I had a chance to sit down with Twitter cofounder Evan Williams at Foo Camp last weekend and talk about the overall state of the company. The first part of the conversation focused on their acquisition of Summize, which we wrote about earlier today (for more on the acquisition, see the Twitter blog, Twitter investor Fred Wilson and Summize investor John Borthwick). Summize, by the way, now redirects to search.twitter.com:
But we also talked extensively about the Twitter API and potential revenue models in the interview.
The Twitter API
Twitter has been criticized for severely limiting API functionality and discriminating against most developers by only allowing four partners (one of which they just acquired) to access the “firehose” XMPP feed. Williams was very forthcoming in the interview and said they are still working through data policy issues. Friendfeed has the full XMPP feed, for example, and is using it to quickly gain market share at Twitter’s expense.
From the transcript below:
MA: Is [the XMPP firehose] too efficient? I mean they are a competitor. It’s great to be open with open data and all, but they are a competitor of yours, and you are handing them all of the data.
EW: Well it’s not ideal, and there are definitely business implications to giving all our data away to anyone, and we are somewhat nervous about that.
These concerns about handing data out wholesale also lead to performance limitations on popular Twitter clients. Twitter could give AIR app Twhirl a XMPP feed and let it talk to each of its installed desktop clients, but that gives Twhirl a lot of competitive data. The current implementation is to have each installed client talk to Twitter independently via the API. Having a XMPP flow to each client isn’t feasible from a load standpoint, says Williams.
We’ll do a deeper dive on the XMPP/API conversation over at TechCrunchIT later today.
Twitter’s Revenue Model
Williams went into far more detail that I thought he would in answering my questions about Twitter’s business model. We talked about a lot of possible business models, but Williams focused on charging for commercial activity on Twitter, and keeping normal usage free. An example is an ecommerce site – Twitter may charge them per follower, per message and/or per sale. And they may also give users the ability to opt into “special offers” from commercial users, and charge for lead generation. From the transcript:
MA: What is your revenue model? Do you know yet? Have you thought about it?
EW: We’ve though about it. We had to do some thinking about that to raise a bunch of money, but it’s not actively in development right now. The broad strokes on the matter are obviously Twitter is being used for a lot of commercial purposes right now, in addition to social purposes. We think that works pretty well. We think there’s a lot of companies that we’ve talked to that seem to be getting a lot of value out of it. If that continues, if that becomes a rich world for users and the companies, we think we can extract some revenue from that.
MA: It might be difficult to define commercial activity versus not, right?
EW: It might in some cases, but in a lot of cases it will be really clear. I mean Woot.com is selling stuff. So maybe we just say, this is commercial usage and you need to pay for that and maybe there’s some features you get on top of that, that wouldn’t be as meaningful to personal users. There’s other cases, like yours, is that commercial or personal?
The full transcript is below.
Transcript Of Interview:
Michael Arrington: All right, I’m here with Evan Williams a co-founder of Twitter, We’re here at Foo Camp and we just chatted a little bit about some news that’s coming up, and you agreed to have a brief video interview with me. I’d like to talk about a few things, one is the news that’s coming out today about your acquisition of Summize, tell me a little about why you decided to buy them and what you’re disclosing about the acquisition.
Evan Williams: Sure, so we started talking to Summize 2 or 3 months ago, shortly after they released their twitter search, which a lot of people have seen, it’s gotten pretty popular. There are a few twitter search engines out there, but Summize was just really impressive with their user interface, the quality of the search, and a bunch of features they had. We were actually looking before that to build search into twitter, we were talking to some partners about that, we didn’t think we wanted to take on building that ourselves. And while we were actually going down the road with another partner, which I can’t disclose..
MA: That partner was going to help you build it for yourselves?
EW: Yea, we were basically going to outsource it to them. And it was a big internet company, and Summize came up, and happened to be connected with John Borthwick and Betaworks in New York, who I was friends with. So we started talking…
MA: John was an investor in Pyra.
EW: John was an investor and advisor to Pyra so I’ve known him for 10 years. It turns out that all the Summize guys were all ex AOL guys and most of them were ex AOL search guys, John was ex AOL. So anyways, we started talking and we really liked them and started exploring..
MA: When was this by the way?
