In my previous post, I shared some Artificial Intelligence (AI) use cases for the retail industry. While the smartness of these systems promises to transform customer experience, there is also an equally significant responsibility for retailers to ensure these smart systems are safe and secure for customers to trust. To succeed with AI, retailers need to augment their QA function with robust and intelligent testing approaches.
The QA case for AI is clearly different from the traditional user input system response type of testing. Traditional IT systems are tested by verifying a definite system output against a well-defined set of inputs. But AI and cognitive systems with machine learning and predictive analytics exhibit abilities to understand given inputs contextually, generate non-deterministic output based on a network of algorithms. This difference renders traditional testing ineffective, and mandates flexible and intelligent testing methods that are best suited for AI.
According to TCS latest Global Trend Study titled Getting Smarter by the Day, success with AI depends on four critical factors – security of systems, systems ability to continuously learn, good and safe automated decision making, and employee acceptance of AI insights and decisions. What does this mean for QA teams in large retail businesses? In line with the above, retail businesses need next-gen QA function for AI – one that assures systems ability to Assimilate, Process & React and Learn. Lets understand this bit by bit.
Assimilate: Cognitive systems receive input from multiple sources – structured data from databases, unstructured data from various devices, social media and even IoT sensors. With the blurring of lines between digital and physical channels, customer preference and behavior data and usage patterns are now captured at a number of customer touch points – from social media to digital apps. While this data volume presents an ocean of opportunities to retailers, its important that the data be tested. There is no point if the retailer is basing customer and market data on skewed or incorrect data.
AI data sources must be tested at their capture points, for functionality, compatibility, performance and security. Given the complexity of the incoming data, its also important to validate data for accuracy and right format, before it is consumed by AI systems. Checks and balances at entry include profiling the categories of input data from multiple sources, identifying any synchronization issues across different data sources. In addition, test data models have to be defined covering real life best cases, average cases, exceptions and negative cases as well. Ensuring data privacy and security also falls under the QA gamut.
Process and React: Cognitive systems can think almost like humans, however their processing capability in volume is much beyond the average human brain. The decision making ability of AI systems comes from their ability to replicate near human behavior and logic from a network of algorithms, hence react like how a human would possibly do.
As there is no defined output that can be verified for a given set of inputs, AI test strategies must often employ statistical methods to determine the acceptable level of accuracy in the AI process. During festival and peak seasons, the most critical focus is to ensure right inventory levels. Also, smart supply chain systems can predict procurement delays while automated workflows and robot assisted warehouses (as mentioned in the Amazon Go example in the previous post), can ensure optimal inventory levels for retailers. To assure such smart use cases, QA teams must define and agree on acceptable level and conditions first say, 95% accuracy in predicting inventory levels, then feed the AI system with a defined set of data patterns such as combination of weather, seasonal events, customer behaviors, and buying patterns. Finally QA must capture and analyze results against the acceptable levels, noting and analyzing anomalies and exceptions in detail.
Learn: Learning is at the heart of AI. With ability to learn and refine their abilities, AI systems can improve over time, on their own. But these systems must be trained to learn and the learning process too, must be validated over time making QAs intervention mandatory and inevitable. Besides providing inputs for supervised learning, QA must also monitor the unsupervised machine learning processes.
The AI-retail integration promises to power up customer experience to unmatched levels personalized and voice assisted gift suggestions, virtual dressing rooms, and shopping assistant robots are just the tip of the iceberg. Ecommerce disrupted the retail brick-n-mortar model and ushered in the online retail revolution. AI too will disrupt customer experience, and this next retail revolution has just begun. It is our responsibility as QA professionals to assure an exciting and rewarding journey for retailers.