The Company

thingsTHINKING was founded in 2017 by a four-person research team from the Karlsruhe Institute of Technology and remains located in that city in southwest Germany. CEO Sven Körner is one of the original founding team members, with the remainder holding senior positions. thingsTHINKING raised €4.5 million from Earlybird Venture Capital in May 2021, and we currently estimate the company’s revenues are sub €3 million.

The Technology

thingsTHINKING’s core platform is Semantha, which is designed to provide nearly out-of-the-box processing for text- and data-heavy documents, with the company suggesting that it requires a “handful” of additional documents as examples for new customers to onboard.

As its center, Semantha assembles what thingsTHINKING refers to as a “Semantic Fingerprint”: a multimodal collection of specific identifiers from word embeddings and document layout/formatting feeding into the pre-trained language models and graph/knowledge bases. The result is a meta fingerprint that combines the output of this variety of approaches into a single common exchange format (see Figure 1).

This fingerprint can then be used for a couple of core use cases; document comparison (Semantha Compare) and document validation (Semantha Requirements). Comparison is designed to sit within an automation where, for example, newly arrived contracts can be quickly verified against previously accepted/approved versions, so that any change deltas can be routed for attention, but those without change can pass without automatic checks. Validation use cases include being able to check requirements documents to see if there are new elements that require human intervention (for example within a Requirements Management tool). These results can be fed back into Semantha’s knowledge base to improve subsequent tasks.

At a deeper level, the Semantic Fingerprint itself is an aggregation of smaller fingerprints from within the document at paragraph and sentence level. Of note here is the ability to determine close but negative matches for similarity; for example, “water soluble” vs. “not water soluble,” where the operator entirely flips the meaning of the sentence and therefore the overall meaning of that section of the document.

Using the default models allied with what has been added to Semantha in terms of customer-specific domain knowledge, thingsTHINKING envisages this being packaged as a “Co Worker”: a packaged app that can be utilized in other parts of the organization to extend the use of the platform to a greater range of employee tasks (and, in parallel, the reach of Semantha within that organization).

As such, thingsTHINKING sees Semantha as being a critical point of augmentation in existing processes and has enabled the platform with a REST API complete with more than 150 preset connectors. The company is building out a partner network to help locate and integrate such opportunities. Their alliance with RPA vendor UiPath makes perfect sense as a starting point for intelligence to be added to process bots; so do integration partnerships with Accenture and SVA to enable scaling of the necessary services to grow their initial customer base.

At present, thingsTHINKING says its customer base is predominantly a mix of automotive (including Hella) and insurance/legal (including Heidelberger Volksbank and Gothaer Versicherung). However, the company intends to grow from that vertical base toward broader connections to enterprise platforms – and employee experiences – using that common exchange format as a vehicle to help share Semantha’s multi-modular output to a greater range of use cases. To assist this widened scope, video and audio embeddings will be added to the platform over the next year.