BUS 714 · Peer Learning Discussion 3
Labelbox
A path to product-market fit that was contrarian, and right.
AI training-data platformfounded 2017Series D · $188M+ raised
Presented by Gordon Wu
Venture & founders
Three engineers who lived the problem first
CEO
Manu Sharma
Aerospace engineer; product roles at satellite-imaging firm Planet.
COO
Brian Rieger
Manu’s classmate; data science at Boeing.
Co-founder
Dan Rasmuson
Technical co-founder; shipped “half a million lines of code” in a year.
300–400satellite images a day at Planet, split across people with desktop tools and thumb drives.
Initial attempt · the 4Ps
Buy, don’t build, your labeling tool
Persona
Technical leaders (CTOs, ML engineers, PMs) at AI startups.
Problem
No way to label data together, so teams built internal tools.
Promise
Buy your labeling infrastructure instead of building it in-house.
Product
Cloud labeling app: many users at once, roles & permissions.
Misalignment: built for self-serve small orgs, but enterprises showed up first.
Challenges
“An absurd idea that was never going to work.”
AI experts who had tried to build labeling tools themselves
- Unknown team with only mock-ups; most expert feedback was negative.
- “Quite early to the party”: the market barely existed in 2017.
- The real competitor was “build it ourselves.”
Progress · validate before building
Ask questions that open the conversation
Closed: avoid
“Are you building an internal tool?”
Yes / no. The conversation stops.
Open, with guardrails
“How are you going about labeling your data?”
Open-ended, but stays on topic.
V1Build trigger: a medical-imaging CTO tried to hire Dan to build an internal labeling tool. Commitment, not compliments.
Progress · Level 1 → Level 2
Priced day 1: Free + Pro $100 → $200/moFounder-led salesPersona: self-serve → enterprise
Achieving product-market fit
From a labeling tool to the AI data platform
$100KARR by Apr 2018, ~3 months after beta
$1MARR by Apr 2019
37customers by Dec 2018, incl. Allstate & Bayer
1M+assets labeled per month (Jan 2019)
Labeling→Boost services (2020)→Model debugging (2021)→Catalog data (2022)
Gap: the article gives no retention, margin or CAC figures.
Personal insights
What I’m taking into my own ventures
- Validate the problem before building, especially as an engineer whose instinct is to prototype.
- Ask open-ended, empathetic questions; avoid yes/no.
- A mock-up and the right questions can validate the problem.
- Don’t lock in a persona: planned small orgs, got enterprises.
“Your idea has to be both contrarian and right.”
Manu Sharma, Labelbox co-founder & CEO
Thanks for listening
Q&A
Read the full article: Labelbox’s Path to PMF: Founders Need to Be Contrarian and Right (First Round Review)
Speaker note