Funnels are one of the most basic and important tools in web analytics for understanding how online interactions with customers feed into both on-line and off-line business processes and affect the bottom line of a business. Funnels are prominent in all sales organizations and most business processes as a simple but effective form of predictive analytics and method for monitoring impact of new actions. Funnels capture the notion of time that it takes to complete a process or task and effective milestones in the process of moving leads, RFQ, and RFP through to sales, orders or contracts.

One would think that such a powerful tool would be second nature in web analytics. However because of different implementations among the different Web Analytic Providers (WAP) and potential confusion with paths and channels and different roles of conversions or success metrics play in different analyzes, setting up funnels can be confusing and too often non-productive.  To overcome these short comings, let’s take a fresh new look at this important analytic methodology and determine how funnels are different but yet support our other views of visitor actions as parts of paths and channels.

Also we will look at how you can determine how your particular tool implement funnels and a way of including funnels in any of your custom reports regardless of your tools.  My attitude is that funnels form a compelling prerequisite for action and should be an element of every actionable report. Let me demonstrate.

Funnels vs. Paths

Somehow the concept of funnels got equated to paths.  Almost all of the current web analytic providers do this.  They have implemented rather impressive funnel diagrams that seldom ever represent funnels.  Funnels as the name implies are wide one end and narrow on the other, whereas paths are steps that a visitor performs to complete a task.  Drop-off along the path means the task was not completed and should be remedied.  The objective is not a funnel but a pipe! Because funnels and paths are used interchangeably, users can become quickly confused when results don’t match their notion of either a funnel or a path.

To establish a conceptual difference and a framework for dealing with both funnels and paths, let us consider an example of both outside of the realm of analytics and business.  Consider the funnel described by this Persian proverb – “Of all that pass the garden gate, some will stop to look through the gate and view its beauty.  Fewer still are those that pass through the gate to enjoy the pleasures of the garden, and very rare are those that remain to tend the garden.”  Continuing in the same vain, consider a path in another Sufi proverb – “Once one has entered the garden, there are many paths to its center”.

With this description of a funnel, one feels they are confronted by some profound truth concerning human nature, where as with paths we see the individualism and arbitrariness in attempting to understand that nature.  Of course, the marketer in us immediately is thinking of campaigns to increase the number of persons passing to stop and bring those that stop into the garden.  Maybe we could have special “Tend the Garden Days”, where there are tools and plants ready with instructions bypassing all the other phases and go directly to tending.  As for that path thing, just make one big path to the center; even add a wall to keep them from meandering off the designated path. Of course this may entirely defeat the purpose and beauty of the garden but perhaps does capture the difference between East and West thought, Ancient and Modern approaches, but that would be another discussion.

From this we see that funnels capture an aspect of how human nature affects your business independent of all the ways that human nature can be manifested or affected, whereas paths represent the specific steps that an individual must take to complete a task and the specifics of human nature are at its very essence.  You would not present Chinese instructions to someone who did not understand Chinese.  In the same sense you want to present the most clear message and instructions that will motivate the individual customer to complete the target action.  The funnel will give you an objective view over all individuals and markets, whereas paths will all be different and targeted to individuals and markets.

Another way of differentiating funnels and paths is that funnels are typically associated with the visitor and paths are a characteristic of the visit.   Funnels recognize that the transitions in visitor perception and commitment occur over many opportunities to interact and take timescales much longer than a single visit. The transitions that represent the stages in the funnel should be associated with transitions in what we know about the visitor’s history or current state.  As the visitor interacts with the web property more information is accumulated as to the perception, intent, and commitment of the visitor.  This interaction data becomes the working storage for the visitor upon which rules or algorithms are applied that diagnoses or interprets these actions.  The rules based upon conditions in work storage fire events that indicate transitions in visitor state that in turn expands the working knowledge of out visitor.

This is how Expert Systems in AI perform diagnosis and was the basis for developing what is now referred as conversions and conversion funnels.  Purchases made by individuals are clearly such events, but prior to that action, there are a number of other actions that indicate the visitor’s intent and commitment, such as clicking on the “Buy Now” button.  These visitor state transition or conversion events are different from content related events such as page views in that they attempt to capture the visitor’s “progress” through a sales / business process captured by the funnel.  However the act of visiting a page – such as coming to the homepage for the first time, or viewing the items in the shopping cart, or viewing the confirmation page after a purchase may indicate a visitor conversion event.

Paths on the other hand look at the sequence of pages and content viewed by a visitor during a session.  In web design, the potential paths that a visitor can take are broken down into smaller sequences called web-flows.  A web-flow is a unit of content presented as a sequence of page views where the visitor completes a specific task such as place an item in a shopping cart, register for a newsletter, download and print a coupon, and checkout a shopping cart.  A typical large site can have thousands of web flows by design to cover all of the variations and options presented to a visitor.

