I recently got a chance to discuss EpiAnalytics product offerings and see a brief demonstration. EpiAnalytics provides contact center analytics to improve customer service and technical support business processes using analytics (including text analytics) and decision automation to automate the manual analysis of leads and support emails. The results are then plugged into an engine that can handle routing, scoring, assignment, automated response etc. In particular the product is integrated with salesforce.com – EpiAnalytics has an AppXchange offering launched in October 2008 that essentially puts a layer of analytics on top of salesforce.com workflows – though they are beginning to get requests for other platforms. While not the only hosted decisioning platform out there, this is the first I have seen that comes pre-integrated with salesforce.com in such an interesting way.
For instance, if a salesforce.com user has customers or prospects that send sales query or support requests (any mechanism including emails, web pages or live calls manually entered) then they can use EpiAnalytics. The initial contact gets stored in salesforce.com and a SOAP call goes to EpiAnalytics. EpiAnalytics uses Support Vector Machines and both structured and unstructured data to automate categorization of this information and to generate a predictive score that can be combined with a formula (a rule, if you like) to return a pick list value. This pick list value is stored in salesforce.com and drives subsequent salesforce.com behavior.
For instance, a contact from a prospect might be analyzed to predict how valuable the contact might be and this predictive score is then the basis for setting the field used in salesforce.com to manage prospects to High, Medium or Low. This process can also set reason codes to explain how the score was calculated to help with subsequent actions. EpiAnalytics can use existing data to train the model (leads already scored by hand, for instance) but it also tracks behavior and feeds that into the system enabling it both to evolve models and to create models as part of collecting new data – a user can simply hook up their existing business process to the analytic engine and the system begins to learn at that point. EpiAnalytics estimate that 500 historical samples or so is enough for them to stop simply learning and start going into active mode so even those with low transaction volumes should be able to use it.
EpiAnalytics tightly integrate with salesforce.com and use as much standard reporting as possible to show customers how the engine is helping and to describe its ROI. As new versions of salesforce add support for more complex logic (such as the new version that has decision tree), EpiAnalytics takes advantage of this to allow for more complex decisions based on their analytic models.
EpiAnalytics have an approved salesforce.com data center and can currently handle customers with 10M transactions a month – this approach might only require 500 transactions to start being useful but they are clearly investing to ensure it scales too.