Launching a new site

Well I had to talk about it at some point, so now is as good a time as any. Computer weekly has just had a major redesign (launched at the end of last week) and I am going to talk about how the redesign process worked and why we did it. Just to prove a point though - we aren't the only ones to have a redesign, the Guardian have just had a relaunch where they have changed the home pages one by one. Interestingly I was intrigued by the blog on the Media Guardian where the users gave their impressions of the new site. It's unlikely that those that liked the site or thought it was ok would have mentioned it on the blog though, so I'll give them the benefit of the doubt. I'll continue with media theme by pointing out that the Sun also changed their site recently - mainly positioning it around the MySun section which has been doing well (apparently) , even if it is still impossible to find any of their content on Google.

Anyway - lets go back to Computer Weekly, because I know something about their redesign, whereas I don't know anything about the others. It all started a long time ago, actually, where we got a group of 'experts' to sit down in a room and attempted to come up with some ideas of what they could do to improve. This process involved not only looking at what our users were doing (using web analytics tools) but going out and asking them (user research). We also went into the depth of the content and categorised it in a more effective way, worked out what our key search terms were and built a content creation concept that would allow the content on the site to be found. Also we wanted to do the whole thing in an agile way.


Ok - I jest, but seriously agile is a system where all parts of the project were broken down into small steps which we could then complete in individual 'sprints'. Although our whole UCD approach maybe went in a slightly different direction, but the theory stands (even if we didn't release it all at the same time).

Well the first step in this process was to create some baselines and work out where you need the most focus. Well for Computer Weekly this whole process was much easier than anticipated - they had a broad understanding of the site and how it all worked in the first place. With media sites it is very important to look at the first touch points of the user and then work out where they go from there (or not). Whilst we have known for a long time that homepages are not the be all and end all it's amazing how when you delve down into the stats you realise how little it is. But that doesn't mean you can ignore it, by the way.

So the next step in the process is to do some user research. This is important. If you don't know what your users think of your site, then you won't have anything to baseline it against. The next step of user research is to create personas. This isn't as scary as it sounds - you are creating a small group of users to your site that represent them all. Your personas should tell anyone reading them about a group of individuals that look at the site, why they are there, what they like, how web savvy they are and what we want them to do on our site. This last bit is important - and should come out of the concept model for the site.

Whilst this process is going on, you have your information architect look at the site and work out the best way of categorising the content on the site together. This is difficult for Computer Weekly because there is a lot of it and you may already have some preconceptions based on the current site. So this is also helped by use of the concept model - if you know what sort of content you are going to create in the future, it is far easier to create the categories. Working these into a set of rules to set this up automatically is the difficult bit.

Then we need to bring all this back together and build some wireframes. These are critical because it means that you are not wasting lots of time and effort (read: Money) building real alternatives as pages for you to test with users. And this is the next step, test them with the users - because if you want to know which option is best, then the only way of checking is to ask some users.

This whole process of building, testing with users, building a bit more, testing with users again, ad infinitum means that you end up with a website that is truly user centred. Not only is it user centred, but it is now also garnished with a lot of links that are about the same subject, giving you a nice 'aboutness'. Given some good SEO work (technically and looking at those search terms that are driving traffic), you will also end up with an audience who are looking to engage more with your website. Which is probably a good place to start next time.

Your tables aren't long enough

Like King Arthur and his gang at a kids tea party, I have been left to rue the size of tables recently. Whilst Arthur may have got out of this predicament by getting Merlin to do his magic, I only have the HBX support guys (although they can be just as magic when they want to be). "But this just sounds like you are over egging a contrived gag." I hear you cry. Well you are right, to a point, but there is something a bit more serious behind it all.

In HBX (as in other web analytics tools) suffers from a problem where by they have to limit their table lengths to speed up responses and processing. This means when you set up your HBX account and you discuss how many pages there are going to be, you have to be very careful. Most of the time, this information is ok - for a Commerce website, the likelihood is that you won't have that many pages. For a content based website this is more of an issue.

Ok - lets put it all into perspective - you could probably go for years without having to worry about this sort of thing (some of our websites have done). In fact, if you don't read this post then you could go into the next 18 months before you have to change to site catalyst without realising that your tables aren't long enough. There are far more important things that you could be getting on with in HBX without noticing that your 1001 top landing page suddenly doesn't have any data associated with it.

But let me explain why you should get them increased (apart from the fact that you can) with a couple of examples. Electronics Weekly web Editor Alun Williams wrote about Hitbox in his blog. This story was the 2500th most visited page on Electronics weekly in the month of October. That doesn't sound a lot but Alun works out how well his blog is doing by looking in the content Hierarchy of HBX and these page views add a vital couple to his blog when he's working out where he should be spending his and his teams time.

