I got a chance to catch up with Attensity after a long absence recently (I last blogged about Attensity in 2008). Attensity has been doing text analytics and customer experience monitoring for 10 years or more. Their approach includesfour steps – Listen, Analyze, Relate, Act (from a decision management perspective it is nice to see them have an Act element rather than just presenting data). Attensity’s product is used in customer care, marketing, operations and more.
I got a chance to catch up with Attensity after a long absence recently (I last blogged about Attensity in 2008). Attensity has been doing text analytics and customer experience monitoring for 10 years or more. Their approach includesfour steps – Listen, Analyze, Relate, Act (from a decision management perspective it is nice to see them have an Act element rather than just presenting data). Attensity’s product is used in customer care, marketing, operations and more. Some customers use the analytics to present information about what people are saying to people while others embed this analysis (often combined with other kinds of prediction) in systems to automate the decision-making.
Analyze and Respond are the two core products that support these four steps. Both are based on Attensity’s Information Access Suite with its Semantic Server and KnowledgeBase. This access suite pulls in data from a wide range of sources – internal files or databases, emails or text messages, social media sources and more.
- Analyze generates reports and dashboards that can be pushed to people proactively as well accessed on demand. Even this tool supports Action, however, as the segments or indicators like “would not recommend” that Attensity identifies can be combined with other factors in predictive models.
- Respond is a multi-channel response application based on the text analytics – one UI for social media response/queues and one for multi-channel response including calls/sms/email/social. The former being typically used by social media managers, the latter by call center agents. Customers can define the workflow and rules for what to do and different queues are set up based on categories so that customers can allocate actions to people by queue/specialty. Workflows get created to assign results to the various queues using rules that act on data elements and indicators like the category of inquiry that were derived from the text. Anything about the customer profile can be included (social media profile, impact and reach etc) as well as CRM system data (Whirlpool for instance brings in products owned, past service requests, warranties etc). Easy to integrate, Respond has many interfaces for integration with workflow and contact centers.
One of my areas of interest is the combination of text analytics with other kinds of analytics in decisioning systems (as I wrote about in What, Really, is the Value of (Text) Analytics? Attensity tell me they have been feeding the indicators derived from their text analytics into predictive models for years. There is an export option that takes insights – indicators – and generates a binary flag for each (does this record have this indicator). This dataset can easily be combined with other datasets in an analytic model so your risk score or propensity to renew can reflect not just your structured data but also what your customers have said in their emails to you or on Facebook.
The current release, 5.5, has a particular focus on massive scalability. With 75M web sources and an explosion of social media monitoring, customers had a need, obviously for scalability. Add in some very large customers who want to do longitudinal analysis over many years of data and you get a need for really large scale analytic infrastructure. Release 5.5 (both SaaS and for onsite use) uses a grid infrastructure to deliver massive scalability, standard database interfaces and reduced I/O and cost of storage. An MPP/Column store solution it focuses on executing the queries from Attensity Analyze and scales to hundreds of TBs of text data. Attensity’s own tests suggest it works at 5x to 180x relative to access data in the RDBMS vendors they support while managing 5x-50x the data.