“WE EVEN HAD to host the servers in our own office.” Naval Ravikant laughs as he describes how in 1999 he and some friends founded his first startup, Epinions, a website for consumer reviews. They had to raise $8m in venture capital, buy computers from Sun Microsystems, license database software from Oracle and hire eight programmers. It took nearly six months to get a first version of the site up and running.

By comparison, setting up Mr Ravikant’s latest venture, AngelList, a social network for startups and investors (see article), was a doddle. The cost was in the tens of thousands of dollars, which he put up himself. Hosting and computing power was available, for a small fee, via the internet. Most of the software needed was free. The biggest expense was the salary of the two developers, but thanks to nifty programming tools they were able to do the job in a few weeks.

Mr Ravikant is not the only serial entrepreneur with such a tale to tell. Since the start of the first dotcom boom in the mid-1990s, launching startups has become dirt cheap, which has radically changed their nature. What was once a big bet on a business plan has become a series of small experiments, an ongoing exploration. This shift has given rise to a whole new set of management practices.

Not all newly created firms qualify as startups. Steve Blank, a noted expert in the field, defines them as companies looking for a business model that allows for fast, profitable growth. The aim is to become a “micro-multinational”, a firm that is global without being large. Many of them are simply small businesses that use digital technology. A growing number are “social enterprises”—firms with a social mission.

Perhaps the biggest change is that computing power and digital storage are now delivered online
In the past, startups almost universally began with an idea for a new product. Now the business usually begins with a “team”—often two people with complementary skills who probably know each other well. These “founders” (a term now used in preference to “entrepreneurs”) often work through several ideas before hitting on the right one.

Such flexibility would have been unthinkable during the first internet boom. Startups had to build from scratch most of the things they needed, particularly the computing infrastructure. Today nearly all of the ingredients needed to produce a new website or smartphone app are available as open-source software or cheap pay-as-you-go services. A quick prototype can be put together in a matter of days, which explains the astonishing success of organisations such as Startup Weekend. Since it was created in 2007, its volunteers have organised more than 1,000 weekend hackathons with over 100,000 participants in nearly 500 cities, including such far-flung places as Ulaanbaatar in Mongolia and Perm in Russia.

Perhaps the biggest change is that computing power and digital storage are now delivered online. At Amazon Web Services, the biggest “cloud” provider, the basic package is free and includes 750 hours of server time. And if a new website or smartphone app proves hugely successful, new virtual servers can be added almost instantly for a small fee.

A whole industry of services to help startups tweak their offerings has sprung up, too. Optimizely, itself a startup, automates something that has become a big part of what developers do today: A/B testing. In its simplest form, this means that some visitors to a webpage will see a basic “A” version, others a slightly tweaked “B” version. If a new red “Buy now” button produces more clicks than the old blue one, the site’s code can be changed there and then. Google is said to run so many such tests at the same time that few of its users see an “A” version.

To see how people actually use their products, startups can sign up with services such as usertesting.com. This pays people to try out new websites or smartphone apps and takes videos while they do so. Firms can tell the service exactly which user profile they want (specifying gender, age, income and so on), and get results within the hour.

Round and round we go
Startups today are in a constant feedback loop, which means they have to be run in a different way from their dotcom predecessors. It was only a question of time until someone wrote down these new management practices, just as Luca Pacioli, a Franciscan friar and mathematician, wrote down the principles of double-entry book-keeping as used by merchants in Venice in the late 15th century.

This time there are actually two competing Paciolis, Mr Blank and Eric Ries. They both observed the same company, IMVU, an instant-messaging firm, where Mr Blank was an investor and Mr Ries the chief technology officer. Their approaches differ somewhat, but they come to much the same conclusion: that the old model of launching a startup or a new product, encapsulated by the phrase “build it and they will come”, no longer works. Instead, firms have to find out what customers want. That involves building something, measuring how users react, learning from the results, then starting all over again until they reach what is known as “product-market fit”.

