction, 4% reduction with related searches, and collapsed filters made results easier to read.
Outcome:
The design work that I led in advancing search experiences within the self-help space made a big impact:
AI-driven search solutions combined to save 40+ Million minutes a year on the phone with a support agent. That also saved the company over $40 million.
In 2020, the Intuit help team won the Scott Cook Innovation Award based on the design, innovations, and technology that I created.
5% reduction in customer care calls by redesigning the search results page components. I also successfully lobbied to remove features and page elements that detracted from our goal.
Filed over 30 technology patents based on designs for current and future solutions.
“Intuit has smoothed over the rough edges we found in TurboTax’s own help system ... it quite simply offers a clearer, smoother path.”
More Case Studies
Financial Data Sharing for Couples
Intuit Help System Redesign