Lessons from 2 search labs
I recently started Untitled Labs, a search lab startup, announced here. This got me thinking I should write about the last 2 startups I founded as search labs over the last 12 years. I have spoken publicly about these founding stories but never shared them in writing. Hopefully they’re helpful to other entrepreneurs who haven’t settled on a concrete idea but want to start a company to explore it.
We believed that Facebook, and social software as a movement, would be massive. Outside of joining Facebook as an employee, building Facebook apps was the best way to correlate to that movement.
Ideation & build process
Our idea generation approach:
- Imagine what kind of apps would be native on the social graph, like Excel was on PC or Google was on web
- Find popular apps on other platforms that could easily be ported
- Stay informed by top apps in the market (but don’t copy them!)
Many developers took the “easy” route, waiting for other developers to find good ideas and then copying and out-executing them. We had too much pride for this approach, believing that originality was the path to success. Although I would take that same path today it was likely to our detriment. Zynga used the opposite strategy and it worked out well for them.
In the early days the Facebook app space felt (and looked) like the wild west. Everything grew or died quickly and production values were very low. This didn’t impact the viability of apps; to the contrary people seemed to find them genuine, even endearing. It was like the internet had rebooted. Given the low development costs, we were able to launch 1–2 apps per week using a robust common technology platform.
What we tested
Our experiments generally fell into 3 categories: simple games, seasonal and current-events apps, and viral quizzes and pokes.
Over a period of 4 months we launched the following:
- Animal Racing — re-skinned dot racing flash game
- Bartender — simple adventure game where friends can order drinks from you
- Bounceman — ping pong game where you play against a friend’s score
- Cube Game — fun personality test with shareable results
- Defend Freedom — app to pretend-report your friends as terrorists
- Doodle Board — Etch-a-sketch for your profile page
- Dope Wars — drug dealing RPG
- FB Basketball — basketball skin of Dope Wars
- Holiday Wreath — customizable wreath for your profile page
- Impersonator — impersonate friends and celebrities with fake messages
- Kiss Quiz — romantic personality quiz
- Moon Me — virtually moon your friends
- My Pizza — pizza box for your profile page, allows friends to eat pizza you “buy”
- My Tip Jar — tip jar for your profile page, friends can tip a certain amount per day
- Oh My Shoes! — Pinterest-like app to collect and showcase your favorite shoes
- Snowglobe — customizable, shakeable snowglobe for your profile page
- Squid Hunter — squid spearing flash game
- The Flirt Club — send flirts to your friends
Snowglobe was our first success. It grew 3x+ day-over-day until the servers went down, hard. It turned out we were caching 100 Kb images of each user’s snowglobe which had quickly overwhelmed our disk array. By the time we devised a new caching strategy and got the app back online we had been cloned by several competitors. They ended up winning the holiday app market.
Seeing our first viral hit fail for preventable technical reasons was soul crushing. I remember mourning it and even changed my passwords to a variant of “snowglobe” in memory of what could have been.
Dope Wars was the second and largest success. We stumbled on the idea after looking for popular games in an online Windows app directory that seemed easy to build. One of the top games with several million downloads was “Drugwars”, originally created for MS-DOS in 1984. I had built a version of Drugwars in junior high for my classmates and realized it would be a great candidate for Facebook. After a 1-week build we launched and immediately saw very strong engagement. Our main metric, page views, which tied closely to active users, engagement, and revenue, doubled every few days for weeks. We eventually reached 10M daily page views.
The problem, again, was how to scale it. Every page viewed by a user opened a connection to MySQL and executed a number of read and write commands. The database was massively overloaded dealing with thousands of queries per second. We eventually built a custom caching system for user data which improved game performance dramatically. This system later formed the back-end of Mafia Wars, Farmville, and other major Zynga franchises.
After 2 months of live operations, Zynga reached out and offered to buy Dope Wars. At first we were unwilling to sell. The game was growing well independently and I feared they might steal trade secrets and back out during negotiations. We eventually came to terms and I left Austin to join their 30-person team in San Francisco.
Liquid Labs (2013-present)
We were inspired by how much technology is changing work. Given the “size” of the labor market — trillions — we were confident we could find an idea in the recently-named “future of work” space, possibly leveraging marketplaces or AI.
Ideation & build process
We centered around 4 themes that generally fit the thesis and our founding team:
- Experts on demand
- Fantasy angel investing
- Personal finance
Product ideas were generated hub-and-spoke around the themes, each of which we spent between 3–7 months researching. Overall we burned 75% of a $1.8m seed round with a team of 3–6 before finding the winning idea.
My main take-away from the process was that we should have spent more time on fewer ideas. There were a number of good ideas we doomed to failure by dramatically under-investing in them to make time for the bad ones.
What we tested
Over a period of 15 months we launched the following, roughly in chronological order:
- Natalie — fantasy football meets angel investing, users can make virtual angel investments
- Datecoach — help older divorcees re-enter the dating world
- GetExperts — get online video help about anything
- Sudo — on-demand help for software developers
- Storhaus — simple blogging platform
- Cilantro — do trial projects to interview at startups
- Macronerds — hire Excel macro developers on demand
- Houdini — automated QA testing
- Sudochat — essentially Discord
- Markov — Github for data science
- Prototyper — directory of template application prototypes to build from
- Over Coffee — Tinder for coffee meetings
- Sudo 2 — Tinder for developer work referrals
- Startup2048 — clone of the popular 2048 game using startup logos as the blocks
- Hobby.vc 1 — fantasy investing game built in Google Sheets
- Angel* — web based angel investing game
- Hobby.vc 2 — angel investing game on mobile
- Over/under game — make cash bets on startup outcomes
- Cash 4 clunkers — sell unused junk from your house
- Runway 2 — simplified personal finance calculator
- Gigster — push a button to hire a quality development team
The first 20 ideas we tested had a mediocre reception. We would beg users to try the products, only to have them often drop off within days. After a while our friends stopped responding to requests for beta testers. We forgot what success felt like!
Gigster was immediately a clear winner. We made a simple landing page that said “Hire a quality developer in 10 minutes”, a value proposition we imagined would resonate with many small businesses. We launched it on Hacker News and Product Hunt and in the first 48 hours had received requests for $3m worth of projects. After a week of discussion on the market and vision for Gigster we shut down all other activities and pursued it with the full team.
Bookings grew anywhere from 2–5x month-over-month in 2015, our first full year of operations, ending at $2m in December. We raised a Series A from Andreessen and grew the team. Fast forward to today, we have raised $32m, have 1,200+ customers including Exxon, Nike, Harley Davidson, and Google, and continue to grow revenues quarter after quarter. We hired a fantastic CEO last year after my departure. You can read more about that here.