Within a web-flow, the user experience may degrade sufficiently that the visitor leaves the flow out of confusion, frustration or distraction.  In designing pages and web-flows, one has to consider the appropriate combination of motivational messaging and clear instructions while at the same time communicate unified look feel and brand experience over all flows.  With paths you are at the center of understanding visitor behavior and the web analytics should provide means of first detecting problems (drop off), and data to explain the reason.  A bonus is if the data can also support improvement through testing and adaptive learning.

Channels, Personas and Segments: How to keep them straight.

Since both funnels and paths have an interest in visitors – one captures the state of the visitor and the other attempts to target the visitor – how do aspects of the visitor such as channels and segments come into play and how do we keep straight channels from segments?  In general segments are simply how we group or characterize visitors or visits so that an individual can be grouped with other individuals that share the same segment characteristics.

In a strict sense, channels are a form of visitor segment indicating where the visitor came from prior to your web site or web property.  For marketing, a channel represents an opportunity to reach potential customers outside of your store or web site, or when they are within the store / site to direct them to particular department or property.  For sales, channels have a similar meaning – defining where and how you can reach potential customers to move them through the sales process.  Web analytics channels in general represent where the visitor came from prior to the web site or property entry, called an introduction.  It can be external (affiliate, search, email) or internal (search, micro-sites); associated with specific providers – Google, Yahoo, Right Media, DoubleClick; methods – email, search, banners, advertising; or media – online, TV, radio, print.

Independent of channels, Marketing and User Experience Design also segment visitors for their content – former by need and the latter by capability.   These form segments called personas, where Marketing and UED develop rather detailed profiles from marketing research, lab studies and surveys that name and identify the needs, experience, motivation, and capabilities of different representative individuals that come to the site. Personas define the audience of the web site or product and segment the visitor by their need and capability independent of the site.  It is the function of the web site design and content authoring to address these personas.

Unfortunately, this is not the reality that web analytics directly observes, but instead must infer from the clicks and vars that are collected in the web analytic logs and rule annotations derived from processing these logs.   One ends up with visitor segments of opportunity so to speak, such as new / returning visitor; or bouncing / sticky visitor; on-line purchasing visitor; or ad-viewing / ad-clicking visitor.   The visitor can be segmented by any combination of the hundreds possibly thousands of ways the visitor’s behavior is distinguished in the analytic data stream as given by applying axiom 5 (see How Web Analytics Sees Reality).  These analytic visitor / visit segments are independent of channels and personas and the web analytic professional must work with Marketing, UED, and Sales to map analytic visitor segments to these conceptual segments, because the latter are the primary drivers for the creative design and content authoring.

Even though personas may be based upon best research and science, it must also be recognized that mapping web analytic segments to marketing/design segments will be neither direct nor complete.  This gives rise to fourth kind of segmentation where analysis of the analytic data streams can, through statistical or adaptive testing, and cluster and classification analysis, yield new segments correlated with key performance indicator (KPI) performance.  These are referred to as targeting segments.  To explain the reason and further optimize content for these segments may require adjusting or modifying persona definitions and willingness to experiment with content to reach segments that cannot be clearly defined or profiled.

All this to say that funnels are not segments.

Segmentation within Funnels and Paths

One of the potential confusions in implementing funnels is attempting explicitly or implicitly to setup funnels for different channels.  For example, one analysts states “I have set up four funnels with each having the different entry – search, email, display, affiliate – and the other conversion events the same. The problem is that conversion counts for all the different entries are all the same.  Shouldn’t they be different?” Let us see how this question can be addressed.

First – Declare events that are important to your over all business. Funnels should be above and independently defined from channels and other visitor segments.  This means the funnel entry events or internal transition events should apply regardless of the visitor segment or channel.  Likewise paths should be actual sequences of content or pages with the objective of determining latter which paths are favored by different segments of visitors.

Second – Identify your funnel from the data collected for these events.  As implied from the name, funnels are wide at the entry and narrowest at the exit.  As you collect counts from the events you defined in the first step, you should be able to group and stage them so that they form an actual funnel.  Your funnel entry is the first time that the visitor comes to the site as a new visitor.   There are of course different ways that they can come to the site but you should be able to collect all these into one funnel entry event type.

Likewise you may have an event for downloading a white paper and another for someone requesting more information.  These may be the same measure of the level of intent and commitment and should be grouped as different ways a visitor makes that funnel transition. Latter we can ask of those that made this transition, how many did it through white papers, RFIs or both to understand the more specific behavior.