EW: This was a couple months ago, I’m not really sure of the exact date. So that go interesting really quickly, and then we debated, well if we partner with them…long term this definitely seems like something we should own. Owning it the other way wasn’t an option, we thought we might have to outsource it. So we started exploring that, and we flew the guys out a couple weeks after that, that went well. We started doing some due diligence on both sides, met the rest of the team. About a month ago, Jack and Biz and I went to New York and spent a day and a half in Fed Wilson’s office with the whole team. We got along great, we were super impressed with their engineering and their culture.
MA: It’s 5 or 6 guys total right?
EW: Yea, 5 engineers, all of whom are going to join twitter now.
MA: How many engineers do you have now?
EW: As of today, their all twitter employees, as of yesterday, there were 12 people in ops and engineering. That includes 3 ops guys, so probably 9 engineers.
MA: OK, so you’ve increased it by a third. Was this a done deal at the Steve Jobs keynote with the 3g phone?
EW: No.
MA: Because that was obviously a sign that you were very close to them when you said that you were going to rely on them a little bit.
EW: That was the week after we met them in New York and we had a letter of intent, we had signed a letter of intent in New York. It was on the way, so we talked a little bit about this event coming up, how can we work together.
MA: So, I don’t want to relive history too much, but you originally announced a search function, a very basic one, I think Biz wrote about it on the blog last fall, maybe it was a little earlier than that. And that never launched, was that sort of a plan or did you guys..
EW: Yea, I think that got overestimated in our minds, or someone else’s mind, because we launched track and that was a very quick development, Blain hacked that out one night. And it was really cool and it gave a glimpse of what following keywords on twitter might be like. As you know it was only on SMS or IM, and then we had to kill it recently. And so the idea was always, this part of the vision of twitter was always there. We want to give access to more of this data that’s flowing through the system, not necessarily who is sending it, but based on what the content is and maybe other ways in the future. We hinted at that, because it was part of the plan, but we never really had it built.
MA: How much did you pay for Summize?
EW: Can’t say.
MA: Was it stock or cash?
EW: Mostly stock, a little bit of cash.
MA: and I assumed the employees you brought on with stock options and the standard stuff. Are they all going to move to San Francisco? Because their all over the country right?
EW: Yea 3 of them are in Virginia, one’s in New York, and one’s in Seattle. And they are all moving to San Francisco, 4 of them ASAP and 1 of them will hang in Virginia for a little while, but the goal is to get him over here eventually.
MA: And the head tech guy over there is Greg Pass, is that right?
EW: Yes
MA: He was CTO and now he’s your director of engineering?
EW: That’s right, director of engineering and ops actually.
MA: So he’s your senior tech guy?
EW: Yea, he’s going to be heading all of our engineering, which we’re very excited about.
MA: So you’re taking all of the employees except one?
EW: Yes, the CEO Jay Verde, is going to go do some new things. He started a couple of companies, so I’d be surprised if he didn’t go do that again.
MA: So, let’s going little bit about the API and XMPP. You’ve had some issues with it recently. Part of it seems to be just turning off parts of the API for stability reasons, as you’ve outlined in your status blog. There’s also been some questions about when that might come back on and also with XMPP which is being called a fire hose of your data. It’s only with very limited number of partners, one of them is Summize the other ones are…
EW: Friendfeed, Twittervision…
MA: There’s one other…
EW: Oh, Zappos.
MA: So when will you turn on XMPP for all partners, when will all the limitations on the API go away, just tell us sort of the state of the union there.
EW: There’s a bunch of different stuff that gets conflated there. First of all limits on the API…that’s something that has varied as we have had performance issues. It was at 70 for a long time. That’s kind of the default (that is 70 requests per hour). So a client like Twhirl or Twitterific can make that many requests per hour. We have taken that all the way to 20 at the lowest, maybe 10 at extreme times, but it has been at 20 or 30 for a few weeks. Actually last week we bumped that up to 70, and then 100 on Thursday or Friday, and I believe it is there now. Somewhat experimental, we are seeing how high we can push it. So it’s the highest it’s been.
MA: Are any features turned off right now? Replies work, everything is turned on right now?
