I’m appearing on a panel at the upcoming Semantic Layer Summit, a first of its kind industry event, and I’m excited. Industry panels are fun and Stardog is a leading player in the semantic layer space, having inaugurated the category last fall within the Databricks partner universe.
But more importantly Stardog has a unique approach to Semantic Layer for two reasons.
- Stardog is the leading knowledge graph-powered semantic layer. All other semantic layer solutions are based on the relational data model and SQL.
- A Stardog semantic layer is the best choice for a new, expanded set of use cases for the enterprise.
Relational semantic layers, which just aren’t that semantic from my perspective, are really good at high-speed data aggregates and counting things that fit neatly into relational cubes. In other words, operationally they’re a kind of just-in-time data warehouse. And that’s great! Really. Many Stardog customers query their relational semantic layer using Stardog.
Which leads to the first main observation of this piece: the “metrics layer” – i.e., dynamically computing the core sales and marketing KPIs that are fairly similar across most enterprises – is the oldest and foundational semantic layer use case. It’s also fairly generic; at least, the “semantics” in question aren’t really about what distinguishes enterprises but more like what most of them have in common. In fact the “metrics layer” and “semantic layer” are really just two different ways to say the same thing typically.
But not in every case! Which leads to my second main observation of this piece: there’s a new breed of semantic layer use cases that Stardog and its customers are pioneering. These use cases include:
- Know Your Customer (KYC), AML, risk and compliance analytics in financial services
- Molecule-to-market, preclinical R&D, and supply chain control tower in life sciences
- Digital Twin, Digital Thread, and Product 360 in manufacturing
Are These Really Semantic Layer Use Cases?
These are semantic layer use cases now because a Stardog-powered semantic layer is a different kind of beast. Different because Stardog is a unique Knowledge Graph platform that deeply integrates a semantic graph data model with powerful graph query answering and with both fast, scalable Datalog (i.e., business rules) and statistical inference, as well as a sophisticated data virtualization and query federation capability.
All of which is very well-suited for the wide-range of new semantic layer use cases, beyond the conventional “metrics layer”, which are primarily concerned with connecting enterprise data based on business meaning irrespective of location in the data sprawl of the hybrid multicloud.
These new semantic layer use cases share some key characteristics, too:
- They’re dominated by business questions that require instance data connectivity rather than aggregate data
- Put another way, they’re dominated by business questions about causality which are represented as lineages and pathways rather than tables of aggregated numbers
- For example in Big Pharma, “find a path from a preclinical R&D study to some batch management records for drug X and then on to the commercial distribution of drug X where some people in Indiana had adverse reactions to X”
- They’re perniciously cross-domain with respect to enterprise lines of business and with respect to data type, i.e., structured, semistructured, and unstructured
- They’re comprised of multiple hierarchies of data that intersect more or less randomly (i.e., it’s not really random intersection but very random-seeming)
- In manufacturing, for example, to support the “stop shipment” function of a digital twin-and-thread platform, we probably need to look at some finance, product configuration, raw field sensor data, as well as, say, dealer field reports of Q&A faults.
- They’re richly connective with a lot of relationship complexity between entities in a way that puts a lot of pressure on the very sparse semantics of unidirectional foreign keys between tables in an RDBMS
- For a variety of legacy reasons, and the properties of the data described above
Maybe it’s easiest to say that these enterprise data sources are very challenging to fit into relational data models which, as a result, makes most semantic layer solutions not an ideal fit for unifying over and across them.
The bottom line looks like this from Stardog’s point of view: a metrics layer is about doing math with physically distributed sales-and-marketing data; but the new use cases described above are about making connections across data silos to answer the enterprise’s hardest questions.
Our customers, together with our strategic tech and service alliances, have partnered with us to blaze a new trail for semantic layer adoption and expansion across the enterprise, and I look forward to representing them and our shared perspective at the upcoming Semantic Layer Summit.
Because the semantic layer isn’t just for metrics anymore!