John Marshall
The following is the transcript of an interview of John Marshall, CEO of Clicktracks. A Netscape alumnus who is responsible for two other successful software companies, John Marshall brings 20 years of experience in the software industry to ClickTracks.
It was while running a previous company that John first understood the benefits of good quality web analytics, and yet saw the poor quality of the available tools. The experience drove John to found ClickTracks with a vision of high-quality data presented clearly and visually. John now regularly speaks on the topic of web analytics and marketing.
Interview Transcript
Eric Enge: Let’s start with something introductory about you, which is how you got into analytics and how ClickTracks got started.
John Marshall: OK, I was running a software company. I am a software industry guy, and I have more grey hair from it than I probably should. So, I have always been in the software business because it is my passion. I have been a software entrepreneur for a long time as well. And, I am a very product focused kind of person. I am very interested in what is the business problem that the software tool is attempting to solve. So, I was running a software company, marketing a product online, and I am doing a lot of online marketing myself. I went out and bought a web analytics products. In those days, it wasn’t really analytics; it was really a bunch of stats.
I bought the stats package that everybody else bought at that time. This was in ’99 and we thought that it would be the answer to our dreams. And of course, it wasn’t the answer to our dreams. There were many, many shortcomings in it I looked at that and said, well you know, the problem here is this product that we have bought just really isn’t up to the job. For a long time, I thought that the fundamental problem was my own ability to understand the reports on the tool. After spending a long time on this I realized that I didn’t really understand this data, it isn’t getting me to where I wanted to be, which was to advance my business.
I originally thought that it was my shortcoming that I wasn’t able to grasp it. Then I realized that no, no, no it has nothing to do with me, the tool itself is not optimal; it’s not well designed and not well implemented. Because of my experience on building software I could look beyond the tool itself and think ‘how would I solve this?’. So while that company itself was not successful, I kept the idea for a better web analytics tool in the back of my mind. I also was a big fan of Edward Tufte, who provided me with some inspiration.
I knew that the tool I had bought was poorly implemented, and the reports were hard to understand. It didn’t tell me what I needed to know. In course of realizing all of that frustration that I had, suddenly it dawned on me that if I am feeling that way, there must be many other people who feel that way, therefore there is a market for a product which does a better job. A better job at representing the results of the data, implementation, overall complexity, usability, a whole bunch of things like that. I just thought there has got to be an easier way here. I decided to go and create that product and create that company. So, that is how I got into analytics and then also by extension how ClickTracks the product itself came about.
Eric Enge: How did you get ClickTracks off the ground?
John Marshall: I funded it myself. I so firmly believed in the idea, and our ability to execute on the idea that I just built it. I had prior experience with raising venture capital money for a company and decided that I wasn’t going to do that. There are a bunch of advantages to raising money, and there are a bunch of disadvantages. The key advantage was that it gave us a lot more flexibility to just do the right thing for the products, and for the customers and so on. We didn’t need to perform on this or that in order to meet this objective over here that someone else set for us. We weren’t distracted with that sort of interference.
Not all VCs are like that., but almost inevitably they are like that to an extent. You really just can’t avoid it. Because we didn’t have all those distractions, we were able to focus solely on building the product. We also decided to launch our product at a sort of a mid-market price. We were very interested in the mid-market because we thought that was being grossly under-represented.
Most web businesses are really mid-market customers. We were able to design a product exactly for them. We felt that if we had VCs involved that they would have said: “no, no you got to go after the enterprise”. “You know you got to hit those home runs”. And we just didn’t want to do that. So, I think that we scored quite well by being able to take that approach. But, it meant that we got there a little bit slower. However, there were some things that we could have done faster if we had more funding earlier on.
Eric Enge: Right, so you mentioned earlier Edward Tufte. Can you clarify for the readers of this interview about who he is?
John Marshall: Edward Tufte is a professor at the IAO. His last name is rather unusual and is spelled T-U-F-T-E. So his website is http://www.edwardtufte.com.
