Technology is disrupting business processes across industries at an unprecedented and exponential scale, with several business domains now at the tipping point. Businesses that successfully adopt Artificial Intelligence (AI) achieve tremendous competitive advantage – it is estimated that organizations that extensively use Big Data and Analytics are likely to significantly outperform the market; they will steal $1.2 trillion per annum from their less informed peers by 2020.
At Wipro, we have been working closely with our customers around this imperative. Over the last two years, we conceptualized, created, and then deployed our Fraud and Anomaly detection platform based on Big Data and Machine Learning. Numerous interactions with customer stakeholders, engineering teams and business process owners have provided us with insights that are relevant for Machine Learning implementations across business contexts. We advocate 4 fundamental lessons for organizations making the shift to data-driven decision making.
Change in mindset
Leaders need to commit and buy into the vision to become data-driven, and help set the right expectations with the management. A key issue is managing the organization’s existing culture with practices likely to get impacted with the new approach and shift from gut-feel and eminence-based decision-making to a data-driven model.
For instance, we were working with an asset management firm to identify data quality anomalies in fixed income securities master data and were able to impress the in-house quants with the platform results. The core challenge remained in communicating how we achieved the results without much context into bond pricing and correlation between instruments in derivative pricing.
Not a magic wand
It is also important to note that while Machine Learning allows one to decipher patterns and understand signals better than traditional heuristics, it should not be seen as a silver bullet to completely handle the business problem. Interpreting the results in a manner that aligns with business objectives will make them a lot more meaningful. For example, in a Holmes deployment at a leading insurer for identifying claims fraud, the Head of Claims appreciated that integration of fraud detection results with underwriting process is key to unlocking value – be it differentiation on prices, better evaluation or optimizing turn-around-time.
Complexities of data handling
Often Machine Learning is equated to rocket science and that is partially true! But while focusing on the complexity of the algorithms it is easy to lose sight of the fact that most Machine Learning problems need a foundation of data to deliver results. A key challenge is in identifying, sourcing and managing the right data streams. Tackling dirty, incomplete, and erroneous data needs to be factored into the data handling process and into the results analysis.
In our deployments in the workforce compliance area, (including detecting credential compromise and employee impersonation) we had a need to handle log data (IP logs, physical swipe logs and other digital logs) to the tune of 100 million records on a daily basis. Log data typically is extremely messy with fields and formats that are inconsistent across systems and organizations. Obtaining all the data from different teams, cleansing and transforming it to make it usable is an important exercise that needs to be factored in.
More than just Automation
For far too long, technology has been used as a means for automation, making certain parts in the existing process faster or error-free or more efficient. Artificial Intelligence provides us with the opportunity to use technology for re-imagining the entire process from scratch.
Let me take the example of Wipro Holmes expense claims management platform. We used AI to re-invent the process: with expense reports that write themselves, and need only one-touch validation from the user (employee). The AI powered anomaly detection engine flags any suspicious claims, resulting in targeted interventions – the manager is equipped to act instead of ignorantly approving all expenses. In addition to significant improvement in cycle time, user experience is at the forefront of this change; taking the pain out of expense filing for employees.
While businesses need to work through the change management process and bring in new skill-sets relevant to leveraging machine learning, we expect the rewards from employing Machine Learning to far outweigh the efforts in implementing it. We expect the possibilities for application of Artificial Intelligence will only multiply with value created in innovative newer opportunities. Organizations will need to have the right emphasis on change management and leadership endorsement; along with the crucial ability to integrate technology with business to achieve competitive differentiation.