6 min read
B08

Builders · No. 26

Do Things That Don't Scale

The founders of Airbnb went door to door with a camera. The lesson was never about photos.

· · 6 min · 1,391 words

In short

  • Most startups do not take off on their own, so founders have to recruit and delight early users by hand.
  • Manual work in the early days is a research method, because doing the job yourself shows you what the product really needs to be.
  • Zappos, DoorDash and Stripe all began with founders personally doing work that software would later handle.
  • Unscalable work becomes a trap when it hides the fact that the product itself is not working.
  • Cheap AI makes it tempting to automate from day one, but you can only automate well what you have first understood by doing.

In early 2009, two young founders went to New York with a good camera and started knocking on doors.

Their company was a website where people rented out spare rooms and apartments to travellers. It was not working very well. Bookings were flat, and when the founders looked closely at their listings in New York, they noticed something embarrassing. The photos were terrible. Dark rooms, blurry shots, phone pictures of a bed taken from the doorway.

So Brian Chesky and Joe Gebbia did something no growth strategy deck would recommend. They visited hosts in person, one by one, and photographed their apartments themselves. They talked to the hosts while they were there. They watched how people actually used the product.

That company was Airbnb. The photo trick worked well enough that it later became a full programme of professional photographers for hosts. But the part worth remembering is not the photos. It is that the founders went and did the work with their own hands, in a way that could not possibly scale to millions of listings, and learned more in those visits than any dashboard could tell them.

The essay

Paul Graham, who co-founded the startup programme Y Combinator, wrote about this in a 2013 essay that has become required reading for founders. It opens plainly: "One of the most common types of advice we give at Y Combinator is to do things that don't scale."

His argument starts with a misconception. Many founders imagine that if they build something good, users will arrive on their own. Sometimes that happens. Usually it does not. In the early days, a startup is fragile. It is more like a match you have to shelter with your hands than a fire that will spread by itself. Someone has to go out and get the first users. That someone is the founder, and the method is usually manual.

Graham gave examples from companies he had worked with. One was the Stripe founders, Patrick and John Collison. When someone agreed to try their payment product, they would not send a link and hope. They would say, in effect, give me your laptop, and set it up for them right there. Graham said founders at Y Combinator called this the Collison installation.

Another was Wufoo, an online form builder, whose team sent handwritten thank-you notes to new users. Imagine receiving one. You would remember that company for a long time.

Neither move scales. That is the point.

Being the software before you write it

Graham focused on recruiting and delighting users, but there is a third reason to do unscalable things, and for many founders it is the most important. Doing the work by hand is how you find out what the work actually is.

Think about Zappos. In 1999, Nick Swinmurn wanted to test whether people would buy shoes online, which at the time seemed like a strange idea. Instead of building a warehouse and buying inventory, he went to local shoe stores, photographed their shoes and put the pictures on a website. When someone ordered a pair, he went back to the store, bought them at full price and shipped them himself.

As a business, it made no sense. There was no margin to speak of. As an experiment, it was brilliant. It answered the question that mattered most, whether people would buy, before he spent anything serious on the answer.

DoorDash started in a similar way. In 2013, a group of Stanford students built a simple website listing menus from restaurants in Palo Alto, put a phone number on it, and delivered the orders themselves. There was no clever logistics system. There were students in cars.

What did they learn? Things nobody could have written in a business plan. How long restaurants really take to prepare an order. Which addresses are hard to find. What customers actually complain about. When you are the delivery driver, you learn the delivery business from the inside.

Anyone who has ever tried to deliver something in the Gulf knows this kind of lesson. For years, many addresses here were really descriptions: the building near the mosque, the villa behind the bakery, the second turn after the roundabout. No map or software spec would have told you that. Only doing deliveries would.

In his book The Lean Startup, Eric Ries describes a company called Food on the Table that helped families plan meals around what was on sale at their local grocery store. At the start, it served a single customer. The founders personally visited her each week, brought a list of recipes and deals chosen for her, and worked out her shopping by hand. Only after they had served customers manually for a while, and understood what they really wanted, did they begin to automate the pieces. Ries called this a concierge approach.

When you are the delivery driver, you learn the delivery business from the inside.

Automation is a summary of experience

Here is a way to think about it. Every piece of software is a summary of decisions. Someone decided what happens when a user clicks this button, what the form asks, what the email says, what to do when an order is late.

If you write that software before you have done the job by hand, you are guessing at all of those decisions. Some guesses will be right. Many will be wrong, and they will be wrong in ways that are now baked into code, which makes them slower and more expensive to change.

If you do the job by hand first, the decisions reveal themselves. After you have onboarded fifty customers personally, you know which three questions every one of them asked. After you have handled a hundred orders yourself, you know which step always goes wrong. Now you can automate, and the automation will be a summary of real experience instead of a summary of guesses.

That is why unscalable work is not a detour from building the product. Done right, it is the first draft of the product.

When it becomes a trap

Doing things that don't scale is good advice with a dangerous shadow. There are a few ways it goes wrong.

Hiding a broken product behind heroics. If users only stay because the founder personally calls them every week, the product may not be working at all. The manual work should be teaching you how to make the product good, not substituting for it forever. A useful test: if you stopped all the hand-holding tomorrow, would anyone keep using it?

Never stopping. Some founders get comfortable doing everything themselves and never make the jump to systems. The point of manual work is to learn what to automate, and then to actually automate it. Once you have done the same task by hand the same way fifty times, it has stopped being research and become a job that a machine should be doing.

Confusing unscalable with unsustainable. Handwritten notes to your first hundred users are a gift. Promising every customer a personal call forever is a promise you will break. Do unscalable things for a phase, on purpose, with an end in mind.

The new temptation

There is a modern twist to all of this. Automation used to be expensive. You needed engineers, time and money to build a system, so founders were pushed toward doing things by hand at the start whether they liked it or not.

Now, with AI tools, you can build an onboarding bot, an automated email sequence and a support agent in an afternoon. It is tempting to skip the manual phase entirely and automate from day one.

Be careful. Automating something you do not understand just produces confusion faster. An AI agent that answers customer questions is only as good as your understanding of what customers actually ask, why they ask it and what a good answer looks like. You get that understanding the old way, by answering the questions yourself.

The cheaper automation becomes, the more valuable the knowledge you can only get by hand. Talk to the users. Do the deliveries. Fix the photos. Set up the product on their laptop while they watch.

Then, when you finally write the code, you will not be guessing. You will be writing down what you already know, and that is what software was always supposed to be.

Further reading

  • Do Things That Don't Scale — Paul Graham
  • The Lean Startup — Eric Ries
  • Delivering Happiness — Tony Hsieh

More from the shelf

B01 · Builders · 6 minDistribution Is the ProductThe best thing rarely wins. The best-known good thing does, and one researcher's blog shows how that works. B02 · Builders · 7 minHow Strangers Learned to Trust Each OtherEvery great leap in trade came from a new way to trust someone you would never meet. O01 · Origins · 8 minFrom Pearls to PortsDubai lost the only industry it had, then built for ships, planes and traders that did not exist yet.

See the whole library →