JT on EDM

Attensity is a text analytics company based in Silicon Valley and Utah that is focused on helping companies hear the “voice of the customer”. By allowing companies to effectively process the text in emails, service logs, call notes and, more recently, forums, blogs and wikis, Attensity aims to make companies both aware of their customers’ sentiment and able to act on it in a timely fashion. Companies using their product often find they get an early warning of problems or that they can find “cries for help” buried in the text.

Attensity uses a combination of statistics-based text analysis and linguistic processing to do what they call “exhaustive extraction”. While they can use lists of entities (locations, products etc), the engine will also do entity extraction and disambiguation. Interestingly it also handles pronouns – what does “it” mean in this blog post, for instance. They also handle abbreviations and colloquialisms and provide an interactive environment for analysts to manage these collections of names for the same thing.

Most unstructured data, it turns out, is written in the first person and is rich, specific and personal. By allowing this to be analyzed and then integrated with structured data, Attensity helps form a true 360degree view. In general Attensity is used to add new data and understanding to existing data warehouse and analytic infrastructure – often partnering with Teradata or HP, for instance.

Attensity’s core product is an environment for analysts to explore the results of the analysis using all the available approaches – root cause, search, drill down etc. Attensity also has templates for customer sentiment, faults, dashboards, product launch, early warning, defect detection and other classic cases. Some of these are further specialized for financial services, travel and transportation and hard goods. When working with companies already familiar with the power of executable analytics – the kind of thing we talk about in EDM – their engine produces “pseudo-variables” that can have a value, such as a score, created for each block of text analyzed. Thus the riskiness of the words in an accident description for claims could be represented as a number. While most of their customers are using Attensity’s product in a BI/reporting/query environment, perhaps 30% are already using it to drive automated decision making.

Attensity’s approach of using multiple analysis techniques for text and then combining this with structured data is one I find particularly appealing.