newmobilepro.info

Jack Jia: Sure. Sure Sean, thanks. The Baynote is a three years old company and we have shipping products for the last year and a half. We have about 85-87 customers today. What we do is to help people find content and product quickly on a website. We do two things, one we call recommendations people like this and also like that but very different than what you see in a typical Amazon recommendation. We do not use purchase for example, we do not use clicks as what is useful.

We also do social search for site search. Giving really what has made Google search very effective. The page rank really has made Google successful and we do that for a site search. We use a technology we developed called used rank, to re-rank a typical keyword search result and then make it much more relevant based on the what appears behind the scene on the site.

Jack Jia: Sure. We actually worked with a large variety of different kind of customers. There are three sort of a general bucket we belong to E-Commerce customers so people are selling stuff on the web. We worked with a lot of long tail website, long tail stores for examples USappliance.com is the major appliance website. They do about $100 million a year online selling washer and dryers and refrigerator, except they carry 8,000 different brands and kinds of the same thing. We also worked with a company called eBooks one of the largest, long tail bookstores and probably $180 million revenue on 22 million titles. Amazon only carries 5 million titles. They carry many times more.

We also worked with media companies, publishers and websites. One recent – really rising star is something called education.com. They really serve K to 12 kind of grade students and the parents and everything they need from study to learning to actual health issues and behavior issues. It is all in one kind of place to get those kind of information, very, very deep, very, long tail information. But we also worked with a lot of enterprise customers and we have Cisco, we have Juniper Networks, Intuit, TurboTax. We help them in three different ways. We help the marketing site, www site. We also help the support, people are looking for information, technical information and also we help people have Internet need.

Jack Jia: Yes, they are actually all the same in some sense. What we do is we track on the user behavior on the site. It is really important to track sort of implicit behaviors of the users. We do not ask them questions. Let them do whatever they are doing if the people on Cisco site and people looking for routers and looking for switches and some people are actually looking for jobs.

Other are probably looking for investor related information. Different kind of people have different behavior. Some people are successfully finding information and others are failing. Those are all clues. What is useful and under what context? The key is to learn and build that collective wisdom of like-minded peer groups. The routers, you may have a 1,000 people you stay come in, find good information and build that knowledge and make sure that knowledge does not go away when that person walks out of that site and then you can accumulate that to collective wisdom overtime. You become smart and then people start to really guide each other without them even actively participating it.

Jack Jia: The wisdom of a crowd is really the insight. Wisdom is what we call the wisdom of the invisible crowd because the people who are not telling you anything.

Sean Ammirati: The wisdom of the invisible crowd. You do that at cross sessions or just point in time like – how does that work exactly?

Jack Jia: Yes, the different information has different ways. Some sites that they actually drop fresh party cookies and position cookies and then you can have cross sessions and others do not use cookies they just have session based behavior. We actually do not – one of the big thing is although we track behaviors we let the behavior guide us where they put the information is but fundamentally we do not use individual behavior unto actually target consumer product. It is not like – there is a lot of concern on privacy and that actually is now the main reason why we do not do it.

We actually have done a lot of social science study and mostly at Stanford and really found individual behavior and even to certain extent personalization do not really help on targeting and do not help you to figure out what people want. It is not a good predictor therefore we do not actually care, the session by session, cross session or not and we just use the behavior to guide us what is useable under what context.

Jack Jia: Sure, sure. There is a kind of different way of targeting content and product for people. There is a whole theory of behavior targeting means “behavior concept”. There is a different definition on that but the more common definition of behavior targeting says I need to track individual behavior, so what my last six visits for example can tell me what I may need in the future, today and tomorrow, etc. Maybe my profile, my demographic information, my age, my gender, time of the day and all of those can somehow predict me.