However if he is trying to work out if he should spend more time building links into his blog posts then he will need to look in his Entry pages report and work out how many of those visits to his blog pages arrived at the individual entry pages. However we've just discovered that our page is only the 2500th most visited page and turns out to be lower in terms of entries. This means that if we only look at our top 1000 entry pages (which is what our account was set at) we don't get to include that page. Suddenly this long tail is looking very short.

Then we have to think about what our users do on the page having arrived at the blog posts. Now if Alun is trying to work out whether he needs to write more posts (get more visits), build more links (get more of those entries) or whether he needs to focus on getting those links built internally to keep them on the blog (or the site) then he needs to make sure that his bounce rate is low. How does he do this in HBX? Well he adds up all the entries to his blog posts (if we can get them right), then he adds up the total number of single access pages in his blog posts and he divides one by the other.

Well we've just proved that we don't really know the number of entries. The number of single access will depend on how long that table is too. If it too is cut off then you might suddenly discover that some of your pages had entries and no single access. They all stayed. Really? Do we really believe that? Well no - because some will go away straight away, but they will have fallen off the bottom of the table. So we're making the best performing pages, even better performing. And our not quite so well performing will appear worse because the better ones will appear better.

Ok so here we have a couple of reports that we need to get longer: Entry pages, Single Access pages, (we might as well do all the other page reports whilst we are at it and do page views, visits, uniques, exits, time on page, etc). But we also need to think about other reports that are going to be too long. Referring urls is one that should be increased - this is used for the referring url report for each page. The search term and search engine report also needs to be the right length (otherwise your long tail, etc, etc and so on).

How do you do it? Log into support and ask them to increase the lengths.

Bidding for paid keywords on your own brand

As I was writing a presentation for next week on paid keyword tracking, I was reminded of something that had happened to me in a previous job. Particularly this was bidding for keywords that included the brand name.

Now for content based websites, this shouldn't be an issue. You should never need to bid on your own brand to get traffic into the website (the exception being when you are trying to get users to sign up for something - eg to buy a print version of the product, merchandising around it or a special offer). Remember that you control the text that the user sees below your website name and above your domain name - so if you do want to give your users something unusual to think about, then this can be a good option.

For e-commerce websites, this maybe slightly different, although I agree with a couple of Dave Davis' points on this subject, I'm not quite as convinced by it all, however both myself and Dave are going based entirely on what we think is the best option, rather than what we know is the best option. How can we know whether it is good to bid on your brand (or company name)?

Well why not measure it:

Firstly measure how much traffic you get from your main brand term from the natural listings in a certain search engine. Then you need to work out your ROI of that search term (by looking at the number of conversions it drove and the amount of money that you made on those conversions). Then you need to start bidding on that search term. Work out how much money you have spent on driving traffic to the site, how much traffic you have driven to the site and again, the ROI of that traffic.

Now comes the important bit - you have to make sure you are not just cannibalising your organic traffic. This is important, because if you take your brand traffic on its own, you will see you have a good ROI because those people are already engaged in your brand (hell, they managed to search for your company name - they must know you already). Also you'll discover you have a good ROI because it doesn't cost you much to bid for your brand name (the cost per click of a brand term is much lower than a product term usually), especially if nobody else is bidding for it (which they shouldn't be).

How do you find out if you are cannibalising your own traffic? Look at a brief before and after picture. This is the traffic we were getting before (and an ROI), this is the combined traffic and ROI that we are getting afterwards. Are you getting more traffic now than you were before when you weren't bidding on your own brand? Are you getting more money now than you were before when you weren't bidding on your own brand? Is that extra money you are getting worth the amount you are outlaying on it (because you are bound to cannibalise a little bit of the organic traffic)?

The areas that you might find a better option for bidding on are areas that you aren't currently getting the organic traffic. That means bidding where you would expect the user to think of you, but they haven't quite typed your domain/company name in first. Think about these ones:

• Popular misspellings - most people know how to spell your company name, however there are thousands of ways of spelling it and if you don't pick those pages up with SEO (which you won't, because usually you don't misspell your company on your own website) then you need to stop them going to competitors (or affiliates) who will be bidding on them to get your business
  
• Slogans (past and present) - these are important in the world of advertising. People will remember your slogan or your tag line and will search for it in a browser if they can't remember your company from the ad. These slogans are part of your brand and should be picked up as such
  
• Brand terms with other products - people bid in search engines for products. It's not cheap. They think they can get a good return on it. If someone searches for your brand and that product, they'll likely get other rivals paid ads being shown up. They want to come to your company (they searched for it, didn't they), so you might need to give them a bit of help in this case.

