First Look: Saffron Technology

Saffron Technology is focused on prediction and sense making using associative memories. Associative Memories automate cognitive thinking by building connections and counts from large amounts of data. Saffron Technology uses a NoSQL / schema free approach and have developed technology to reason in real time by recalling the connections in the context of the raw data. They talk about “finding without knowing” and provide both a diagnostic component (that uses a search metaphor) and a prediction component (focused on the heavy lifting of building prediction from hybrid data). Their approach is to combine structured and unstructured data to predict, for instance, customer lifetime value and influencer revenue. They make predictions about each individual customer and use this to improve customer acquisition, customer retention and development.

Saffron’s technology has a number of critical elements:

• Structured and unstructured data are combined without curating the data. Very large analytic graphs get developed from the original data.
• It is Schema free and model free at the time of ingestion – the model is created using a machine-learning approach.
• The models support real-time learning, revising and updating as new data flows in.
• The engine “remembers” anything. It can in principal answer any question as no schema or model has to be developed in advance.

The core tool allows you to do diagnostic and predictive work, finding aliases for instance or doing root cause analysis for maintenance or customer churn. The engine is also transactional though, allowing you to “score” specific transactions (events) using a REST API for instance to predict maintenance problems or fraud likelihood.

Architecturally Saffron’s product has a number of layers

• SaffronAdmin allows both batch and incremental data ingestion (from databases, or social like twitter feeds) and performs natural language processing on unstructured data.
• SaffronMemoryBase builds entity connections (between entities in structured data and entences in unstructured data), maintains counts, builds and maintains the networks
• A set of RESTFul APIs allows access to the memory base
• SaffronAdvantage is a decision support tool built on these APIs to support investigation and the finding of interesting links or connections, and their change over time. This supports investigation of entities and entity ranking, similarity analysis, classification, network analysis, temporal trends and episodes.
• Other solutions like integration to case management can be built on top of the same APIs

Traditionally associative memories did not scale linearly and tended to explode in complexity. Saffron’s secret sauce is a solution that scales linearly and handles the potential explosion of complexity that results when, for instance, a million items must be cross referenced to a million items. The store itself understands semantic triples in an RDF-like way though without the need to define the structure of the triple first. This use of triples provides context for the networks and of counts that are kept up to date as more data flows in, and allows for finding correlations in hybrid data.

Saffron Technologies will be one of the vendors in the next version of our Decision Management Systems Platform Technologies Report and you can get more information on them here.

4/5/2013 Minor correction made