This is the third of three articles discussing Startup School 2008, a free conference that happened at Stanford on April 19. The first article covered Paul Graham and David Heinemeier Hansson, who talked about building the right product and how to make money from it. The second article talked about VC funding. This one covers five remaining presenters.

Listen to your users – but not too much

Paul Buchheit, creator of Gmail and FriendFeed, talked about the process of listening to users. User feedback is important, of course, but when it comes to this feedback, listen != obey. Often, users’ feedback needs to be decoded and interpreted; a user may ask for one thing but really need/want something else. At Gmail, for example, a lot of users asked for the ability to reply to a message right from the inbox. After listening to users, though, Paul learned that what they meant was that it took too long to view a message and reply from there, which means that the real problem was that Gmail was too slow. So they sped up the system, and users stopped asking for this feature.

Similarly, users sometimes ask for conflicting things. 5% of your users may ask for Feature X, but what if Feature X makes the system less useful or more cumbersome for the other 95%?

The long and short is to care about user feedback, but don’t be a slave to it. Users don’t always know what’s good for them, and what’s good for one user may not be good for another.

EC2 is interesting

Jeff Bezos of Amazon pushed AWS. Hosting is basically “undifferentiated heavy lifting” – it needs to be done, but it doesn’t matter who does it. In other words, your site will be worse if it is hosted poorly, but once you reach a baseline of quality, your site won’t be any better or worse based on who does your hosting. Jeff compared this to electricity: 100 years ago, a factory may have its own generator for creating electricity. But with modern electrical grids, most businesses would be crazy to create their own electricity.

I’m not sure if EC2 is quite ready (though persistent storage and elastic IPs help), but it definitely has promise. Indefinite auto-scaling of the application layer sounds great. Unfortunately, EC2 doesn’t offer much to help database scaling, which is a harder problem anyway; so capacity for thousands of web/application servers might not mean anything if your database is the bottleneck. And you almost certainly want your database and your application servers in the same network.

There is one thing that EC2 is perfect for, though: asynchronous processing. If you need to do a lot of processing in the background (see Zencoder, NY Times, Animoto), then you have an easy choice between EC2 and purchasing thousands of servers to meet a short term need.

Tell TechCrunch a Story

If you want to get on Techcrunch, you’re in good company – just about every startup would like a positive writeup from Michael Arrington, and the traffic that follows. But TechCrunch has grown to the point where it can’t write about every startup, so it is picky. Michael talked about how to stand out from the crowd. The solution is what any good PR firm knows: tell a story. “Please write about GoodReads” is not a story. “GoodReads Launches” is a story, but not a very interesting one. “GoodReads gets 1,000,000 users” is a bit more interesting. “GoodReads to publish Harry Potter 8 online” will be picked up by TechCrunch, Mashable, CNN, NY Times, and everything else.

Michael also cautioned that sometimes the stories that are most interesting to the media (TechCrunch included) are the ones that you don’t want written. Leaks, rumors, and scandals play well in the press. So you might want to think about how you can make the most out of gossip and “negative” stories if they arise. And if TechCrunch bashes you, says Michael, engage them – don’t ignore or throw stones.

Finally, Michael recommends that startups not pay for PR, but rather engage the community through blogs, Twitter, discussions, etc. Personally, I don’t see what’s wrong with combining traditional PR help with social/community relations, but talk to me in another year or two and I’ll have a better perspective on the matter.

Be so good that they can’t ignore you

Marc Andressen has several big successes under his belt and is currently working on Ning. He did a Q&A session with Jessica Livingston, asking questions voted on by Startup School attendees. The discussion was good, though as Q&A it didn’t really have a single theme. One thing that stood out was Steve Martin’s advice to young comedians wondering how to break out: “Be so good that they can’t ignore you.” This is great advice to startups, who sometimes get caught up in issues of funding, pricing, staffing, marketing, networking, etc. All of these things are important, but if your product is good enough, the others will fall into place.

Data provides the best feedback

Peter Norvig closed the day and talked about learning algorithms, not surprisingly. But of course, he connected them to startups. A good formula for a startup goes something like this:

1. Start small
2. Go fast
3. Gather feedback *
4. Iterate

Learning algorithms can help with #3. The right dataset plus 20 lines of code can tell you important things about your users, your market, etc.

Conclusions

• Build something worthwhile, and build it really well.

• Build something you care about.

• If you need funding, go in with your eyes open and with the right advisors. And don’t let the quest for funding kill you before you even get out the door.

• Build something that people want and need.

• Iterate quickly and frequently, based on user feedback; but dig into the feedback rather than following it blindly.

• Engage the community.