Designing a Zero-to-One Incubator Inside a Giant
Fix the incentive system and the time horizon first — everything else about internal incubation follows.
- Difficulty
- Expert
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 93%
Most corporate innovation labs are killed before they ship anything by the parent company's own operating system: semiannual performance reviews, standard leveling and comp, mandated infrastructure, and legal barriers to talking directly to customers. Lovinsky, who helped build Facebook's New Product Experimentation team, argues the design problem is almost entirely about which company processes you exempt the team from — and that the payoff should be scored on more than 'did you find the next Instagram.'
Origin
Noam Lovinsky, from co-founding Facebook's New Product Experimentation (NPE) team, org-sponsored and firewalled by Mike Schroepfer ('Shrep'). Nike's incubation lab is cited as the best external example he encountered.
Core principles
- 01Smart people game the incentive system whether or not they intend to — so design it first.
- 02A semiannual performance cycle is the wrong time frame for zero-to-one and creates adverse selection in who joins.
- 03Startups run on outsized-impact and outsized-incentive horizons, not on promo cycles.
- 04Zero-to-one requires doing things that don't scale — running an experiment with 100 people is not hard at Facebook scale, it's impossible.
- 05'Did you build the next Instagram' is the wrong bar; the discovery is close to a lottery, and the byproducts have real value.
How to run it
- 1
Design the incentive system and time horizon before anything else
This is the number one thing. If your zero-to-one incubator is evaluated and rewarded through the parent company's twice-yearly performance management process, you have already killed it. The time frame is wrong and it creates adverse selection in the people you attract.
Pro tip Pull toward startup-shaped incentives — outsized upside tied to outsized impact over multi-year horizons — because you are competing against those people leaving to start their own thing.
Watch out You cannot half-do this. Telling a big org 'we're throwing out leveling, comp and performance process for this group' is genuinely hard, and if you flinch you get the incentives back.
- 2
Get an executive who will firewall the team
The exemptions only stick if a senior leader will absorb the organizational pushback and say 'nope, they're going to get to do this, sorry.' Without that sponsor, every exemption erodes back to the default within a quarter.
Watch out The sponsor must be senior enough to overrule infrastructure, legal and HR — not just supportive in principle.
- 3
Free the infrastructure choices
Let every team pick whatever stack best solves the problem in front of it, on the explicit understanding that most of the code will be thrown away. In a Facebook- or Google-scale org this is one of the hardest exemptions to obtain and one NPE did well.
Pro tip Say the quiet part to the team: this code is disposable. It removes the instinct to build for a scale that doesn't exist yet.
- 4
Restore direct customer contact
At scale, PMs, engineers and designers essentially cannot talk to customers — everything runs through recruiting agencies, third parties and reports, layered games of telephone, hemmed in by legal constraints on what can be said to whom. Zero-to-one work is a hunt for faint signal, and faint signal does not survive indirection. Build the environment where the team sits next to the user.
Pro tip Restore the ability to run genuinely small experiments too — with hundreds of users, not millions. Community products in particular can't be born at a million users.
Watch out Legal and privacy constraints at scale are real, not bureaucratic theater. Solving this needs a designed exemption, not a workaround.
- 5
Score the full portfolio of outcomes, not just the lottery ticket
Set the light on the hill as 'find the next Instagram,' but budget realistically for champagne-level versus nice-dinner-level outcomes, and count the byproducts: features that get absorbed into mainline products faster than the mainline could have validated them, a first-class recruiting magnet, and a mirror that exposes what makes zero-to-one impossible in the core org.
Pro tip The Nike model: run a completely separate operating model, different people, different incentives, for product discovery — then plug the finding into the parent's distribution, marketing and growth arms once fit is found.
Watch out Judging the incubator solely on whether it hit the jackpot is judging it on whether it won a lottery. It will always fail that bar, and you'll shut it down having ignored the value it did produce.
In the wild
Lovinsky was one of the early members of Facebook's New Product Experimentation team, firewalled by Mike Schroepfer, with the charter of incubating big new ideas protected from the larger Facebook org. Teams chose their own infrastructure, worked small, and operated outside mainline product development constraints. Ideas like Threads lived inside the team.
→ It never produced the next Instagram, but it validated and built features faster than the mainline org could, served as a strong recruiting tool that seeded founders, and exposed which constraints were making zero-to-one work structurally impossible at Facebook scale.
Nike's lab runs a completely different operating model from the parent: a different type of person recruited, a different incentive system, and a fully separate product discovery process. Nike's core organization only enters once the lab has something with fit.
→ The parent's distribution, marketing and growth arms then plug in to scale the finding — separation for discovery, integration for scale.
Common mistakes
Running the incubator on the parent's performance cycle
Twice-yearly reviews are the wrong time frame for zero-to-one and select for the wrong people. Lovinsky's verdict is blunt: do that and you have already killed it.
Judging success only on hitting the jackpot
Discovering an Instagram is close to winning a lottery. Scoring the incubator solely on that discards the recruiting value, the mainline features it accelerated, and the organizational diagnosis it produced.
Trying to do zero-to-one at parent-company scale
Starting a community product with a million users, or trying to run a hundred-person experiment inside Facebook, isn't hard — it's impossible. If you don't design the ability to be small, the incubator is just a slower version of the mainline org.
Leaving customer contact mediated
Building a product from day one without sitting next to the customer means hunting faint signal through layers of indirection and telephone. At scale this is the default, and it must be deliberately exempted, not tolerated.
Is it for you?
Best for
Executives and product leaders chartered to build a startup-within-a-startup, internal incubator or new-product group inside a large, process-heavy company.
Not ideal for
Actual startups (you already have the incentives and the small scale), or companies whose leadership will not grant real exemptions from performance, comp and infrastructure processes.
From the transcript
“so one is I would say think really hard about the incentive system you know smart good people”
“it's just like it's the wrong incentive it's the wrong time frame it creates adverse selection problems for the sort of people that you bring…”
“just do what is best for the problem you're trying to solve in this moment knowing that you're likely going to throw away a lot…”
“I think in terms of was it a success or not because it didn't build the next Instagram I think is a little bit of…”
“what time horizons right like when you're when you're starting a company you're not thinking like in the next six months I'm going to get…”
From the episode
The happiness and pain of product management
Noam Lovinsky (Grammarly, Facebook, YouTube, Thumbtack)