💪🗣 About BoldVoice

BoldVoice (YC S21) is an AI-powered communication coach that helps non-native English speakers improve their pronunciation, clarity, and confidence. The app combines video lessons from Hollywood accent coaches with instant AI feedback as you speak, powered by our proprietary speech models. It has been downloaded by 5M+ users globally, is one of the top Education apps on the App Store, and has been featured in Forbes, TechCrunch, Wired, and Business Insider.

We recently raised a $21M Series A led by Matrix Partners, with Flybridge, XFund, and Y Combinator. We're a small, fast-moving team with a mission to help 1 billion people speak English clearly and advance their careers.

📊 About the Role

We're looking for our first Data Scientist to own how BoldVoice measures itself and decides what to build next.

This is not a reporting role. You will be the person who decides what "working" means for a feature, designs the experiment that tests it, calls the result honestly, and then digs into the data to find the next thing worth building. You'll sit at the intersection of product, growth, and our speech ML team, and your analyses will directly change the roadmap.

There is a large, unusually rich dataset waiting for you: many users, tens of millions of scored speech attempts, 100+ native language backgrounds, and a full consumer subscription funnel.

👩‍💻👨‍💻 What you'll do

Own experimentation end to end: help design A/B tests, choose the metrics up front, run the analysis with proper statistical rigor, and make the call on what ships
Build and own the metric layer the whole company runs on: activation, trial start, trial to paid conversion, retention curves, LTV, and learning outcomes
Run the deep dives that change the roadmap. Why do speakers of one native language retain twice as well as another? Which lessons actually move pronunciation scores? What separates a user who sticks around from one who churns in week one?
Turn behavioral and speech-scoring data into models the product uses: churn and conversion propensity, lesson recommendation, and personalization of the learning path
Partner with growth on paid acquisition measurement: CAC by channel and creative, payback, incrementality, and where the next marketing dollar should go
Work with the ML team to measure model quality in the wild, including how our speech models perform across accents and whether a model change actually improved user outcomes
Make data self-serve, so a PM, designer, or engineer can answer their own question without waiting on you

😀 Who you are

You take a question from "we don't know why this is happening" all the way to a recommendation the team acts on that week
You have strong statistical judgment. You know when a result is real, when it is noise, and you say so plainly even when the answer is inconvenient
You care about the product, not just the analysis. Your goal is to change what ships, not to produce a spreadsheet or deck
You want to work fast, you want to not get interrupted by meetings, and you want to not need to ask for permission to do things

✨ Requirements

At least 5 years in data science or product analytics, ideally at a consumer subscription app or a fast-moving startup
Deep experimentation experience: power analysis, variance reduction techniques like CUPED, handling multiple comparisons, and recognizing when a clean A/B test isn't possible
Strong SQL and Python (pandas, scikit-learn, statsmodels or equivalent). You have worked directly against a production database, not only through a BI tool
Fluency with the consumer subscription funnel: install to trial to paid, cohort retention, LTV, and paid acquisition measurement
Experience with product analytics and experimentation tooling. We use Mixpanel and GrowthBook on top of PostgreSQL
Up to date with latest developments in using LLM tools like Claude Code, Cursor, Codex or similar to rapidly prototype and ship code quickly
We work out of our office in downtown NYC, and want our team to be in person at least 3x a week
Nice to have: causal inference beyond A/B testing, exposure to speech or audio ML evaluation, or experience measuring learning outcomes in education products

🎯 What success looks like in 3 months

The team trusts one shared definition of our core metrics, and you own it
Experiment velocity is up and experiment quality is higher, because you set the bar for how we read results
You have found at least one non-obvious insight about our users that changed what we build
You are the person people come to before deciding whether a feature worked

🎁 What we offer

Senior Data Scientist at BoldVoice