ASG is now part of Rocket Software! Please visit to learn more. Follow our journey
Blog > November 2018 > Offensive Data Strategy Trends

Offensive Data Strategy Trends

If you are a CDO working on your offensive data strategy, this blog series is for you.

Corporations investing in artificial intelligence (AI) – especially in areas like fraud detection, anti-money laundering (AML), chatbots and product recommendation engines – are dependent on secure, trustworthy datasets. Datasets that are governed, transparent and easily assessible. Gartner research indicates that by 2020, "85% of CIOs will be piloting artificial intelligence programs through a combination of buy, build and outsource efforts.

Successful AI is not about the technology that is deployed. It’s about the data that is used and the “end-result” of that data. The first focus needs to be: Are you answering the right questions, and are those questions aligned with the business problems and challenges your company has? Data scientists have PhDs, but do they really know your business? Are the data scientists’ projects aligned with your business objectives? Are they incented to work with business users? Is business logic being tied to the code?

agreement-arms-business-1081228.jpgWe have found that many machine learning (ML) models are built and extended in a “black box.”  If the results are fragmented or ambiguous, will your analytical users be able to easily trace the origins of the data used in the model? Will they know what rules and biases have been used? Or are we just relying on the Scientist Community? Tracing through the test data is essential.
Once the data is clean and understood, it's important to monitor and govern the information. A solution like ASG's Data Intelligence (ASG DI) categorizes your datasets and ensures that they are meaningful to the business. Collaboration features enable users (scientists and business analysts) to rate the data and track how the data is used and leveraged. If the insights are questionable, the end user can trace back to the originating systems and manage the data feeding the dataset.
For example: Fraud is a billion-dollar business and is increasing every year. In 2016, 15.4 million people experienced some sort of fraud – almost a 20% increase from 2015. Given the rise and advancements in ML, the industry “standard” approach has shifted from a rules-based fraud detection system to modern ML-based solutions.
ML models can easily scale to larger datasets and process them in real time. These models can automatically find hidden and implicit correlations in data – requiring minimum manual interaction and providing significant cost savings for businesses. However, despite all the advantages ML models provide, if not managed and governed properly, they will become another black box for business users.

How Do You Govern a Machine Learning Model?

The most important aspect of governance is the data used to train the model. Do you understand the data? How is the data collected? What is its quality? Its origins? Is the data representative of the real business cases? Do you have permission and consent to use the data for this purpose? ASG DI gives you visibility into the datasets – including data quality, data lineage, definitions and data distribution. ASG DI also ensures that data is in compliance with the ever-growing and strict privacy regulations.

How Do You Utilize the Machine Learning Model?



 Does the business understand how the model works? What assumptions have been made? What data elements does it evaluate? How is the actual model tied to the business rules? How will the output change as you change the model or training set?  
ASG’s Data Intelligence ties the business logic to the actual source code and ensures they remain in sync as the business requirements and model evolve. It’s dynamic. This provides significant competitive advantage for organizations.
ML models are not always predictable. The same input will not always generate the same output. So, how do you use machine learning models in a regulated environment? How do you monitor and benchmark the results as the models evolve?
We would love to hear your feedback and learn about your ML experience! Contact ASG on Twitter, LinkedIn or Facebook.