Why we Invested in Blockbrain

Knowledge management is an unsolved trillion dollar problem!

Knowledge is one of the biggest assets for many enterprises. It touches almost every aspect of a business – from knowing the right person internally, finding the best supplier externally, finding a problem with a machine or finding the best candidate for a position. A lot of information is stored nowhere, often only in the brains of experienced people. And even if it is stored it is often very difficult to find. Many enterprises hoped that the increasing digitalization and cloudification would solve this problem, but it did not. It just made the problem worse: nowadays every enterprise produces an enormous amount of digital data, almost impossible to organize and use. This ineffective management of knowledge leads to a productivity loss of $ 1.3trn and it is estimated that every enterprise loses on average 20% productivity (Source: McKinsey).

Traditionally, knowledge management was either solved manually (e.g. having an excel list of the right people to approach or an overview of relevant data) or solved with platform based approaches. Manual approaches only scale for smaller companies. Platform based approaches try to collect and index data either semi-automated or automated. They are typically very hard to retrofit within a large company since there is typically still (quite substantial) manual tasks involved. There is either manual labor needed to pre-categorize data before indexing. Or there is manual labor needed after the indexing to filter out wrongly categorized data. In addition, typically only knowledge in data is captured (=loss of knowledge from employees).

So how can this problem be solved? There is currently a new “cool kid around the block” – called large language models (LLMs). Almost everyone knows ChatGPT, Gemini and other LLMs. LLMs are great for summarizing, researching, coding…so why not use them to organize knowledge and data? There are simple reasons why this is a bad idea: first, LLMs are text models, they are not very well suited for images, audio, video… so which company has all its knowledge in text form? Second, they are costly! LLMs are priced per token analyzed. 1500 words are roughly 2048 tokens – more or less one Medium article. The price for ChatGPT’s latest model 4o is roughly $5 per 1 million tokens. This would result in $5 per 666 Medium articles. That sounds cheap but imagine how many pages would need to be analyzed in an enterprise. And this is only one search. Imagine thousands of employees querying millions of pages every day. Third, LLM’s are non-trustable. Have you ever tried to convince ChatGPT that the earth is flat? It is doable! Often times LLM’s face the problem of hallucinations, misinformation or even hate speech. Not the kind of solution that you want to use in your enterprise setting. In addition, there is a huge data security and privacy risk: 61% of business professionals are not trusting AI solutions due to that reason (Source: KPMG).

What now?

Blockbrain: secure and private knowledge bots

Blockbrain solves all the mentioned problems of traditional knowledge management and purely LLM based solutions. The solution consists of 4 main components. First a unique data pre processing pipeline retrieving, sorting and pre analyzing data of almost any format. Secondly, the data pipeline structures the data and creates a knowledge graph and makes the data highly accessible in different types of databases depending on the type of data stored (e.g. audio, video, text…). Third, a user facing knowledge bot that is made for a particular use case (e.g. sales automation, legal reviews…). This bot has several ways to retrieve data from the knowledge graph (and update that data with user input), depending on the user request and type of data needed. Data is then collected and forwarded to any LLM that is desired (e.g. Gemini, ChatGPT…). The LLMs only job is to format the data retrieved nicely. Lastly, there is a large emphasis on data privacy and security. The solution includes features like for example tokenization of sensitive data, or a unique approach to make the AI decision process transparent and automatically detect data drift.

The Blockbrain solution combines a “traditional” data pipeline and retrieval approach with LLMs. Why is this a good idea? First, it is way more cost effective: the LLM is only used for its initial purpose, being a text interface, not as analysis tool. This saves an enormous amount of tokens use by the LLM and lowers the costs. Second, accuracy is way higher. When only using the LLM as text interface it does not have to deal with other data types it can’t properly understand – video, audio etc.. It reduces the risk of hallucinations and other malfunction of the model significantly. Third, the indexing of data is fully automated and the user only deals with the chat interface, solving his or her specific problem. Lastly, the solution implements rigid privacy and data security measures, preventing the upload of sensitive data to public LLMs.

And the best: since it’s so easy to use it can also retrieve know how from people directly. While interacting with the system, the knowledge graph is not only fed with hard data but also with user knowledge. Imagine the system actively asking how to treat customer X – and deriving a recipe for future employees.

Aren’t there other similar solutions in the space?

Yes, there are other solutions. Let’s try to segment them a bit. There are traditional knowledge management solutions out there with the problems mentioned in the beginning. There are also many LLM based solutions out there. In our opinion many of them are simple “wrappers”, offering a nice user interface and simply calling the APIs of existing LLMs. This does not solve the problem – it still has the same problems as the LLM itself. Data formats ingested, data privacy and security etc.. Obviously there is also direct competition, however many of these companies come from the US, typically with lower data privacy and security standards. We know that especially for scaling in Europe, a EU headquartered company is very advantageous if not necessary. To get a better idea for competing solutions, we benchmarked many of them with a standardized test procedure. Our test clearly showed the outstanding quality and ease of implementation of Blockbrains solution. None of the competitors got even close to their results. Obviously we don’t know the roadmaps of the – often well-funded – competition, but we value Blockbrains current solution and their ambitious roadmap. Adding to that we believe that this market will not be a “winner takes it all” but rather a market with multiple “winners”.

Our investment thesis

Blockbrain convinced us from 3 angles. First and foremost the outstanding team. The company is led by 3 co-founders. The CEO, Antonius (“Toni”) Gress has extensive experience in the European market, especially with medium sized corporates. Currently the core market for Blockbrain. Toni shines as energetic doer. Toni is complemented by Honza Ngo, managing operations – a young talent with VC background and an excellent HR network in south east Asia. Mattias Protzmann, the third co-founder, leads strategy and vision of the company. He is a serial entrepreneur with 4 successful exits. One of his former companies include the well known statistics platform “Statista”. Statista shows parallels to the current solution of Blockbrain: analyzing and structuring fragmented data and making sense of it. Mattias is outstanding in shaping the vision and steering the company towards success.

Secondly, we believe that market timing is perfect now. Many companies are seeking AI solutions to improve their efficiency – a growing and increasingly large market. Many of the especially European companies however, fear data privacy and security issues. This can be addressed perfectly by Germany based Blockbrain, emphasizing data privacy and security. Blockbrain follows a “land & expand” approach with prepackaged use cases. This helps with rapid implementation, value can be shown almost immediately.

Lastly, we see a clear exit path for Blockbrain: first, big tech being very active in the space for now mainly investing either in their own LLMs or in private companies like OpenAI. Adding a (European) use case layer can be a logical next step for them. Second, companies like OpenAI themselves. We see them focusing on LLM development and simple use cases. Adding an enterprise grade implementation would make sense. Lastly, other knowledge management companies consolidating the market. Leading players are highly acquisitive.