In fact, the more the system knows me the better they can personalize my needs and target content. That one school of thought. We actually do not subscribe to that. We do not believe understanding someone fairly or inside out can actually improved our recommendation or targeting that much better. The reason is pretty simple actually we have way too many profiles and we have way too many different characteristics. The more you know me, so on surface I am – is where I am talking as a context of CEO.

You think, okay you can talked to me CEO related information but also I like many other things. I like wine and I travel. I like sports. I am a father. I am a son. I am a brother and the more you know me, the more confuse the engine will be, the recommendation will be. Actually what the hell I am going to recommend you. Given also our patience is very short, the user patience is very,very short and three recommendation if not on target, I am gone. By profiling people and personalize people may not be the best vehicle to get you what you need.

Jack Jia: There are hundred of moods in a given day or week. It is very hard and people switch mood within five minute range. I will switch to maybe internal staff meetings soon.

Jack Jia: The opposite of that however, there is a solution and that is not something that is necessary proprietary to Baynote but it is sort of a social scientist and the brain science have told us people are animals of context. We are not so unique after all. Personalize to me does not really help because people under the same context, 95-98% of people where actually need the same thing and that is really a fundamentally determining by our brain structures. Our intelligence sort of information and so therefore understanding can understand the context, you can actually get right most of the time.

Jack Jia: Behavior helps to define what is useful under a context. There are two different things. Our behavior helps to kind of taking peoples action at the clue what is useful instead of asking them. Action speaks louder than words. We do not ask them, use behavior to approximate what they truly need. Right, that is kind of one. Then based on that the new people come in. We do not need to care where they have been. What is there historical interest and all we need to know is your current context and then we can predict all the other people with that same context whether they need it.

Sean Ammirati: I guess what is part of a person’s current context and what is not. It is the page they came right before us is that part of their current context?

Sean Ammirati: But sometimes it is and sometimes it is not?

Jack Jia: Yes, if I am on Best Buy and we are looking at TVs and I will send them down with TV I am going to cameras, the context has completely switched. Right. Just make a general assumption that the TV and the camera has no overlap of common interest and therefore my TV behavior may not guide me what I need on the camera set.

Sean Ammirati: Sure. If you are looking at four cameras, the camera you looked at before might help the context of the next camera you look at?

Jack Jia: Absolutely.

Sean Ammirati: Okay. That is very consistent with how I would view the market as well. I guess one thing that I am curious about though is that it seems like there are times though. Just to make sure I am understanding the statement you made. You are not saying then really that behavioral targeting does not work so much as you are saying behavioral targeting by itself does not work.

Jack Jia: I guess it depends on the definition. The current definition – at least most of the definition behavior targeting or personalization focus on understand this individual and their profile and their buying behavioral, even their wealth, all of that. It is not really a good indication of what this person needs next. We actually said throw that information out, do not use it. That is kind of maybe the difference.

Jack Jia: Sure. It is just basically what we are saying is that this recommendation – the system you wanted to deploy do not try to personalized people, do not try to track this people with their histories and their buying power and got to know them. Right. It is a notion of personalization. I want one to one. I know you so well, so I would know what you want because I have been tracking your entire life. There is a way in tracking you. We do not need to do that. A stranger comes in without knowing, this persons name, without knowing their interest in the past. We can still predict what this person need based on the context they are in.

Sean Ammirati: Sometimes. Sure, but think about like a political – there are times where the context would not be that helpful either, correct? I mean, surely our pages where context is not that helpful.

Jack Jia: There is always exception but generally speaking that is the case. Give me an example that you think it is not helpful?

Sean Ammirati: That context would not be helpful?

Jack Jia: Yes.

Jack Jia: Yes.

Sean Ammirati: That is the place where behavior probably will render a more effect to that than context.

Jack Jia: You can argue both ways but in terms of - even that in that context that might give you probably more worse context sort of example where you know someone, you reverse you do not know someone. Google ad word or ad sense does not know the person. They do not care who you are searching. The ad they are presenting is strictly based on the words you type in. That is sort of defines the context. The ad effectiveness is very high we know about because it will make a lot of money.

