The goal of web analytics is to make it easier for people to complete the tasks both of you want completed.
This article assumes that you do not have a lot of money to spend nor are able to hire a dedicated team for web analytics. You just want to educate yourself on web analytics 101. Other blogs to learn about web analytics:
- Avinash Kaushik, the main source for this page
- www.liesdamnedlies.com, a Microsoft perspective
- Analytics Talk, contains code snippets
- Analytics.blogspot.com, the official Google Analytics blog
- Blogs.commerce360.com
- Blog.instantcognition.com
First: Why do you have a web site? Lead generation, e-commerce, customer support? As a distribution channel, for brand building, for sharing information about your company? Or because everyone has one?
Second: what do people want on your site? This is hard to answer in general, as they all are different.
What can you measure?
1. Clickstream data, what people click on in your site. The traditional province of web analytics, showing you what happened.
2. User feedback from surveys, testing, usability studies and so forth, helping you to understand why it happened.
3. Business outcomes like revenue, leads generated, lowered support and distribution costs. Tells you if what you do with the website is actually doing any good.
When you measure data on the web, where you can track what people do in a level of detail never before experienced by the unwashed masses, remember one thing: data quality still sucks. Also, don't bother with real time data. Looking at stuff a day or two later is more than enough.
How do you measure success?
The visitors who did what you'd like out of all visits are your "conversion rate". Typical conversion rates are about 2%, because many people are on your page for reasons entirely unrelated to being converted. In practice you can only measure where someone clicked, so you often measure some proxy for conversion rate real outcomes.
For e-commerce or lead generation conversion rate is straightforward to measure, you tag suitable pages like "Thank you for ordering" as a conversion. Use this to segment your other data, so you can understand what makes people convert or how traffic from search engines does compare to pay-per-click traffic. If you can assign a monetary value to your conversions, you also will be able to calculate a rough ROI on link advertising.
For support or brand building, you need to use surveys. You never know if someone read a knowledge base article and left because his problem was solved, or because he just got frustrated and gave up. The best way around that is to ask them through embedded surveys and hope they will answer.
Surveys are annoying, so keep them minimal. The simplest, and best would just run:
1. Why are you here today? (Always provide the ability to answer free text together with any predefined options.)
2. Where you able to complete your task? (Yes/No.)
3. If not, why not? (Again, free text.)
Other "soft" data come from observing customers either through click-density overlays, or on the web (Userfly), or in usability tests, the hallway kind where you pull in an unsuspecting coworker, or formal ones with studio recording equipment, or in follow-me-home studies. The costs increase here.
For collecting click-stream data, there are many options, but honestly, just sign up with Google Analytics. This uses JavaScript tags, small code snippets which call a logging function on a remote server. Advantages: only measures stuff people see in the browser, can be highly customized, decouples the data collection and processing from your site, so less dependence on IT folks' mercy, sophisticated, powerful analysis reports for free. Disadvantages: your data is stored by someone else, who may not give you access for in house analysis and integration. You will be blind to people who have switched off JavaScript, about 6%, of average users. (Another free for the basics is www.statcounter.com, there are also many commercial ones.)
An alternative are Server logs. These have the "advantage" that you also see robot traffic, disadvantage that you need access to them and the cooperation of whoever runs your IT. (Free tool: AWstats.) There also are exotic or outdated solutions like "beacons" (embedded image URLs pointing to a third party server, think tags, but less functionality) and ISP packet sniffing.
How to analyze
Data without context is powerless to provide insight. Compare data against past performance. Look at different time frames to see if there are seasonal fluctuations. Segment your data, for example by source, to find areas of weakness.
Compare against your competition. There are several sources of free competitive information:
Panel based: this works best for sites with millions of visitors per month and can be detailed. A number of people sign up by to be monitored and the data is extrapolated. (E.g. comScore, or Alexa, free and Compete, partially free).
ISP data based: data is harvested from raw traffic and is more superficial. (E.g. Hitwise, also not free).
Search engine traffic. Again this will not tell you how people behave in detail on some web page, but a lot about what they are interested in (E.G. Google Trends, free, or Microsoft AdCenter Labs, free.)
Link based: ranking popularity by links to a page (Marketleap, which also has a report ranking how you and competitors do for a key phrase on several search engines, or just Google with the link:yoursite.com syntax.)
Other interesting sources for external benchmarking (mostly for bigger firms) are the American Customer Satisfaction Index, or the Fireclick Index of typical web site metrics.
Adoption
The best way to get started is getting your hands wet with a practical tool. Produce a report. This will allow you to gather feedback on how to change that to support decision making and work out problems with your site's URL structure (which may impede measuring the same page as the same page) and which features you need, should Google Analytics not suffice.
Technical hints
Tag all your pages, best at the end of the page and not in tables or such, so the tags execute and do not hamper loading. Make sure the tags work as you'd like them to. Run a link checker (eg www.relsoftware.com/wlv, not free) to make sure all pages are tagged.
Identify your unique page definition. What is a page? Care about link coding; parameters and session ids tend to muck up URLs.
Be careful of redirects, use 301 permanent ones wherever possible, as only these pass along referrer information. Be careful about links wrapped in JavaScript, which make clicks on them hard to track. Also about rich media like Ajax or Flash.