EW: IM is still off. That’s the other thing. So API limits are the highest they have been in terms of HTTP requests. XMPP – there are two different things people talk about there. One is using Twitter over IM, over Gtalk, or Aim (which hasn’t been on in a while), and that’s all just regular usage. There’s also an API over XMPP. So in theory, a client like Twhirl has been built to support XMPP, which means same functionality for the user, except the data is coming in over the Jabber protocol. This is much more efficient, and there is a better user experience because things are coming in live and you don’t have to wait for the poll. That is definitely in the plan, I don’t have a date for that, but it is definitely something we want to support. We are going to turn that on as soon as we can. Then there is XMPP, the Firehose as you said…
MA: Which is what FriendFeed and Summize have.
EW: Yes, that is what those guys have. That is just taking all of their public updates and sending them over the wire to a single recipient, and we are getting a bunch more requests for that and there’s a bunch of issues, that’s why we don’t know exactly where that’s going to go in the future. We are offering it to Summize, even if Summize was still a separate company. We would probably want to offer it to them because it offers clear value. To offer search over all public updates they need all public updates. FriendFeed needs all the public updates for all the people on FriendFeed, so technically they don’t need them all, but it’s the most efficient way to give that to them right now.
MA: Although certainly FriendFeed could just read the RSS feeds as well.
EW: Yeah
MA: But this is more efficient for both sides?
EW: It is.
MA: Is it too efficient? I mean they are a competitor. It’s great to be open with open data and all, but they are a competitor of yours, and you are handing them all of the data.
EW: Well it’s not ideal, and there are definitely business implications to giving all our data away to anyone, and we are somewhat nervous about that. We are not nervous because of FriendFeed, we understand what they are doing and it makes sense. We don’t necessarily believe that anyone should be able to access all of this data for free and do what they want with it. Even aside from business purposes, we don’t see this as the best thing for users because they have sort of an implicit agreement. They are putting their data on Twitter, and even if they mark themselves public, that doesn’t mean their data should show up wherever someone may choose to put it all over the web. So we need to get a little bit more structured around that. And, ideally, a lot of the people asking for that firehose feed, there are alternatives for that, and there are more alternatives now that we have the Summize API. Summize has had an API for a while that offers their full search and all the operators and everything. You can get as Json or multiple flavors of XML, and that serves a bunch of the needs, a bunch of people want full feeds so they can filter down. And now you will be able to ask Summize or Twitter search, just give me the stuff already filtered. So we will still do the Firehose in certain cases for some companies that we have agreements with. The last factor on that is that we simply can’t offer many more right now because it is a ton of data, and there is a cost to that. And so even people that we like what they are doing, and would be fine to offer to right now, we just don’t have the capacity right now.
MA: Although we talked a couple of days ago about that fact that you probably wouldn’t give an XMPP feed to Twhirl because it would be a separate feed to each client. But you could give an XMPP feed to Twhirl the company, and let them talk to their clients.
EW: We could do that, technically, and we wouldn’t rule that out, I’m
just not sure that’s better for anyone. It is more complicated.
MA: Well all the complication is on there end. And you only have one interaction. The problem is you are also dumping all your data into Twhirl’s database, and that is not necessarily good for competition.
EW: Right. Possibly not, and we certainly need to understand that better. We prefer that Twhirl users are Twitter users, they are connecting to us, they are getting the data from us. It seems like a better arrangement. There are other cases, we are talking to Gnip about how we might be able to offload. And I think their intentions are on the up and up and it could be a good relationship for everybody if we are going through them in some degree.
MA: So it sounds to me like you are working through stability issues, as you mentioned on the blogs, but you are also working through data policy issues and what you want to be and what you want to do with the data. Which, it’s funny, sometimes you refer to it as user data, sometimes as our data, and it is essentially both. But you don’t really know where that’s going, other than the fact that you will probably continue with some partners with XMPP on the full feed when it makes sense, but you may limit it on others.
EW: Right, and the important thing there is not that we, there has been some speculation on what we want to charge on that data. We don’t think that’s our revenue opportunity. There are strategic questions, basically we want to understand and have some agreements on what people are going to do with it for our own and our user’s protection.
MA: What is your revenue model? Do you know yet? Have you thought about it?
EW: We’ve though about it. We had to do some thinking about that to raise a bunch of money, but it’s not actively in development right now. The broad strokes on the matter are obviously Twitter is being used for a lot of commercial purposes right now, in addition to social purposes. We think that works pretty well. We think there’s a lot of companies that we’ve talked to that seem to be getting a lot of value out of it. If that continues, if that becomes a rich world for users and the companies, we think we can extract some revenue from that.
