The Case for Semantic Technology in Financial Services

In my last post I discussed some of the common use cases of non-relational database technology within our industry. This time I’d like to turn to the use of a specific kind of non-relational technology, namely Semantic Technology.

Semantic Technology is specific set of data management tools that aim to provide meaning and context to the data. This is achieved by defining facts about the data, in the form of a Subject-Predicate-Object triple, e.g.: John Smith –> marriedTo –> Jane Doe. These facts can be asserted as they’re discovered, without the need to fit them into any sort of a rigid schema. So, if we later find out that John has a brother named Jim, we can add that fact without the need to have a sibling table in place: John Smith -> brotherOf -> Jim Smith. Furthermore, we can now infer that Jim Smith -> brotherInLawOf -> Jane Dow, and immediately add that piece of information into the database.

This is a very brief high level overview, but it should give you a sense of this technology, and the reason many people are excited about it. For the rest of this post I’d like to focus on the use cases that this technology addresses, and the ways its unique features can provide business value around them.

Customer 360: You probably guessed from the example above that this might be a use case for semantic technology. The ability to store data about a customer as the data is discovered and ingested from a myriad of diverse sources, without having to go through extended, costly ETL cycles, is a key benefit of this technology. This is especially true when some of this data is non-relational (e.g. onboarding documents, communication records, etc.), and would therefore be quite difficult to stuff into a relational database.

Data Provenance: Due to the increased focus on data governance and regulatory compliance in recent years, there’s a growing need to capture the provenance and lineage of data as it goes through its various transformation and changes throughout its lifecycle. Semantic triples provide an excellent mechanism for capturing this information right along with the data it describes. A record representing a trade for instance, can be “decorated” with information about the source of the different elements within it (e.g.: Cash Flow -> wasAttributedTo -> System 123). And this information can be continuously updated as the trade record changes over time, again without the constraints of a schema, which would have made this impossible.

Reference Data: Somewhat related to the example above, reference data management can also benefit from semantic technology, by modeling the connections between instruments and legal entities associated with them using triples. Here the richness of semantics as a way to model the real world is key. For instance, modeling the complex relationship between a mortgage backed security and the derivatives built on top of it, or the relationships between legal entities that are affected by M&A activity, can be nightmarish using entity-relational models. Semantic triples represent a much more agile and flexible way to capture facts such as Smith Barney -> acquiredBy -> Morgan Stanley, or CDS_123 -> wasDerivedFrom -> MBS_xyz.

The Enterprise Data Management Council has been developing such a semantic model to represent the reference data universe. It’s called the Financial Instruments Business Ontology, and you can find more information about it at www.edmcouncil.org/financialbusiness.

Pre-Trade Analytics and Decision Support: Extracting facts from free-form text is an important aspect of providing information to traders and other decision makers, weather the text comes from news feeds, syndicated research articles, tweets, or any other unstructured source. In this case the facts contained in semantic triples represent the context of the unstructured text, e.g.: article123 -> mentions -> Apple Inc. But it can also go further to represent aspects such as the sentiment in the text: article123 -> isBullishOn -> Apple, Inc. This is done using sophisticated tools that can extract facts from the text, so that it can become immediately actionable, without a human first sifting through it.

Compliance: Regulatory legal text can be quite difficult to understand, and tracking the policies that would satisfy the regulations can also be an onerous task. Here again, semantic technology and its ability to analyze text and establish relationships within it can provide a huge benefit. The network of rules within the regulatory text, and the policies that satisfy the regulation, can all be represented by a semantic model (e.g. Form W9 ->satisfies -> IRS Requirement xyz). And this model can keep evolving with the policy and regulatory changes,. Thus workflows that are affected by these policies can be automated, alleviating the need to manually check each step. The onboarding process of a new firm for example, is governed by an exorbitant amount of regulations and internal policies that map to them. By using a sematic model to capture these, the onboarding process can be fully automated, and become much more efficient and expedited. This dramatically improves both the customer experience and the cost associated with the onboarding process.