He is a guru on data presentation, and visualization, and data persuasiveness. In the end, data needs to be used to persuade somebody else of the something, and he is extremely knowledgeable of how you go about building data that does that. And there are all kinds of subtleties in this, but you know probably one of the classic examples is Pie Charts. Pie Charts are grossly overused in data presentation, and they often serve only to obscure the facts, not to clarify the facts. And, anything you read about Edward Tufte will confirm that type of thinking.
Eric Enge: So, you mentioned before that you were targeting the mid-market with ClickTracks. Can you describe how you would classify the market overall, and how would you label those different segments?
John Marshall: There is free, mid-market, and enterprise. Those are probably the three divisions that exist. We particularly like the mid-market. Because most websites are mid-market customers. These are companies that will probably have less than five hundred employees, and they probably have a rather simple IT infrastructure. They are not using an Oracle data warehouse over at the data center, for example.
We particularly like that customer, because the web analytics data has such an immediate impact on what their business does. It is just like a direct feed into the bottom line of the company or the top line actually. Our products are designed for very quick implementation and a very quick understanding of reports. We very consciously try to avoid overcomplicating things.
There are a whole lot of design decisions that we make and they really reflect the fact that the mid-market customers represent the majority of websites. There are companies which are small enough that they can move fast, so that’s really the beauty of web analytics data. We give you data and in theory, you are supposed to take that data, act on it and do something about it. And, the mid-market company can actually do that.
So, the success rate of the products is very high. In the enterprise space, all the data in the world can’t change the slow decision making that then happens. And, you know we have very high pride of ownership in what we have built, and we just don’t like to see shelfware. We just don’t like to see products sitting on the shelf which don’t get used, and larger companies I think are more prone to do that because they just simply can’t absorb the data that we have given them
Eric Enge: How would you characterize the unique value propositions of ClickTracks?
John Marshall: I think there are three categories. There is implementation and implementation time, and there is flexibility of the data that you are going to get, and which is somewhat connected, and there is an ability to understand the data. These are somewhat interconnected with each other. We have made a product, which does less. When you have too many features or reports, it just doesn’t work in the real world. So, we don’t implement those features. And sometimes, we lose a sale, because we don’t do path analysis.
Path analysis is something that has been out there for a long time. It is kind of surprising that ClickTracks product doesn’t do path analysis, but we feel that this report as traditionally implemented doesn’t work, so it’s not implemented in CT. Overall the product does less than other products do, and because of that, we can make each individual feature more focused on solving the business problem that the customer has.
For example, a stripped down reporting system might be one way to look at that. We strongly believe that reporting is, in general, too complicated for the customer in most web analytics products. So, our reporting is made to try to solve that. And, we certainly use some visualization attributes that help us get that. We set very high standards for ourselves in making the product easy to use.
Implementation is sort of a dirty secret in the world of web analytics. The customer often just doesn’t know how ugly the implementation is going to get. And, that is because of the salesperson just tells the customer that all you need to do is implement custom Javascript on your pages, and you are off to the races.
So, the salesperson is able to get the order knowing that the ability to actually collect the data is completely punted on, and the customers are going to have to solve it. And, the customers almost really can’t argue with it, because the salesperson did tell them that they were going to need to populate these JavaScript variables all over the place and deal with whatever nonsense in their consent management system to get it to work.
Eric Enge: When you have a tens of thousands, or hundreds of thousands, of pages that becomes a messy proposition.
John Marshall: That’s right. So, our product is designed so that for most websites, every page has the same JavaScript, if you are doing JavaScript data collection. Now, we also like doing log file data collection, because it completely removes this problem from the equation, and can easily give data on historical dates, even though using log files is a little out of fashion. There are many benefits to that approach. But even then if the customer is going to do JavaScript data collection and they prefer that for multiple reasons, we make the promise that the implementation will be easy. We meet that promise, I think a little less than I would really like, but I think our implementation is a lot easier than other products 95% of the time.