MA: It might be difficult to define commercial activity versus not, right?
EW: It might in some cases, but in a lot of cases it will be really clear. I mean Woot.com is selling stuff. So maybe we just say, this is commercial usage and you need to pay for that and maybe there’s some features you get on top of that, that wouldn’t be as meaningful to personal users. There’s other cases, like yours, is that commercial or personal?
MA: I think clearly not.
EW: Yeah…
MA: …in my case.
EW: I’m not too worried about that, because there aren’t too many TechCrunchs in the world.
MA: If you were to do that model, what would make sense (I think) is to charge based on number of users following.
EW: Yup.
MA: Is that the logical thing to do?
EW: That’s the way I’ve thought about it. As we start exposing the data in different ways and followers (search results and stuff), it might get a little bit more complex, but also there’s more opportunity. So, one scenario that’s kind of interesting and I’m sure you’ve noticed, [there’s a lot of] questions and answers over Twitter. You combine instant answers and mobility and the fact that people are traveling around me asking a question. Like, “Where should I have dinner in the part of San Francisco?” Or, there are companies now that are sending, they’re little cafes that are sending their lunch specials for today. If you can make that querriable or just allow users to opt-in to receive that kind of stuff, then…
MA: Opt-in advertising slash/offers…
EW: Yeah, yeah.
MA: Do you think we’ll ever see advertisers being placed on the Twitter page as we see with the Japanese version of the site?
EW: I can’t rule it out, just doing more traditional web ads. They’re not that interesting to us. Something probably a little bit better would be the Facebook kind of “in the stream” thing, like Pownce does as well.
MA: Does that mean you might be texting people ads at some point?
EW: I can’t rule that out, especially internationally.
MA: I know we’re just speculating, but [what about] offset versus a premium version of the service where people pay to get rid of that.
EW: It could be. I don’t want people to go off speculating or saying this is their model, but since we don’t know our model I can’t say we absolutely won’t do this. Definitely, what we vastly prefer and think we can achieve is a win-win scenario where it’s like I’m opting in to receive this information and a high enough percentage of it is commercial and the companies are getting enough benefit out of it that it works for everybody, rather than a cost to using the system.
MA: So Woot, for instance, (obviously if you follow Woot) you could charge them just for that.
EW: Yeah.
MA: And it could also be a lead generation for them with users who’ve opted in and say they want (we’re using Woot randomly here) Amazon or Woot offers or something like that. You could also tax that or…
EW: Yeah.
MA: Ok.
EW: The other thing that will certainly come out with the Summize news is putting ads in search results is a pretty well understood monetization strategy where ads can work. I don’t suspect ads for Twitter results would work like web search ads, but we’re definitely going to experiment with that and see if that makes sense.
MA: So, last question since something is going on behind me. The last question is on the issue that’s come up with how much you moderate data and when you sensor and when you don’t. Have you guys come up with the final plan on that? It seems like it’s a little unclear where you want to go there.
EW: Which type of stuff are you talking about?
MA: Well first of all, if trademark infringements or personal brand infringements [come up] you’ve always shut that down when people have requested it. Usually that’s with account names, but recently I think Biz was talking on your end about the issue with what might be called offensive material on the site. I believe you took it down in this case, or did you?
EW: It depends on which one. I mean we have taken some down.
MA: How do you define…it seems like you guys are still working through that policy.
EW: There was a big issue awhile back where a lot of people were attacking us. We were being accused of not taking some stuff down. We looked at it, we didn’t think it crossed a line and we didn’t take it down. The user that was being upset by that thought it was wrong and that we should be doing something about it. That’s really tough, I mean we don’t have any black and white answers for it.
MA: It’s tough when it’s not black and white.
EW: Exactly. We’ve taken some stuff down. There are some clear lines where there are personal threats that are specific and some forms of speech that are…
MA: Borderline criminal.
EW: Yeah, but we don’t have a policy against calling someone a name or saying they’re stupid. Oh, personal data, if addresses and telephone numbers are not public information we’ll definitely take them down, but with insults and fights a) we don’t want to get into that and b) it’s neither feasible nor right necessarily to try to create a communication mechanism or system can’t have disagreements. So we take a liberal stance in general on that, maybe more liberal that other “communities” but we’re still figuring it out I think.
MA: Ok. Thanks very much for your time.
EW: Sure.