2023
Intuit: Machine learning recommendation experience
I developed account task recommendations for an AI/machine learning model. This model was designed to extract insights from customers' historical account activity, learn from these patterns, and subsequently identify the most suitable task for each customer, precisely timed for optimal effectiveness.
What is the Intuit Account Manager?
The Intuit Account Manager platform receives 21 million annual visits, empowering customers to manage their data across various Intuit products they use. Additionally, it serves as a collaborative platform where teams integrate their ecosystem experiences, providing customers with functionalities such as managing payment methods, sign-in and security credentials, address details, and more. This central hub presents actionable tasks to prompt customers to address crucial account-related matters.
— Business Problem
Increase customer engagement through an AI-driven solution
Managing "hard-coded" rule-based recommendations for surfacing account tasks poses significant challenges for developers, necessitating extensive rework whenever new actions need to be incorporated.
By integrating AI/ML technology into the Intuit Platform team for the first time, how can we circumvent explicit "hard-coded" programming and instead harness machine learning algorithms to analyze vast datasets, derive insights, and subsequently make informed decisions?
— Customer Problem
I just want to ignore these tasks…especially if they keep coming up and I’ve done them already.
“As an Intuit customer, I receive important task recommendations to complete, but I don’t always take action. I see them repeatedly until I finally dismiss them, which makes me feel forced to click on something I have no interest in or may have completed already, so I end up seeing this as annoying and aggravating than something of value to me.”
— Proposed Solution
How might we categorize the recommendations so the model can learn the right task to show at the right time?
In contrast to the existing "hard-coded" rules for presenting account tasks to customers, the new AI/ML model would rely on algorithms and customer data to introduce innovative solutions to this challenge.
It stretched our team to think differently about using learning algorithms to replace rigid business logic, using AI/ML to learn from past data, and increasing the accuracy of the output.
— The Design Process
Review data to learn about tasks customers are trying to complete and narrow a set of task recommendations that could best fit/work within the AI/ML model into themes.
Took a stab at writing content to communicate task benefits more effectively so customers are more inclined to take action and not ignore them.
Assess the AI/ML model's strengths/weaknesses, take our learnings, and propose a recommendation framework that would govern how, when, and why specific recommendations surface to customers.
— Results
What we learned.
Things in the future.
KPIs: We achieved our 5% KPI lift to increase account task recommendation engagement because of the new AI/ML at work.
My first time with AI: Machine Learning algorithms are only as good as the data they are trained on. The model can create new flavors of old data but will always be limited. How might we think about presenting generative content to customers?
Future: Push the boundaries and think about personalizing experiences.