Eric Enge: So, you mentioned the JavaScript thing, I remember you did a presentation at SES Chicago. And I believe, if I remember what you said, is that in terms of contributing to errors, erroneous JavaScript implementation is about a 5% factor, is it that, do I remember that right?
John Marshall: I think so. I must admit that the number in that presentation was just based on anecdotal evidence and my experience of doing this kind of stuff. But, yeah I think that incorrectly implemented JavaScript is a major contributor to problems.
Eric Enge: But, if you had the patience in getting the JavaScript implementation right, you do get a slight accuracy gain with Javascript over log file collection. But how much it matters is a whole another question.
John Marshall: Yeah, exactly. We often are asked by our customers which approach do you recommend and we generally say, if you are worried about the accuracy of this kind of stuff then you know you are really asking the wrong question. It’s the standard stuff that people have been experiencing. I think really the key point here is that the JavaScript’s implementation is so complex that people just kind of underestimate what a complete rat hole that’s going to end up being. Because of the way the world works, it is impossible for the customer to be sufficiently aware until they get into the implementation. And, they can’t get into the implementation until they have bought the product, by which point it’s too late.
So the website development process for a customer is kind of like a treadmill, and it runs at a certain speed, and the web analytics product has go to run on the treadmill at the same pace because otherwise it just falls off. So even once you have got the implementation done because you have got that brittle connection, with the web analytics implementation, has to be considered part of every website development project, even when you are making fairly simple changes. Inevitably what happens is the customer changes something on the website, it breaks the JavaScript, and the customer goes to their web analytics vendor saying “hey you guys stink, this is completely broken again”.
The web analytics vendor has a fire drill, they fix it and it’s just this constant round and round battle. That process forces the web analytics vendor to charge a lot more for the products because they have got to have more money in order to deal with this constant round and round problem. We are trying to avoid that, by not having stuff that is that complicated in the first place. We just feel like it is not valuable enough for the mid-market, so the benefit of all that is that the product is significantly lower cost.
We are trying to deliver to the market an iPod (Editor: i.e. the iPod equivalent for the Analytics market), that’s really what we are trying to deliver. An iPod is remarkable for the features that it doesn’t have. If you look at it compared to another MP3 player it has fewer buttons, it has fewer features, it just does less. But, the elegance there is very strong, and it does things in a little bit of a strange way, it’s got a hard disc in it, for example, that’s a bit rare for a device like this, but there are so many other benefits that you get from that.
So, we kind of take this similar approach, we are not attempting to ship you the cheapest possible MP3 player. We are not attempting to create an extremely high end and high fidelity experience, we are not attempting to be Bang and Olufsson. We are shipping a good quality product into the mid-market.
Eric: So, one thing I have seen that ClickTracks does is uses Perl compatible regular expressions to allow you more power in categorizing groups of pages or things that you are trying to analyze. Whereas other products in the market would require you to tag those pages with unique Javascript, sometimes just labels or things like that, but the labels that need to be in their Javascript are the kinds of things that need to be maintained. When sites are redesigned or changed, this is the sort of stuff that’s breaking.
John Marshall: Yeah, that’s exactly it. Our intent is to collect all the data at cleanly, once, and store it in this big bunch of binary files, and we then go back and apply our code against that data, and come up with numbers. Suppose you want to build a funnel. Building that funnel ends up being just very, very complicated for many people.
We feel like, if you need to define a funnel by passing a variable from the JavaScript, that’s just completely broken, because no marketer can possibly come up with all the pages and get all the JavaScript in there, and get it absolutely definitely correct, before they want to start collecting data, it just doesn’t happen. You got to separate out the process of collecting data from the process of processing that data. We strongly believe that the customer of our product is a marketer, and marketers are very used to working through things iteratively.
