Case study · Intuit QuickBooks

Uncovering
tax savings
with AI

I led the shift in how QuickBooks uses AI for self-filing business owners: from exporting data to surfacing tax savings. The result: real savings, more confidence, and a natural upsell path to TurboTax self-filing. Win win.

2x

Tax filing upsell CTR

$1,000

Avg. tax deductions added

100%

Leadership alignment on the vision

AI feature shipped

Catching the deductions businesses often miss

Context

Make QuickBooks' tax hub more valuable & stickier—powered by AI

Tax is the elephant in the QuickBooks room: the reason why everyone gets QuickBooks, yet mostly invisible. The main touchpoint: a year-end TurboTax cross-sell and export for the CPA-less business owner.

Intuit wanted more: year-round engagement and a larger share of the self-filing market.

Team: 1 product manager, 1 content designer, up to 9 engineers.

My role

Led the end-to-end AI design, and earned the next bet

  1. 01

    Initiated the founding research

    Led 25+ SMB sessions, pulling novel insight from weak signals and sparking the deductions agent idea.

  2. 02

    Led scrappy sizing experiments

    Shipped an in-product fake-door test to back the business case.

  3. 03

    Designed the new AI deductions agent, end to end

    Owned the design and the agent conversation model, defined and tested prompts across models, and iterated through 20+ SMB sessions.

  4. 04

    Spun up a higher-stakes AI tax savings agent

    Validated the riskiest assumptions, then coached a senior designer through execution (in progress).

quickbooks.com
Impact

A paradigm-shifting win-win

  • 2x CTR to tax filing upsell

    The confidence the feature builds doubled upsell click-through, lifting ARPC.

  • $1,000+ new deductions

    Average incremental deductions per user. A standout result, featured in the Q2 CEO investor update - QuickBooks starts paying for itself.

  • 11% above target

    Helpfulness beat an ambitious Intuit-wide target, with one of the highest AI engagement rates in QuickBooks.

  • Earned the next bet

    The strong engagement and results gave product leadership the confidence to invest further in AI powered tax savings.

Decision spotlight

Designing for relevance when you can't know the answer

A key design challenge was to deliver initial relevance without overwhelming users, and under a hard constraint: we couldn’t reliably know which deductions a given business was actually missing.

Explorations for the landing state

After

  • 01

    A deliberate "wide net"

    Since the AI couldn't filter "missing deductions" down to a precise few, I leaned into a deliberate "wide net" strategy to prompt users to recall deductions they didn't know existed across all tax categories. A second use case reinforced the approach: for new business owners, the same content doubled as education on what's even deductible.

  • 02

    Progressive discovery

    Rather than front-loading everything, I let users explore at their own depth — a confident few up top, with more categories or detail revealed on demand. I refined the AI prompt and evaluated responses across models to maximize relevance and specificity in the top-3 examples that drive first engagement.

  • 03

    Simple, but not generic

    I explored and iterated relentlessly to strike the content density balance, validating with users for detail that felt specific, not generic. Hence the concrete examples on the front screen.

Learnings

What I'd carry into the next AI product

  • 01

    Explore AI assumptions early

    AI capabilities and latency evolve constantly, and they shape the experience more than almost anything else - so probe them before committing to a design direction.

  • 02

    Revalidate interactions you reuse

    I assumed reused in-product experiences would hold up; some did not.

Next case study

Bringing clarity to tax owed