Kontomatik Presents at FinDEVr London 2017


July 18, 2017
Kontomatik Presents at FinDEVr London 2017

This June our team has attended one of the leading European events dedicated to the development of the financial apps – FinDEVr 2017 in London. Previously we have spoken at Finovate and FinDEVr events about banking APIs and their integration, mobile onboarding of the customers, and this time we decided to take a completely different approach. Fintech seems to make lots of noise, but there is almost no sense behind the vast majority of the buzz words. This is why our CTO, Piotr Włodarek, decided to go ahead and present a Fintech dictionary that helps to understand the world of Fintech the right way.

Well, enough said. Click the Play button below and check it out.

Below we will provide the text of the Key slides.

Big Data => Small Data

  • Whenever you hear a “big data” think Excel spreadsheet.
  • Most “big data” companies process thousands or millions of data records. That may sound like a lot for a layman but your home Excel spreadsheet supports that scale. This is not a big data.
  • Tip 1: Avoid “big-data” technical solutions unless your scale is in billions.
  • Tip 2: Contrary to popular belief for many tasks you do not need big data to have great machine learning results.
  • Tip 3: Big data is big trouble (security, performance, …). Start with small data.

Kontomatik reads from banks ~1m financial transactions daily w/o using any “big data” solutions.

AI => Algorithm

  • Most companies claiming “AI” efforts are actually building classic programs with manually created decision trees
  • It’s just that “AI” sells much better these days
  • Whenever you see “AI” in a text, replace it with “algorithm”

Kontomatik cleanses and normalizes banking data across the world using non-AI algorithms.

Machine Learning => If-Else

Whenever you see “machine learning” read it as “if-else rules”.

Tip: if company claims “AI / ML” solution, ask for technical details:

  • specific ML algorithm employed
  • basic technical configuration for the algorithm
  • how data is cleansed and prepared
  • how they test and iterate

When faced with secrecy, you will know the answer: there is no significant machine learning involved. The most effective machine learning algorithms are all public: https://en.wikipedia.org/wiki/Machine_learning

Kontomatik actually uses machine learning 🙂

Schemaless DB => Implicit Schema DB

  • Schemaless databases allow for arbitrary structure data to be saved ad-hoc, claiming enhanced flexibility
  • The issue with that is that schema is still there – implicit and very hard to infer. When you need to migrate data to other structure you are in left the dark. Risk of breaking things is much higher than with explicit schema enforced by database
  • In your source code, whenever you refer to specific key name or assume specific data type, you are implicitly assuming the schema
  • Tip: Go schemaless only when can treat data structure in a fully generic way through your whole tech stack – and then use bjson type in PostgreSQL or similar

Kontomatik uses PostgreSQL database and enjoys full control over the shape of data by employing explicit data schema

NoSQL database => Inconsistent database

  • NoSQL databases “scale better” by relaxing consistency guarantees.
  • ACID transactions are replaced with “eventual consistency”. The thing is, between now and “eventual”, things are likely inconsistent, and you must write lots of code to deal with that. Reasoning about the system gets complex and things get buggy very quickly.
  • NoSQL databases give significantly fewer guarantees and less power to the programmer.
  • Tip: your good old RDBMS scales much better than you assume. In 99.9% cases, you will not need to drop down to the lower-level NoSQL databases. Consider NoSQL a C of databases – when was the last time you had to drop down to C?

Kontomatik avoids NoSQL hype and enjoys data consistency guarantees provided by ACID database transactions.

Bitcoin => Cash-as-Information

Tip: probably the easiest way to think about Bitcoin is this: cash-as-information.

Enterprise Blockchain => Distributed Database

  • Inverted properties of the Bitcoin Blockchain
  • Whatever good you heard about Bitcoin is not present in the enterprise blockchains
  • Distributed database
    • Centrally governed
    • Trust based
    • Permissioned
  • Big overpromise

FinTech Dictionary

  • Read big data as small data
  • Read AI as algorithm
  • Read Machine Learning as if-else rules
  • Read NoSQL as inconsistent
  • Read schemaless as implicit schema
  • Read Bitcoin as cash-as-information
  • Read Enterprise Blockchain as boring distributed database
  • Read Kontomatik as the quality banking API