Third – form a baseline of the general trend for the funnel. You should be able to form and report a general funnel providing the counts for each transition in the funnel.  Though the numbers reported in each funnel count report will differ eventually the ratio of the different counts will trend to transition ratios that represent the likelihood of a visitor making the transition from one phase to the next.  As a predictive tool, this report gives with a measured level of confidence, the likelihood that a new visitor coming to the web site or property will eventually complete the funnel – e.g. make a purchase.

Obviously, we want to go beyond this report and understand what is happening underneath – drill down to the visitor segments and channels to determine if any have better transition rate metrics than the general trend.  Thus,

Forth, workout the attribution between visitor segments and funnel transition events.  Basically the transition event inherits the visitor state at the time of transition including any previous channels the visitor came from to the number of previous purchases the visitor has made.   Almost all WA tools will do this automatically for special forms of variables called marketing parameters that you add to tracking URLs submitted to marketing channels such as Google or Yahoo.  As a result every transition event can be attributed to the last set of marketing parameters for the visitor.   The result of separating out the funnels based upon marketing parameters is another form of funnel called a conversion funnel, which you should realize is different from the general funnel we constructed above.

Conversion Funnels

OK, so funnels maybe aren’t so simple! Bear with me and we will get this all straightened out.  To help keep track of the different funnels, call the general funnel that started with new visitors to the site as the web site or business funnel since it is looking at the transition of new prospects to real customers regardless of whether or not these transitions take place on-line.  The conversion funnel on the other hand attributes and measures the likelihood that a specific marketing campaign or channel will lead the visitor through a transition in our business funnel.   Each transition regardless of where it occurs within the business funnel is a conversion.

So one of the things we need to know about conversion rates is first how many potentially could have converted from the marketing campaign and how many of those did convert FOR A DESIGNATED CONVERSION TYPE.   The ratio of the two measures gives the conversion rate for the transition attributed to the marketing campaign.

Think about it, if a marketing campaign is attributed a purchase conversion that does not mean that the same campaign also brought that visitor through the other stages of the funnel.  In most cases another marketing campaign introduced the visitor to the site and yet another brought them to the site to download a whitepaper.  Therefore when looking at a conversion funnel, you are looking at the propensity of the marketing campaign to generate funnel transition events and not final funnel purchases.

If your marketing campaign has a high conversion rate for a specific business funnel transition but leads to few or no purchases, the campaign still has value in moving prospects through the funnel.  Even though the campaign cannot be justified on direct sales or purchases, it has widened the funnel that will lead to more purchases if your business sales ratios (conversion rates) are correct.  The key point is that you first have to believe that the funnel represents your business process and then confirm that it does by collecting the data.  Then if you widen the funnel at one point you should observe it eventually increasing conversions at the end of the funnel.  If you can demonstrate that your funnel can respond to “shaping” then you will have a truly powerful tool that models your business process and from which you can monitor and understand changes.

This tends to bend the mind a little to really appreciate the value of these metrics.  The causal relationship between marketing and business processes seems indirect. Because of this web analytic tools, such as Omniture SiteCatalyst, add a restriction that the marketing campaign must have attributed to each of the transition events in order.   The result is that the number of conversions attributed to the campaign is much less for any of the funnel transitions.  Also total attributions of final funnel events, e.g. purchases, to marketing will be less than the total number of conversions without attribution.  The reason is that very few marketing campaigns (in a properly defined business funnel) can meet of the requirements of satisfying all funnel transitions in order. On the other hand, this is precisely what you need to understand drop-off in a path.

Page Paths

So what may be a lousy funnel report may be an excellent page path or flow report.  You can determine which one your tool implements by comparing the total number of conversions for each type to the sum of all conversions by type attributed across all marketing channels.  These should be exactly the same for a funnel report.  What may happen is that most of the conversions go into a non-attributed bucket that when added to attributed counts does give the same total count.  In that case, determine if the non-attributed conversion counts are reasonable.  If you have an active marketing effort, non-attributed should be smaller that all your attributed counts, since non-attributed channels cover visitors with no referral information and these are usually a small percentage of overall traffic.

With paths one is playing croquettes where there are penalties for missing brackets out-of-order.  The sequence of pages in a path do not necessarily have to be contiguous but most be reached in the order defined in the path sequence.  This is great within a web flow where there is possibility that the user will go backward in path to view previous information but will still go forward thru the sequence to complete the task.  Or maybe the user while going through a checkout does other searches or reviews previously loaded content to confirm an option or technical detail about the product before returning to complete the checkout.  Another strategy is to set up alternative paths and determine which path sequence leads to the most completions.  All these are possible by defining paths that are specific sequence of events and comparing the effectiveness of different sequences.  As useful as this approach is, the paths are not funnels but a funnel will incorporate many of these different paths.