Sean Ammirati: From search clear areas you are talking about.

Unfortunately, the ads that fly by, you know those banner ads, there is probably 58,000 to 100,000 have gone by in the last 10 years. I have not use a single one. I did not even look at any single one of them. It is not because those content are not importantly to me, those ads. There BMW cars are actually fine by many times and I own three BMW cars but it just not at the time when they started surveying. It is just really out of context. All right. That is the key thing I am trying to – context is king. Without deciding the context. It does not matter how I am involved in that thing. You can profile me, it does not really matter.

Jack Jia: No, I did not pay attention.

Sean Ammirati: It was a classic example of work contextual advertising in my opinion falls down; they were ads to buy real estate in New Orleans, right. It is very unfortunate thing. I am not trying to make light of it.

Sean Ammirati: I think that is a good guess. Sure.

Jack Jia: No, we use behavior so the content is whatever it is, right that we do not care. The context we are basically saying – what we are trying to say is what is content is about?

Sean Ammirati: You do care about behavior. You care about the behavior of the community not the behavior exclusively of one individual.

Jack Jia: Right. We do not care about individual nor – what we are trying to use is the behavior to define what community call this. That is what we call context. I will give you a really example that US appliance that would shed some light.

Sean Ammirati: That is wonderful.

We observed what is the interest ultimately where they actually find value is that the cook top they find value so we connected those dots. Then that intent to find stove in the context. The behavior simply help to define what this thing is called and then under that stove context we do not care who comes in. Joe can come in. Mary can come in. We do not care who they are.

Jack Jia: Use rank.

Sean Ammirati: Use rank. I am sorry. Use rank.

Jack Jia: Use rank is really the foundation. It is really our engine in the sense we call it affinity engine. Basically, it serves both social search and recommendation.

Sean Ammirati: I see how that could have – how you could figure out that stove means range if people where entering keywords “stove” and then clicking on things about ranges. How do you do it though when you do not have a query?

Jack Jia: You do not have query, you have other behaviors. You have link text. You can borrow.

Sean Ammirati: In that case that is behavior, right. That would be page navigation or something like that.

Jack Jia: Yes, page navigation. Link text itself, right there are words in it. They are searches on Yahoo, MSN, Google that is the where they come from or they are going. There is all kind of things. We are getting to how we implement this. There is all kinds of behaviors that we actually pattern it about 24 some behavioral heuristics that could tell us – we call them the fingerprints of behavior.

Jack Jia: That is right. There are 24 other behaviors that will actually tell us collectively not just individual behaviors but the light minded pear behaviors and that is where we ultimately figure out what is useful and what is not.

Sean Ammirati: Cool. All right. This is very cool. Philosophically I think similar though we are doing it in a different problem space but it is very similar. Tell me the what the reaction, I mean I imagine then on the recommendation side, you guys signed up. Who would you do as your competitors? Whether than fill that in for you? Who would you do as your competitors?

Sean Ammirati: That would be basically ad servers. When you talked about competing with behavioral targeting companies, you are talking about competing with behavioral targeting ad servers.

Then there is the home-grown system. We compete probably whole lot more to the home-grown system then actually commercial vendors. Most of these commerce sites or media site have some kind of recommendation then they are based on purchase. People bought this also bought that. They sometime use click. People click on this also click on that. What we are basically to say – those are good start you are not even getting near the kind of revenue lift that you are getting. You could have got with like-minded peers the community driven sort of approach that is the way to be testing and then you figure out what are the difference are.

Jack Jia: In theory, we should be competing with them but we actually have varieties. I have been competing with them a couple times and we won most cases. The reason is again the approach is very different. They take a very more of a behavioral targeting kind of approach and we do not.