However remember the rule above: If you are still cannibalising your own traffic and you aren't getting the ROI it is not worth it. Bidding on search terms is all about Return on Investment and with your web analytics tool you should be able to measure that to the nth degree.

The end of HBX is not quite nigh

It's all been happening recently. Well, if you live in the world of Analytics like we do then it has all been happening. Firstly there was news that Omniture and Visual Sciences have merged (more of that later), then suddenly the whole of the WebTrends board mysteriously stood down (more of that later as well). Now suddenly whilst we are contemplating the number of tools going down by one, Microsoft have made it go back up again with Project Gatineau. Wow, nothing happens in the field for months and then three come along at once (I won't go on about London buses).

Well lets start with Visual Sciences and Omniture, because that was quite big, wasn't it? Well $396m good if you believe the statistics (and I know we can do anything with statistics in this industry). But what does the outcome mean firstly for the web analytics industry and then for those VS and Omniture customers?

Well for the industry as a whole we have been talking for ages about consolidation so this whole thing shouldn't come as a surprise. Are there too many different tools in the industry? Well yes (from a personal point of view), because every time I start working for a new company I have to work out how it works, why it is different, how it is implemented, how the interface works, which segments work best, etc. If it takes me ages to work out how to do it, imagine how long it takes someone who isn't a Web Analyst to work it out if they move companies? Then think about what happens when some smart arsed Web Analyst comes along and tells them to change - you need a whole team of Web Analysts to tell them how to use it. Now we have a bunch of reluctant users even more reluctant to use the web analytics tools.

So consolidation is good for the industry. Although obviously not too much consolidation because if you don't have competition then product development goes down the pan, as does support, etc. But this whole deal may not be the best thing for the customers of the aforementioned companies. Lets start with Omniture customers - everyone is assuming that they won't generally be affected. So our SiteCatalyst friends will still get to use SiteCatalyst and their support will still be the same phone number (is that how it still all happens?). Assuming that their support centre's number is clogged with HBX users who have migrated.

So these HBX users (and there may be more of them since VS is going to continue selling it until the deal goes through). What are their options? Well they could hang on for as long as possible in HBX if they weren't thinking of changing their system. They'll probably discover that as clients slowly move off HBX onto other tools (possibly SiteCatalyst if they give a good deal to ex HBX people), HBX stops moving forward product wise and support slows down and eventually there aren't enough left for Omniture to continue support of the data centres and it gets closed down giving you no option. The trouble is that there are so many of these HBX users that this could take well up to three or four years to happen. The advantage of HBX had always been that it was relatively easy to use for the average users (this is why so many media companies use it).

So now we have to worry about the thing we were just talking about - moving to a new system and training your users how to use it. Do you go for SiteCatalyst which is a bit more complicated but more malleable for the Web Analyst? Do you go for WebTrends which is a little less malleable, but they reckon it is a viable alternative to HBX. Or hey - what if you are using the Visual Site upgrade on HBX? This uses HBX tags, but transports the data into a different data centre, what will they do with that. Maybe this is why they are taking 6 months to finalise the deal so that they can think about it.

Now the main challenge of changing tools is not the training of the new users (including yourself). Implementation is something so challenging that entire analytics consultants have made a living from it (and if they haven't they could do). More to the point if you do it wrong, not only do you have a bad solution, half the time you are missing data for a period of time whilst you've taken the old solution off and started on the new one. I've also been through enough implementations to know that you'll forget about something at some point that you'll have to go back and do later, so this is an ongoing job. Keeping your HBX tags for as long as possible is suddenly sounding more appealing.

Well we've mentioned that WebTrends was a viable alternative. Everyone at some point in their life has used WebTrends. There is something that is quite satisfying about arriving at a new company and discovering that at some point in their past they have used it and nobody liked it. that's a bit OTT - the reason that it generally gets bad press is that you don't have a FTE to look after it and tell you what it means. However what is going to happen now that they have a new board and why did they do it? Is it because they want to have a fire sale now that they've lost out to VS in the sale? Is it because they want to change their direction given a slipping market share? Do we think that they were trying to get sold and the board didn't manage it, so they've paid the price? I think this one might come out in the wash over the next couple of weeks.

As for Project Gatineau (can we please come up with a better name for this? It's going to annoy the hell out of me because only the French will actually be able to pronounce it properly). This could be the dawn of a whole new level of Analytics. Demographic information in your WA tool? Conversion by age and sex? Acquisition sources by age and sex? Telling your boss that you're missing Mosaic Group E and you need to focus on increasing your traffic from this high spending group could suddenly become very powerful.