B2B2C & consumer products · 0-1 to scale · AI-native

Emily Yeh

Harnessing AI to build demos, not memos.

9 years taking products from 0-1 to scale, with a focus on consumer growth and AI. As founding PM at Bolt, I built consumer products across a 10M+ user network and grew account activation from 2% to 70% through behavioral data and experimentation. Today I design, build, and ship my own AI and LLM-powered products end to end.

Emily Yeh

Selected Work

Guest checkout, then logged-in 1-click · ~10% conversion lift

The 1-click checkout pivot

Founding PM · Bolt

Partnered with engineering, design, and data, and brought leadership along.

Problem

Checkout was tuned through endless A/B tests on a guest flow; conversion had plateaued.

Insight

Fullstory behavioral analysis and SQL cohort work showed logged-in checkouts converted ~50% higher than guest. The lever was identity, not another button test.

Decision

Championed a strategy pivot from A/B-tested guest checkout to a 1-click, logged-in checkout, and brought leadership along.

Outcome

Shifting traffic to logged-in checkouts lifted overall checkout conversion ~10%, with logged-in converting 50% higher than guest.

What was hard

Leadership had invested in the guest-checkout A/B program, so the hard part was making the case to stop optimizing a local maximum and re-bet the roadmap on identity.

Bolt ID: a verified shopper identity card linked to a network of merchant, fraud, and verification signals

Bolt ID, a universal identity layer

0-1 build · Bolt

Partnered with engineering, design, data, and risk across the merchant network.

Problem

Every merchant had its own siloed accounts, so shoppers re-registered everywhere and drop-off was high.

Insight

A shared identity network across merchants could make sign-in 1-click and lift trust and approvals at once.

Decision

Built Bolt ID 0-1, a universal identity and verification layer across the merchant network, and redesigned registration with embedded phone-based auth.

Outcome

Grew account activation from 2% to 70% across a 10M+ user network, where account holders are 15% more likely to make repeat purchases than guest shoppers, and reduced fraud while raising payment approvals.

What was hard

A shared identity layer only creates value if merchants trust it and adopt it, so the hard part was earning cross-merchant trust and driving adoption as much as building the auth.

ValonOS: System of Record, Workflows, and Task Management feeding one unified platform

Workflow & task management for loan servicing

Senior PM, Platform & 0-1 Products · Valon · 2024-2026

Partnered with engineering, operations, and a beta enterprise customer.

Problem

Mortgage-servicing operations ran on rigid, hard-coded processes and a generic helpdesk, so changing a workflow meant an engineering project and agents worked out of a tool that did not fit servicing.

Insight

If the work were modeled as composable servicing primitives, operations teams could configure their own automated and human steps (and AI-driven ones later) without waiting on engineering.

Decision

Designed and shipped a no-code workflow-orchestration layer and a task-management system built on those primitives, owned from customer discovery through company-wide rollout.

Outcome

Task management replaced the incumbent helpdesk (Zendesk) and reached 200 support agents across 15+ teams.

What was hard

A no-code platform only earns its keep if operators trust it and adopt it, so the hard part was change management and validating with the real persona, not just building the engine.

15K+
volunteer sessions supported across nonprofit programs

Bella, a volunteer-management CRM for nonprofits

Founder · 2017-2018

Designed and built the product end to end.

Problem

Nonprofits run on volunteers but manage them with a patchwork of spreadsheets, paper sign-up sheets, and email. Applications, scheduling, training, and attendance each lived somewhere different, so volunteer managers had no system of record.

Insight

Volunteers bring goodwill but limited time, so they only contribute well when the work comes with clear instructions, concrete goals, and timelines. Nonprofits, in turn, have to report volunteer hours for grant applications and rarely have an easy way to track them.

Decision

Built a 0-1 lifecycle platform rather than a static database: volunteer portals for instructions, training, and scheduling, manager dashboards for program performance and volunteer-hour reporting, and internal growth tooling, including metrics, support tools, and an A/B testing platform to track feature performance and monitor user behavior.

Outcome

Supported 15K+ volunteer sessions across nonprofit programs.

What was hard

Selling to nonprofits: long sales cycles and budget-constrained buyers. What wound the company down about a year in was market economics, not the product.

AI Product Lab (Independent)

Gist

Native iOS · private beta

Voice notes, transcribed and summarized by AI.

A native iOS app that records your voice notes and uses AI to transcribe and summarize them, with full-text search across everything you've captured. The bet is that capture should be effortless and retrieval instant, so the tiered subscription gates depth of AI summarization rather than basic recording.

Swift · SwiftUI · AI transcription + summarization

Available on request →
Orphea app screens: the feed, a show with its rating and review, and a profile

Orphea

Live on App Store

Letterboxd for live theater.

A consumer social app for logging, rating, reviewing, and discovering live theater, with a diary, a "want to see" list, recommendations, and a friends leaderboard. The product idea is that theater fandom deserves the same social memory film already has, so the design leans on the diary and the friends feed.

iOS · social · diary · recommendations

View on the App Store →

Clara

Personal build

Turns a wedding inbox into a plan.

A full-stack LLM tool (GPT-4o mini) that reads your email and extracts vendors, deadlines, budget, and to-dos into structured, trackable plans, with AI summaries, a Stripe-powered registry, and guest and RSVP management.

Full-stack · LLM-based (GPT-4o mini) · Stripe

Built end to end, hands-on with Claude Code.

Earlier & Recognition

Before the product work

  • Earlier Software Engineer at Index (acquired by Stripe) and OkCupid, where I built an interest-based recommendation feature that inferred each user's interests and surfaced the most relevant profiles to each viewer (featured in New York Magazine).
  • Education B.S. Computer Science, Carnegie Mellon.
  • Recognition Executive 10X Impact Award, Bolt.

Open to senior and lead product roles. Reach me at xemily.yeh@gmail.com.