Sean Ammirati: Okay. Got it. Interesting. If you were to project out now, say 18-36 months in the market place and obviously you are the CEO trying to change that market place with your company. Where do you see the market place for this stuff in general going? Maybe stepping outside your role as CEO for a moment just talking about where you see the industry moving?

I think a year ago, there where very few people who talk about recommendation and especially use with crowds. For the first time, Forrester had a publication on this in December and start talking about – actually there is space. A few years out I think people will realize that recommendations is very, very important. Largely because the sites are getting larger and more content and more product, right.

We will see more and more sites start to go, taking advantage of this crowd sourcing and the best crowd is your invisible crowd. They do not subject to survey bias, you know, what the psychologist often tell us. The visible crowd, the social network is kind of crowd, the surveys, the reviews, the foreign blog, all suffers the survey bias. They kind of – what we call the three kinds of people that will give you feedback. You got few people too much time on their hands, too opinionated or someone who has ulterior motivation. Those are not good feedback.

When like man appears starts to – what they would do is push them into higher and higher price product because that is something they just not like but they love. They end up buying $500 bicycle instead of a $200 one. Much like the real world. I am a golfer for example. Every time I hang out with my golfer friends. I end up buying clubs that I did not think I need it but I end up in love with those things and I tell others friend then they do the same things. That is the power of community kind of recommendation.

Jack Jia: The smaller better but too small then you have got noise that you do not have the ‘wisdom of crowds’. There is a lot of noise going on. Too big is not good, not either. Like Cisco, like router club is actually mean we have real data. There is probably a million of people who care about its routers and switches and that is just too big of a crowd for you to truly drive people to niche information and niche need. You need to further divide that big router into subdivided into stuff community.

What they are trying to drive is use self-service of course. Every call, the company cost them a lot of money, $30 per call. They have good content on their site. Before they went live with Baynote, they are trying to improved their search and navigation, their average sort of search utilization radar was about 15% before they start to improve it. It was very low tech 15% they said that is too low but actually industry averages, which is actually too low if you compare with how many searches has to get use on this first site search.

Overnight it jump from – this is over a year ago by the way. They have been with us for a year now and jumped from 18% to 35%. The community started to come in. They started analyzing. Basically 3% of the query where really – the top 100 query only accounts for 3% of the total queries. There are so many different questions in terms and these are the microcommunity that we are talking about. They have different needs and different content.

Jack Jia: We are software as a service company. We provide technology as a service much like Salesforce or Omniture. We licensed our service to any website who needs our technology including actually some – for example nasa.gov uses us to actually recommend galaxies. People like this galaxy also like other galaxy and we do their search as well. We do star clouds, what kind of topics people care about. The homepage and the deep content as well.

Jack Jia: Yes, we have a flat fee model as well as pay for performance model. You can pay a fixed fee, depends on that. That is mostly for non-revenue generating site seeds based on your profit, they can be apprised. Or you can – for revenue generating you do not have to believe our word just try it and it is very easy, you can go live. The actual work when you go live is actually very minimal like the hours of technical work and then you can go live. Since they do AB testing and you build up, and once you see the lift then we have a bigger percentage of the actual lift.

Sean Ammirati: Interesting and just a little bit about your company. You said you have been around for three years. Or you venture backed?

Jack Jia: Yes. We are venture backed. We have done two rounds of financing. First round was kind of a from JMB Capital and the second round was actually unsolicited round so it came from Steamboat Venture which is a Disney group of venture arm. They invested in us as well about a year ago. We are talking about $15 millions all together.

Jack Jia: Yes. They have been great. Disney has been great. Absolutely, certainly the media space – they are a big player in that space. They are being extremely helpful. He was also being extremely taking the capital, being extremely helpful from opening up – channels and the company is doing really well. We are growing our revenue and just had a phenomenal raise.

Sean Ammirati: That is great. Jack I really appreciate you taking sometime to give our audience a little overview on your services and tell us a little bit about Baynote.

Jack Jia: Thank you very Sean.