2023
Intuit Account Manager Task Recommendation Redesign Using AI/ML
Led design strategy for Intuit Account Manager's first AI-powered task recommendation experience, replacing static, rule-based prompts with personalized guidance. I defined the recommendation framework, aligned engineering, data science, and product leadership around it. The work now shapes Intuit's recommendation strategy across 21M+ annual account visits within the Intuit Account Manager.
Within 60 days of launch, task completion rose 5% relative to baseline, with fewer repeated prompts and more completed high-value identity tasks.
The Intuit Identity Platform
The Intuit Account Manager platform empowers customers to manage their data across various Intuit products they use. It serves as a collaborative platform where identity management capabilities are contributed to provide customers with functionalities such as managing payment methods, sign-in and security, profile management, and more.
Customer Insight Informing Our Work
I was able to search and surface that over 140,000 customers say that they see the same account prompts repeatedly, even if they did not apply to them. This created frustration and reduced trust in account-level messaging. At the same time, the business needed a scalable, intelligent way to surface the right actions to customers across millions of account sessions without requiring manual targeting rules.
Team Opportunity
How might we use AI to help customers complete the right tasks at the right time?
Intuit prioritizes an AI-driven future, encouraging designers to identify opportunities where intelligence can enhance customer experiences. This shift enables the Platform team to move from rigid, rule-based logic to adaptive, data-driven models that learn from real behavior allowing recommendations to evolve rather than remain fixed.
Customer Problem
“I want account recommendations that actually apply to me, not tasks I ignore because I’ve taken action on them already.”
Customers often saw the same account prompts repeatedly. This created frustration and reduced trust in account-level messaging.
At the same time, our team needed a scalable, intelligent way to surface the right actions to customers across millions of account sessions without requiring manual targeting rules.
How I Collaborated With Stakeholders
Design Process
Data Engineering Capabilities
Worked closely with data science to understand how it could work for our use case and the weaknesses that exist within the model.
Recommendation Framework Definition
Built a framework for when and how recommendations appear, grounded in customer feedback themes and adapted to each user's account signals.
Pilot Launch + Measurement
Launched with a focused pilot audience to measure real-world interaction patterns and optimize before broader rollout.
From > To
What AI/ML Brought To The Intuit Account Manager
| Before | After (AI Driven) |
| Generic prompts shown to all users | Personalized recommendations based on behavior and account state |
| No explanation of why the prompt appeared | Clear contextual messaging explaining relevance |
| Repetitive experiences that led to frustration | Dynamic, adaptive recommendations that evolve over time |
Design Decision: validate with one
high-value task before scaling
We could have launched recommendations across every account task at once. Instead, I pushed to start with Confirm your phone number because it's tied directly to sign-in, recovery, and security, so improvements would be easy to measure and meaningful.
The tradeoff was designing other account recommendations 30 days later around the framework already proposed. But it gave the Identity team clean data to prove the model's value, and that result secured buy-in for expanding the framework to design other account recommendations 30 days later.
Content: proposing to the team how we say things
Starting with phone number confirmation was the right call, but it raised a new question: which motivation actually moves customers to act? I'd test three versions of the recommendation, each built on a different reason to act, with its own headline, body copy, and buttons. I drafted the content myself as a starting point, to be refined with content design before testing.
Final Recommendation
I chose the quick check messaging version. Our pilot showed that repetition drives dismissal and that identity tasks carry emotional weight, so the lowest-effort version seemed most likely to get action without adding anxiety. The protection version risked feeling alarming, and the convenience version promised a benefit customers might not value until they need it.
Additional Deliverables Outside of Scope
Designed New Account
Recommendations To Support
The Framework
The proposed account task recommendations introduced new, meaningful actions that expanded beyond the typical identity-only tasks.
AI Powered Framework
Account Task Recommendation Framework
We used Voice-of-Customer insights to identify the most common account tasks and grouped them into a set of simple, meaningful themes. I partnered with a Senior Content Designer to ensure the framework was easy to understand and supported clear, consistent messaging.
| Critical Tasks | If the customer doesn’t take action will this compromise their security or identity? |
| Celebrating Milestones | If something is achieved are we taking time to engage in celebratory moments? |
| Product Features Not Being Utilized | If the customer isn't aware of a feature and its benefit did we inform them? |
| First-Time Customers | If I'm new to Intuit what tasks should I be handling first before all others? |
| Support Retention & Revenue | What tasks do I need to complete to prevent disruption of product use |
Research Insights
What We Learned
Three key patterns were identified that informed the design decisions:
Repetition drives dismissal - when customers see the same prompt multiple times, they don’t assume relevancy and stop trusting guidance. In a pilot we observed a 30% drop in prompt dismissals (based on internal logs). Long term, I'd also watch for false confirmations, meaning customers who tap "Looks good" without really checking, since a version that wins on taps but leaves outdated numbers in place isn't actually winning.
Context matters - customers respond better when the recommendation clearly explains why it’s being shown. “Customer feedback post-launch noted that the recommendation felt “clearer” and “more helpful” (verbatim quote taken from customer feedback after the pilot launch).
Identity-related tasks carry emotional friction - updating account details (phone, email, password) requires trust - customers need confidence that the prompt is legitimate and safe.
Takeaways
This project marked Intuit Account Manager’s first foray into embedding AI/ML into customer identity experiences. Instead of hard-coded rules, the system now adapts to customer history, surfacing the most relevant task at the optimal time. Our design work not only boosted engagement but also revealed insights about how customers perceive value from AI recommendations: trust, timing, and clarity matter more than volume.