In his book “Four Steps to the Epiphany” and his more recent “Startup Owner’s Manual” Mr Blank tells readers how to tackle what he calls “customer development” (as opposed to product development), exhorting them to “get out of the building” and find out what people really need. Mr Ries’s “The Lean Startup” is more of a manual for continuously improving an online offering.

Even more than Mr Blank, Mr Ries has given startups a vocabulary to describe what they do. They should start with a “minimum viable product”, or MVP, a sort of trial balloon to gauge the audience’s interest. They should always test their assumptions, aiming for “validated learning”. And if their strategy does not work, they should “pivot”: in essence, throw in the towel and start again with a different product. Mr Ries even prescribes a new form of accounting for innovation: startups should keep meticulous track of their experiments and how these influence “meaningful metrics” (not just a rise in the number of users, but what they do with the product).

Startups also often use a related method called “objectives and key results” (OKR). It was invented by Intel, a chipmaker, and later adopted by such firms as Google and Zynga. The idea is that all parts of a company—the department, the team and even individual employees—not only set themselves clear objectives (increase sales by 25% in the next quarter), but pursue “key results” that help them get there (hire two new sales people or increase clicks by 10%). “Within that framework we can then iterate, which allows us to stay lean,” says Carl Waldekranz, the chief executive of Tictail, a Swedish startup, which makes it easy to build online shopfronts.

Pacioli’s ideas spread because the printing press had just arrived in Venice. His treatise about double-entry book-keeping was one of the first books printed there, in 1494. Mr Ries’s book has also spread his message, but so have his many speeches, YouTube videos and what he calls the “lean movement”. More than 1,000 lean-startup groups worldwide meet regularly to discuss the approach. Outfits such as Lean Startup Machine organise workshops. Others, including Luxr, sell teaching materials. Yet others offer tools to track a startup’s performance continuously.

Having a lean time of it
The lean methodology has caught on quickly, but implementing it is not easy. “Once you get going, there’s no way you can sit down in a relaxed state of mind and think about the next test,” explains Shawn Zvinis, co-founder of Tab, a service to let people keep a virtual tab in London shops. It folded in December, in large part, he says, because it added all kinds of bells and whistles that users did not want.

When building Buffer, a service based in San Francisco that lets users put tweets on hold to be sent later, Joel Gascoigne, the British founder, largely stuck to Mr Ries’s methodology. Even today, two years after the launch of the service, new features are first tried as an MVP and any changes A/B tested. Buffer now has more than 1.2m users, with 13,000 of them paying at least $10 a month for extra features. Yet staying lean has been a struggle: “As an entrepreneur you’re meant to be bullish about your opinion. But lean means that you constantly have to remind yourself that you could be wrong.”

Some people are wondering whether founders of lean startups are still entrepreneurs in the conventional sense, rather than empiricists who try to find a profitable niche. Others question whether lean startups are capable of significant innovation. “Lean provides a useful toolkit, but it can bias you towards the incremental rather than the transformational,” says Scott Nolan of the Founders Fund, a venture-capital firm that makes big technology bets, such as an investment in SpaceX, a space-transport company. “You cannot simply iterate your way into orbit.”

Mr Ries admits that his methodology has limitations. In his book he warns of “analysis paralysis”, when founders lose sight of the strategic forest for all the testing trees. Yet he certainly thinks big. His model is Frederick Winslow Taylor, the father of scientific management in manufacturing. Taylor aimed to reduce waste in material resources; Mr Ries wants to avoid squandering the mental kind. In Taylor’s time, Mr Ries argues, most innovation was devoted to increasing the productivity of workers and machines. Today the world has the capacity to build almost anything imaginable. “The big question of our time is not ‘can it be built?’ but ‘should it be built?’,” he writes.

Staying lean certainly helps to refine founders’ ideas, but to industrialise the rapid creation of companies they need something else: accelerators.