Even though these path reports are not funnel reports, they are difficult reports to define and generate so the good news is that most analytic tools provide support to generate these difficult reports. Funnel reports are relatively easy to generate and incorporate in custom reports or spread sheets since it is just conversion counts ordered by funnel transition sequence attributed to marketing and channel variables.  To normalize the conversion counts within each channel, divide each transition count by the total number of impressions or clicks at the beginning of the conversion funnel to compute the conversion rate for the channel.

Getting the total number of clicks for each channel maybe difficult if the WA tool treats this data as another form of session parameters.  For example Omniture only allows marketing parameters to be attributed to events, page variables such as domain do not automatically become attributed to events.  A work around that I have used that works for most tools is to define an introduction event that is declared when referrer and landing page URLs have different domains (axiom 4) and then immediately associate marketing / channel parameters to this event regardless of whether or not they start a session.  Introductions have 100% conversion rate for clicks so can be used to compute conversion rates. Introduction counts then give the start of a conversion funnel and subsequent events can be attributed to the introductions, which then can be further segmented by introduction parameters.  In this way, the visitor state at the start of the conversion funnel will be consistently attributed to all the subsequent transitions within the funnel including the end conversion event.

Alternative Measure of Visitor State

To bring this all into better focus, let us consider a way of designating the visitor state with respect to the business funnel.  We have already identified that entry into this funnel begins with a new visitor, no previous visits and no previous passes through the funnel to the end conversion (n=0,m=0).  Now every time a visitor returns to the site via a new introduction, we continue to increment the first count (n++, 0), which becomes the number of previous introductions.  So a returning visitor is any visitor with more than one previous introduction (0 < n, 0).  When the visitor eventually completes the funnel, the visitor state will reflect the number of previous introductions to current introduction that is attributed to the conversion (N, m++).  The pair of counts representing the visitor state (n, m) supports some valuable metrics including the average number of introductions before a conversion <N>.  This metric is important for understanding how the visitor is truly interacting with your web site and where your marketing campaigns are effecting this interaction.

Upon reaching the end of the funnel, the visitor state is reset to (0,1) for the next introduction.  The visitor starts a new business funnel for the next purchase, upgrade or any other action you wish the customer to complete.  This is referred to as an opportunity touch point funnel where each introduction is an opportunity to market the visitor and each visitor state (n, m) is a touch point between the visitor and the business funnel.

This funnel shows how marketing campaigns, keyword searches or channels contribute to the business funnel as distributions of visitor state (n,m) within the funnel.  One might find that keywords contribute to bring the visitor to the site early in the funnel and aid in transitioning the visitor in the funnel, while others appear to be closing terms that make the sale at the end of the funnel.  In a sales funnel, the early transitions represent qualified leads that have a specific value to the sales force.  The more qualified the lead, the higher the close ratio and the more valuable the lead.   By assigning values to the various transitions one can define an upper limit to the cost per acquisition (CPA) that one can tolerate for marketing campaigns or search words that create the leads.  Using CPA for qualified leads further into funnel instead of Price Per Click (PPC) at the start of the funnel or Return On Investment (ROI) at the end of the funnel can provide better opportunities to manage both marketing cost and business revenue.  For example if one optimizes to ROI, one might find excellent ROI on closing terms but poor revenue due to not expanding the number of leads earlier within the funnel. With CPA one can better manage increasing revenue by shaping the funnel and yet control cost to meet an overall target ROI objective, where investment includes all marketing spend.

Summary

• Business funnels are not paths but instead are means for describing how your business and web site converts new visitors into valued customers.
• Conversion funnels are not business funnels but a means of attributing marketing efforts to transitions in the business funnel.
• Paths are neither business nor conversion funnels but a means for observing and classifying individual behaviors while on the site and indicate when users decide to drop off from the path.
• Channels are not funnels or paths but are instead are sources for the various entries to your funnel or path.   Channels deliver audience that represents an opportunity to market and engage the visitor outside of the store, the web site, or the internal property.
• The audience that is delivered thru a channel can be further segmented such that marketing and product content presented to the visitor can be tailored to the segment.
• Using visitor segments derived from analytic data one can form targeted segments that are demonstrated through testing and analysis to be correlated to KPIs.

If you can keep these terms separate and in proper relation, then you will be well on your way of having a very productive web site capable of supporting performance marketing, multivariate testing, behavior targeting and site optimization.  If this still has not yet all jelled then at least remember this one rule of thumb:

“When two things with different names are said to mean the same thing, then question and look deeper, for most likely they are truly different and are given two different names for a reason.”