It’s very difficult to just know for them to know in advance exactly what they are looking for. Marketers rely on, here is what I have done, what worked or didn’t work, what are we going to improve next time, etc. So, it is very difficult to say, in advance, where is the funnel I want for my website. They start to build it, they look at the data, and say “that’s not right; there is this page over here that we didn’t think of, that’s really part of the process, isn’t it, okay let’s include that”.
Eric Enge: Right. So let me switch the course here a little bit and talk about some metrics and implementation details. How does ClickTracks count visitors currently?
John Marshall: It’s pretty standard stuff; you know there is sessionization against either the IP address, or end-user agent, or if there is a cookie, we would use the cookie to define the session. There are certain things that ClickTracks does, that are a little bit biased toward search engine marketing. For example, when somebody comes in from a PPC campaign, we will start a new session, always, even if the same visitor was in the site twenty seconds earlier looking at other pages. We just say, okay, if he clicks on PPC ad again, it’s defined by us as a unique session. Beyond that I mean it is all just pretty standard stuff.
Eric Enge: And, how does ClickTracks count page views?
John Marshall: Yeah, ClickTracks has a little bit of bias away from page views, and somewhat more towards how many people see this page. So, the default reporting ClickTracks before you reconfigure things is not how many page views does this page get. The default report is how many people see this page at least once. That is different compared to other analytics vendors, who are actually measuring slightly different things. With page views, if an individual visitor sees page A, goes to page B, goes to page C, and then goes to page A again; that’s two page views on page A.
ClickTracks’ point of view is going to be that there was one person seeing page A. We de-duped the second view. If you are a media company selling ad banners and that’s what you want to measure on your website, that’s the wrong thing to do. If you are selling impressions of ad banners, you want to know how many page views you got because that’s your revenue source. So, there is a configuration option in ClicksTracks’ that provides a way to get that, but most people don’t do that. The reason is the whole ad banner thing has gone away.
What people care about much more is, which page on my site is effective; and which pages on my site do people look at. When somebody looks at page A, and then goes to B and C, and then looks at the page A again, their brain says, “hey I have already seen this page”. You are really only using page A as a jumping off point, you don’t actually want to read all of this same stuff again. You have already looked at it, and you are most likely using it as jumping off point to go somewhere else.
Eric Enge: What do you do if the cookies are rejected by the user?
John Marshall: default back to the user agent and the IP address.
Eric Enge: Okay, so you just use that as the tracking of whether it’s a unique user or a new session?
John Marshall: I am sorry, there are two different things here. You are asking a unique user or new session. If you reject cookies, we do not attempt to determine if you are a new unique visitor. We just can’t perform that calculation; because we can’t unless you have got a cookie. The data without a cookie becomes very questionable for unique visitors, so you just kind have to bypass it.
Eric Enge: So, you don’t know it’s a unique visitor when they come in?
John Marshall: Yes, that’s right. For sessions, it’s not nearly so bad. You can use IP address, or user agent to establish a session and reestablish a session and it far works a lot better.
Eric Enge: So what percentage of users you figure reject cookies?
John Marshall: It depends very much on the demographic, unfortunately. If you are a Christian bookstore it’s probably small; and if you are an online gambling site, it’s probably much higher. So, somewhere between 5% and 40% are the two extremes.
Eric Enge: Right.
John Marshall: And it also depends of course on what interval you are measuring it across, right. If you are measuring cookie rejection across a month, you are going to get one number, and if you are measuring it across the week, it’s going to be a lot lower.
Eric Enge: What do you do if JavaScript is not enabled?
John Marshall: If we are using JavaScript, but it’s not enabled for the user, we do collect some data, but we don’t insert that into the data that gets analyzed, so the data collection engine knows about it, but it doesn’t actually attempt to collect that user data. So in fact, those users become opaque, they just go away.
However, turning off JavaScript is rare.
Eric Enge: 1-2% of users or so?
