What is OwlDQ
What is OWL?
​
OwlDQ is an intelligent service that automates data quality using ML and data science to profile and cleanse a data set. Owl was designed with 4 KPI's in mind:
Speed, Scale, limited maintenance, and blind spots. This will assure that data is valid, complete, accurate, and consistent. Data Quality is derived from intuitive clicks, not unnecessary code. Our goal is to deliver DQ in minutes not days or weeks, on a simple interface available to a wide range of business users.
​
​
​
OwlDQ core components - Owl provides a data quality assessment that scans 9 dimensions of a data set to assure the integrity of that data. The 9 dimensions are behavior, rules, outliers, pattern, source, record, schema, duplicates, and shapes. Owl offers a full DQ suite to cover the unique challenges of each dataset. Complete coverage and consistency drives trust. A single scoring and reporting framework with 9 features that can be activated in a custom DQ pipeline. Owl is horizontally scalable, it can scan data from any location and scale as the data set grows in size and complexity. Data quality needs to be abstracted from data ingestion for management to have a single normalized view of data health. Owl will scan your data with the same frequency, that you load your data.
Owl scans 9 dimensions of DQ out of the box
Behavior - imagine a column going null, automatic row count checks - does you data behave/look/feel the same way it has in the past
Rules - assures only values compliant with your data rules are allowed within a data object
Outliers - data point that differs significantly from other observations
Pattern - recognizing relevant patterns between data examples
Source- validating source to target accuracy
Record - deltas for a given column
Schema - columns drooped
Dupes - fuzzy matching to identify entries that have been added multiple times with similar but not exact detail
Shapes - infrequent formats
​
Use cases
FX Rates (Outlier Detection)
​https://docs.owl-analytics.com/use-cases/financial-fxrate-data​
Reference Data (Pattern Detection)
​https://docs.owl-analytics.com/use-cases/security-reference-data​
Cloud Migrations (Validate Source)
​https://docs.owl-analytics.com/use-cases/copying-or-moving-data​
Wealth Management (De-duping)
​https://docs.owl-analytics.com/dq-visuals/duplicates​
Buyers & Users
Primary buyers and end-users of OWLDQ (alignment, where applicable, to use cases)
Strategy, Analytics & Deposit Products
Consumer Credit Risk Management and Loss Forecasting
Data Analytics, Business Intelligence and Data Governance Practice
Commercial Banking Data Analytics
Data Analytics Strategy & Planning - Global Markets Technology
Business Analytics / Automation
Data Quality and Data Governance
Enterprise Data Management
Cloud Architecture / Big Data Architecture
​
Competitive Landscape
In order to avoid getting lost in the latest marketing jargon, a fundamental description is provided under each of the 4 stages. There are many ways to ingest and transform data; the descriptions are not meant to be exhaustive. Imagine a scenario where data is loaded in either a batch or stream, then joined to another dataset with some column transformations, and finally made viewable in a BI tool for consumption. But what about quality? What checks and verifications are in place to guarantee data accuracy and completeness? After all, showing someone a housing report with incorrect estimated housing values or a stock report with the wrong stock prices won’t go over well. Figure 2 below has popular company logos overlaid in each stage to bring more context to the discussion. There are easily 30+ software companies in each of the 4 stages.
Background:
OwlDQ spent about a decade on wall street dealing with DQ problems. We used rules based tools like IDQ. And ultimately found that we were spending a lot of time writing rules, that were hard to maintain and did not scale. We wanted to change those economics. My group in particular was a team of data scientists, back then they called us quants and we built models to catch things like spoofing, ramping, front-running and insider trading. We really just wanted a simple mechanism to tell us when things are going wrong. We decided to start using the phd level data science that we were applying to our models, but instead apply it to the data itself. We found that using behavioral analytics we could remove 60% of the rules but also remain adaptive.
Simple Scenario:
So, a really simple scenario is, let’s say you have a dataset that has 1 Million rows. We could write a rule that checks to see if it has less than 1 Million rows. But by next week it grows to 1.1 million and 6 weeks later it grows to 1.5M. Then suddenly it drops to 1M. The rule still passes but 30% of your data is missing.