Collect source info (website, campaign, search engine etc); page info; and user info through persistent anonymous id cookies so you can identify repeat visitors. Use first-party cookies with your own domain name, not the vendors so they survive stricter cookie security settings. Google Analytics does all that automatically (except campaign tracking). Measuring cookie rejection rates gives you a better idea of the quality of your data. You can also collect if someone is logged in, part of a test etc.
What metrics are useful?
Overall traffic ("unique" visitors and "total" visits). Be careful that while unique visitor usually refers to someone identified by a persistent cookie, so that repeat visits would not increase the count, some tools incorrectly record each session as a unique visitor.
For most of the following metrics, segmentation is needed for understanding, because they are mashing together users with different intent.
Conversion. Without some way to measure success, knowing all these other things will still find you helpless what action to take to improve them. Try hard to come up with some proxy for this. Don't try and measure it by page or link, it makes no sense, but it is possible and very useful to see what pages converters visited before, because these may be the ones with the convincing arguments.
Top referring URLs. They help you understand where your visitors come from. Segment them by conversion rate, to understand the quality of traffic they send you.
Top search phrases and top viewed pages. These give you a good idea what people are looking for. For the top search phrases, do a competitor analysis to see how you are doing. Especially important are category terms, that represent the general term, not a vendor specific one, like car instead of BMW. Have a look at search funnels to understand what people searched before and after that phrase, you can even get a forecast on what might become popular with demographics. Some additional sources to check what people are looking for: Google Insights for Search, which is great in showing also related searches and changing trends. For bio-scientifc work, GOPubmed is a great tool to track trends by publications.
Search phrases in your on site internal search engine are also a good source for this, and may hint at things that are hard to find with normal navigation. On average, 10% of users use site internal search to navigate a page.
Top entry pages. As a lot of traffic tends to come from search engines directly into a page deep within your site, the top ten entry pages are probably just as important as the home page. Less than half of your visitors ever see the home page, so do not waste to much time in optimizing it, better spend that time on seductive navigation. Exit pages, in contrast are pretty useless, unless you are looking at a multi-page checkout process, because people can exit for many reasons, and exiting is not necessarily a bad thing.
Bounce rate. A "bounce" means someone came to your page and did not click anything but directly left within 10 seconds. Bounce rates of 40% are common. There is no way (at least without Ajax) to see if someone still has a page open, and if the page is open if they are still reading or left for dinner. Therefore, usually tools decide that if someone doesn't click on anything for half an hour, they have left. If someone left from their entry page, this is then counted as a bounce. Bounce rate is critical because every bounce is a valuable visitor wasted. Time on site suffers from the same malady as bounce rate that you have no idea how much time people spend on the last page they visited, unless they navigate off that page through a link.
Abandonment rate. This is only relevant for e-commerce sites, and describes the rate of people who leave after putting stuff in the cart or initiating checkout. This can be measured on a small number of pages, and affects people who already wanted to buy - ripe ground for optimization.
Pay-per-click (PPC) advertising. Measure raw page impressions, click-through-rate, bounce rate. If you also have conversion rate, with monetary value attached from overall cost by conversions you can get the cost per conversion, and revenue/cost for PPC gives you ROI. You also should make a reality check the PPC cannibalism of your natural or "organic" search traffic. In the long run, a more helpful and appealing page (and product!) is a better investment than PPC.
Dashboards
When you create dashboards or reports from these metrics, make sure that numbers show context (against the past, against a benchmark), that it is limited to an overview, and that you annotate it briefly with insights and recommended actions. If it takes more than one screen/page at normal font sizes and borders, you have a report, not a dashboard. Use graphical representations like pie charts, bar charts or scatterplots to make the data intuitively accessible. A good test if a metric should be on a dashboard for management is: can you clearly explain how it maps to a business objective? If popular exit pages can not, they should not be on.
Testing
A great way to improve your site without annoying the users with questionnaires they do not like to answer anyways is variant testing, where you show different pages (A/B testing) or different elements on a page (multivariate testing) to different users. If you have some way to measure success, you can then see which of these performed better, and use that in future. This is really awesome: it is observing users to give them what they want, and it is non-intrusive and cheap. It also frees you from assumptions or following the HiPPO (highest paid persons opinion -- you know, when the VP of thisorhtat really likes the mauve pages with puppies, that's how it will look like).
You can test different images, lots of text vs little text, flashing moving stuff vs quiet elegance, a big hero image on the home page or three smaller ones ... the possibilities are unfortunately endless. There should be a hypothesis behind the different things you try. Start out with one single goal and with a simple test. A great way to get buy-in and interest is to have people bet on the outcome. Start to a high-traffic page.
A test plan should contain:
1. The hypothesis, eg "Customers come for scientific soundness, so lots of formulas and text will work better than a smiling babe in a bikini."
2. Business case. What are you trying to optimize, and how does it help the business goals.
3. Audience. Who will be in the test: what percent of traffic, what source of visitors, people who do what on the web site?
4. Details: How will the alternatives look like (Mock-ups), what pages are affected, what tracking code needs to be put into place.
5. Success measure: what are the metrics, how do they look now, and what will count as success for one of the test options?
6. Outcome actions: who needs to be notified, what will be implemented?
No comments:
Post a Comment