John Marshall: Yes, so much of the web just doesn’t work if you turn off JavaScript. I mean our CTO is the kind of guy who turns off ad banners, he has the special little proxy server that he doesn’t accept persistent cookies, he has filtering software so that flash doesn’t come in, and ad banners are turned off. I don’t mean any of that in a critical way. This is just how he chooses to drive his internet. However, JavaScript on the browser’s turned on, because the web is just too hard to use without it.
Eric Enge: That is your thinking about solutions that rely on data warehousing as an approach versus alternative approaches?
John Marshall: I am skeptical about data warehousing as a solution here because data warehousing has a history of failure. Data warehousing also doesn’t do a good job of applying statistical sampling techniques to the data, because databases are not architected around that type of idea. In the database, everything is known perfectly, or not at all, which is a production line is a right approach, but data warehousing tends to want to tie together absolutely everything, and it becomes very slow and cumbersome, because of that. So, ClickTrack’s product has a custom data store which stores everything from every user across all time. But that’s not the same thing as a relational database and trying to get an RDBMS engine to store it for you.
With data warehousing what the customer often really wants to do is combine their online and offline data. That turns out to be very complex, because there is so much online data, and so much of it is completely useless, that you really just don’t want to take all of your online data, and throw it into a data warehouse. You just get major breakage if you try and do that. So, if you somehow get a subset of the data from your online tool into some kind of data warehouse, it still doesn’t work.
Data warehouses are set upon the assumption that it’s all loaded in there and it will sort itself out. You spend a hundred thousand dollars on this data warehouse, it is going to be perfect, and it is just not. I think that until companies get a bit more comfortable with doing things like a statistical sampling of the data, that before it goes into the warehouse, there is not going to be much progress on that front.
Eric Enge: Can you take the offline data and bring it into your online storage environment?
John Marshall: You could, but the online data storage environment is not set up very well to provide the kind of queries that people are used to having. You know certainly the volume of online data is a major problem, and that’s why bringing it into the offline world is tough. Taking it the other way it is possible, but I don’t know anybody who does that. Companies do get a bit leery of data privacy, you know if you are buying an online service. They get apprehensive about who owns my customer list.
The key to online data in the ClickTracks repository is that it’s much more efficient because it doesn’t use traditional data warehouse approaches. It doesn’t have a row in table type of mentality, because it is just too slow.
Eric Enge: What are the key challenges do you think are facing the analytics industry in 2007?
John Marshall: I think customers are tired of looking at reports. Reports don’t really tell you what you need to know. They are looking for insight and insight is hard to get unless you apply a human being to the problem. And, I think more solutions to the online marketing problem need to be delivered less by vendors. The market for analytics is pretty healthy. I don’t think any vendor is going to go away. I don’t think you are going to see consolidation.
But, what customers do want to do is spend money on solving their online marketing problems. But, they don’t what to spend money on buying analytics only solution. So, products that do more than just analytics are important, and I think it is difficult to deliver that type of solution to customers through partnerships.
Finding good web analytics is a problem, but it isn’t the only problem that we have got. We also need to worry about click fraud, I need to worry about the email marketing, and I need to worry about what’s going on with the new content that was pushed on the website.
Eric Enge: Right. So, to some degree, the big thing is how do you get the information out that they really need with the minimum amount of expense and effort?
John Marshall: Yeah.
Eric Enge: What would you say to a webmaster who asked you “how should I go about making a decision on an analytics vendor?”
John Marshall: I would look at the complexity of implementation, and usability of the product. I think those are the two weak points, and I assume that whatever you think you want to know about your website today, such as the kinds of reports that you think you want, you are probably wrong. You probably need to first understand what you don’t know about your website, and then attempt to work that out from the web analytics.
Web analytics are very rarely a process of let’s have these custom reports. It’s a constant iterative process of examining different data points, and the successful company is one that starts out the process of understanding that. The company that isn’t successful is the one that goes to the web oversite committee, and works on the RFP for what the products need to do, and puts together a Chinese laundry list.
Eric Enge: Great! I appreciate your taking time for the interview.
John Marshall: Thank you.
