How to Start a Startup

How to Start a Startup. An essay by Paul Graham, March 2005.

Three things, all doable

Graham gave this talk at Harvard in March 2005, months before Y Combinator funded anyone. Its stories come from his 1990s startup, Viaweb, so the numbers are dated.

His recipe has three parts: good people, something customers truly want, and spending as little as possible. Most startups that die fail at one of them.

Each is hard, but none takes genius. Do all three and you'll probably succeed, and since success tends to make founders rich, wealth is within reach too.

Ideas are cheap

Start with the idea. It needn't be brilliant: what people use is often so bad that beating it is easy. Google's plan was just search that didn't suck.

So take something people already try to do, and make it not suck. In 2005, Graham pointed at online dating.

An idea alone is worth little. Most VCs won't sign an NDA just to hear one. And plans change: Microsoft started out selling programming languages.

An idea is really a starting question for its founders. That's why the team matters more: good people can rescue a weak idea, never the reverse.

Hire animals

One of Graham's best hiring tricks: would you call this person an animal? That is, so serious about their work that they pass professional and become obsessive.

Picture a salesperson who never takes no, or a hacker up at 4 a.m. over one bug. To test someone, say it about them. If it sounds funny, they're not.

For hackers, three more tests: truly smart? Gets things done? Bearable? The third rarely mattered: the truly smart don't posture. They say I don't know.

The founding team

Startups usually begin with friends, and universities are where smart people meet. Don't schmooze, though. Just build your own projects with people you like.

His ideal: two to four founders. Alone, the load is too heavy; with eight, arguments harden into factions. And include hackers: business people can't judge them.

Do you need business people? Graham found business no great mystery. What you do need is someone who understands users, and hackers can learn that.

Because not understanding users sinks more startups than anything else.

Make what customers want

Can you name a restaurant with truly great food that went under? For startups, the death certificate says ran out of money. Usually, nobody wanted the product.

The cure: show users a rough version early, then improve it. The alternative is a Hail Mary: plan for a year, spend $2 million, and find nobody wants it.

Your first plan will be wrong somewhere; building shows you where. Viaweb expected web consultants and catalog firms as customers. It got small merchants.

Watch your users

So Viaweb pushed hard on ease of use. Hawking's editor said each equation would halve his book's sales. Reverse that: a bit easier can mean twice the sales.

To learn what users want, watch them. Viaweb used trade shows less to sell than to build real stores with visitors, and see what they needed.

You're rarely the typical user; software for people like you tends to be open source. So here's an idea generator: what do people unlike you want?

Aim at the low end

For each famous startup like Google, twenty more live in niche markets or hidden infrastructure. The best odds are there, and where things suck most, like corporate IT.

And sell to small companies first. You can outhack Oracle, but you can't outsell its salespeople.

Small also wins later. Cheap, simple products keep gaining power, like water filling a room, until the pricey ones hit the ceiling. Sun did that to mainframes.

So own the low end. If anyone gets underneath you, you're the one in their sights.

Seed money

You'll likely need investors. Graham sees a startup as pass/fail: a bigger share of a failure is worth nothing, so trade stock for better odds.

In 2005, seed money meant tens of thousands of dollars from angels, rich individuals who often decided within a week. Viaweb's was $10,000, from a friend.

The business plan can be short: what you'll do, how you'll make money, who you are. And you've split the stock right when everyone feels slightly shortchanged.

Ask about IP

When you incorporate, have every founder sign that their ideas belong to the company, and that it's their only job. Then ask what else they've signed.

Viaweb learned this the hard way. Mid-sale, they found one of their people had once signed his ideas over to the company paying for his grad school. The deal froze.

Cash ran low, their investors played hardball, and the buyer walked away.

It ended well: six months later, Yahoo bought them for much more. But ask about IP first.

Valuations and VCs

Selling part of a company prices all of it: 5% for $100,000 means $2 million. Early on, that price is just a bet on ideas and work still to come.

VC money is bigger: millions, months to decide, more strings. Some VCs install a polished CEO, a type Graham's team called newscasters. Viaweb stuck with angels.

Still, you have more leverage than you think: there's more money than good startups, and a VC's dollar is just a dollar. His pick: the most, soonest, fewest strings.

Get big slow

Once the money arrives, Graham's advice is blunt: don't spend it. Whatever else went wrong, running out of cash is how most startups die.

The dot-com bubble said get big fast, but being first matters less than it seems. Google came late to search, in 1998, never advertised, and buried the leaders.

Viaweb had about 70 users by late 1996. That slow year was a lab: talking with users, they pulled far ahead of every rival.

That's the key: understand your business. Google understood search; Yahoo didn't. Please users first. Graham's name for the approach: get big slow.

Stay cheap, hire little

VC millions make you feel rich, but only revenue makes a company rich. Graham's model: grad student, not law firm. Viaweb's chairs were so cheap their arms fell off.

His advice: rent an apartment, not an office, since hacking is easier at home. Pick a spot full of restaurants, so the team eats together, then keeps working into the night.

Above all, says Graham, don't hire. People are a cost that recurs, and they slow you down. Hire only for work you need but can't do yourselves.

Headcount impresses people, but for equal revenue, fewer is better. Viaweb had 20 people to its main rival's 140, and kept beating it.

Who should start one

Should you start one? Graham thinks far more people could than realize it. He was terrified himself, and only did it because, as a Lisp hacker, he had few options.

His sketch of the right founder: a good hacker, roughly 23 to 38, who'd rather make their money in a few intense years than slowly, over a lifetime.

Younger, and you haven't seen a real business from inside, and no one takes you seriously. Older, and stamina fades, and risk is harder to bear, especially with kids.

Forty years in four

And you have to want it. A startup packs a career into a few years: about four, flat out, instead of forty at a normal pace. For Graham, that meant little but work.

The grind is no worse than a normal career, he thinks, just compressed. What it buys you is time. And failures fail fast, so even grad students can go back.

If it's what you want, then, do it. No business degree required. Graham's summary: “Build something users love, and spend less than you make.”

In short

Start with good people. They matter more than the idea.

Show users early versions, and build what they actually want.

Raise money for better odds, then spend as little as you can.

None of it takes genius. If you want it, it's doable.

The original is long, and better than this. Read it: paulgraham.com/start.html