AI(Artificial Intelligence) on BI(Business Intelligence): An inevitable entity
Traditional BI or big data BI is all about industry’s domain knowledge being converted by programmer into data insights for businesses. The business understands the data insights from combined knowledge of its own subject matter experts (SMEs). This significantly limits the insight in traditional BI to handful of human imaginations. In contrast, AI on BI with its deep learning capability is machine dependent with unlimited imagination power. The insights of AI on BI are from neural network layers of AI. Models in AI are trained with synonym and antonym data sets from heterogenous data sources. AI on BI insights are not only descriptive but could also be predictive in nature.
When it comes to AI, it’s not new to the world that we understand. AI in a way has been known as computer aided system from 90’s and is still being used in a lot of businesses. A small improvement in computer aided system algorithm was giving businesses 10-15% of profit margin. AI now in new avatar is not an incremental change but a gigantic shift for businesses. AI is disruptive. Ignoring AI would be nightmare for businesses. Business insights using AI on BI is a fundamental opportunity for businesses to seek dark data insights, which may not be fully human articulative. Possibilities of deeper pattern identification, classification and segmentation using AI is a great value proposition to any business.
AI on BI adoption is pervasive and for many businesses adoption has significantly resulted in major profit margins. Businesses in US and China are early adopters of AI on BI. The AI on BI adoption in regulated industries such as banking, security, healthcare is more fulfilling when compared to other industry segments. AI on BI brings precision and personalization to the business solution with its significantly rich features.
Challenges and possible approaches:
1. Challenge with AI on BI is its inability to explain underlying reason for a business event. Why such event occurred may still be AI’s “I don’t know” area in few segments. For example: AI can help in recognizing fraud / predicting fraud but may fall short on identifying the reasons why such fraud happened. Perhaps this segment of AI is what is still in lab research.
2. Data used for training AI can be biased. The notion that AI learns from data means that skewed data can result in biased insights. The solution to biased insights is to make data collection an independent unbiased activity. Train systems with synonym and antonym data (positive and negative data set) and tune AI system on how to react to negative data set. For example – how AI should react to unrelated obscene conversation. The thumb rule in identifying the right path for AI unbiased system could be to validate whether insights reinforces the best practices of the industry segment.
3. Data collection itself is elaborate and sometimes painful phase in AI enabled systems. The likelihood of having no historical/decision support data can result in stereotyping the business solution. The possible approach for no data AI on BI system is not assuming data but to transition system from traditional BI to AI on BI system. Once again, the thumb rule in identifying the right transition path could be to make insights human articulative to machine articulative while corpus data is collected.
Business need not resist the disruption but try to be on right side of business disruption by embracing AI on BI. This may mean a fundamental change in business culture and priorities for few businesses. A holistic approach for AI on BI is needed that would help businesses to land safely during this healthy AI universal shift.
Happy AI