Companies can save a lot of money if customers use self-service systems like Web sites, kiosks and interactive voice response (IVR). But they can lose even more money if the systems are so annoying that customers take their business elsewhere. Since the systems run unattended, special efforts are needed to understand how customers interact with them and identify potential problems.
The key tool in such efforts is a “funnel analysis.” If that tracks how customers enter and exit particular paths, such as a service request or checkout process. Funnel reports are built by identifying the sequence of stages a customer passes through during a process. Analysts pay special attention to transition points where customers drop out or take undesired actions such as asking for live help. They must also look at the process as a whole to ensure that changes, which appear to be improvements at one stage, do not harm performance somewhere else.
Most self-service systems maintain some type of log file that records the individual events and associates them with customers. But the structured query language (SQL) used with conventional relational databases does not easily identify the pairs of records that indicate movement from one funnel stage to the next. This means that special analytical tools are needed to convert the log files into meaningful information.
For Web sites, vendors including Coremetrics, ClickTracks, Omniture, WebSideStory and WebTrends have built such systems. They capture the required data either by reading the Web server log files or by tagging selected Web pages with small Javascript programs that notify a data collection server when the pages are rendered. Either way, page views are tied to individual customers with cookies (small files placed on the user’s PC by the Web server) that send a session ID or customer ID with each page request. The analysis system uses these to knit the page views into a picture of how customers moved from page to page during the session. To create a funnel report, users identify the specific pages or groups of pages that represent each stage. The system then extracts movements among those pages from the larger set of data. Page views are often supplemented by additional information, such as purchase amounts or customer attributes, which is captured in other systems and matched back using the customer or transaction IDs.
ClickFox, a product of ClickFox Inc. (www.clickfox.com), provides customer path analysis across many types of self-service systems, although its particular focus has been IVRs. It classifies the events captured in the system’s log file, using a model that shows how each event fits into the general flow of the system. ClickFox then imports the system logs, which show the events experienced by individual customers. It maps these onto the model and displays a visualization of their path through the system.
The key to this is building the model ClickFox reads the log files to build an initial map of the application, which it presents to users for clarification and fine-tuning. ClickFox engineers then create the actual models. Models also can be imported from third-party flow design tools such as Cisco Audium.
ClickFox can display the path for a single customer interaction, paths for similar customers or paths for similar interactions. Users define customer segments and interaction types with a combination of event log data and information imported from other sources. Such information might include customer attributes, transactions, revenues or costs.
Users can also identify a set of events that make up a particular funnel or task, such as opening a new account or placing an order. Other events can be labeled as task outcomes. This lets users analyze particular customer activities and identify problem areas.
One ClickFox model can track customer activities across different systems, as long as events within the individual systems have been mapped and interactions relating to the customer can be linked. If the different systems use different customer IDs, ClickFox can maintain a cross-reference table that captures the relationship. Creating the table itself – that is, identifying IDs in different systems related to the same customer -must be done externally.
Analyses can combine segmentation and task definitions to examine whether different sets of customers react differently in particular situations.
ClickFox reports can show any behaviors captured in a model. Interaction systems cannot provide many of these because the behaviors are defined by model categories. These include task success rates, distributions of outcomes and behavior by segment. The system can generate financial evaluations such as return on investment models, although this requires some custom development. It can also issue email alerts when specified conditions occur. Although log files can be uploaded frequently, the system does not report on results as they occur in real time.
The system can compare tests of different interaction system rules, so long as the test cases can be separated into distinct segments. But, because its reports are limited to actual log data, it cannot perform “what if” analysis to estimate the impact of proposed rule changes.
ClickFox also has some “automated intelligence” that flags behavior patterns that appear to indicate problems. These might be frequently skipped stages in a standard process or frequent cycling between two stages. But the vendor reports these features are used less often than data visualization to determine which situations to explore.
The system holds the log data in a proprietary file format for better performance. It has processed IV Records from more than 20million calls per month. Although the primary focus has been on IVR data, ClickFox says about half its clients now combine logs from more than one channel.
The company was founded in 2000 but until recently has worked mostly on consulting projects. It is now expanding aggressively and has about 20 active customers.
Most customers use a hosted version of the system, although the software also can be licensed and run in-house. Annual fees can range from $150,000 plus services to several million dollars, based on the number of sessions tracked.
