Digital Asset #3 - Advanced Analytics Capabilities

Most organizations don’t need to be convinced about the huge importance of advanced analytics and AI to support their business to thrive. Combining the integrated data from an organization’s unified digital platform with powerful analysis tools creates an effective strategic advantage.

Advanced analytics capabilities support users to make more informed, precise decisions, more quickly, reducing uncertainty by delivering the right information at the right time, eliminating false positives and false negatives, and helping to distinguish weak signals from noise. The goal of building Advanced Analytics Capabilities is to create a process of continuous learning within an organization, as illustrated below.

Click to enlarge image

Assign outcomes, amplify with Artificial Intelligence
© https://www.raywang.org/blog/2016-09/mondays-musings-understand-spectrum-seven-artificial-intelligence-outcomes

This said, many organizations today struggle to harvest data and transform it into actionable insights. Those that do engage with analytics do so at varying levels of intensity. Some are only starting to make better use of the data available from their transactional systems to make more informed decisions and react more quickly to changes in their operations.

More advanced organizations are leveraging their integrated data to engage in more powerful analytics to determine what happened and to deliver more real-time insights into what is happening.

The most advanced are building predictive models about what’s going to happen, planning their actions accordingly to manipulate their desired outcome. These latter organizations are able to build ‘business graphs’ that describe individual customers’ context, in detail, at the point that customers make buying decisions.

Experience shows that building advanced analytics capabilities like these is difficult, requiring advanced skills and cultural change around the decision-making process within the target organization.

Establishing new ways of working is difficult. ‘Unlearning’ existing ways of working often proves much more difficult. To implement Advanced Analytics Capabilities, for example, it’s important to introduce a ‘data-culture’ that clearly distinguishes uncertainties created by ‘the unknown’ versus ‘the unknowable’. Too often, uncertain decisions are based on entrepreneurial intuition or ‘gut feeling’. These don’t involve unknowable uncertainties; they are simply unknown.

An effective data culture demands individuals to examine and reduce the ‘unknown’ by generating relevant insights through advanced analytics. If data is not readily available to answer relevant questions, then effort must be spent on finding the data (including externally), bringing it together and discovering new types of patterns through AI and analytics. This type of discipline is where an enterprise can build data supremacy over its competition.

• The combination of a unified digital platform and advanced analytics opens up new possibilities for organizations. Increased integration of BI and AI has matured to a point whereby BI consumers are asking for predictive AI capability within their analytics solutions, enabling them to review past, present and future.
• Conversational, semi structured BI approaches – including ThoughtSpot, Looker and AWS Quicksight – deliver analytics to users in a conversational fashion.
• Enterprise-wide analytics and the use of AI (including for operations) are now mainstream, supported by the foundation of modernized integrated cloud platforms and applications. These enable organizations to own and govern their own data. Next, we will see the integration of BI and AI, and more advanced ways of using AI to accelerate and improve the quality of development and QE, like data-driven engineering leveraging ML (‘DevAssist’).
• Algorithm performance is now optimized by reinforcement learning, a subfield of ML that is nearing maturity. Reinforcement learning deals with how machines learn to optimize their behavior in response to feedback. This type of learning is different from supervised and unsupervised learning in that it doesn't rely on predefined labels or categories for input data. Instead, reinforcement learning algorithms attempt to learn how to behave in a certain way by observing the consequences of their actions.
• Knowledge graphs are a way of representing data that connects entities. In a knowledge graph, entities are represented as nodes, and the connections between them are represented as edges. This allows for the easy identification of relationships between entities. For example, a knowledge graph could be used to represent the relationships between people, places and things. One of the benefits of using a knowledge graph is that it can help to improve search results. Additionally, knowledge graphs can be used to power other applications, such as voice assistants and recommendation engines.
• On the horizon is federated learning, a ML technique where multiple devices, often in different locations, collaborate to learn a shared model. Each device in a federated learning setup trains its own copy of the model, sending updates periodically to a central server. The central server then combines all the devices’ models into a single, final model. This technique is often used to reduce the amount of data that needs to be transferred between devices and to improve the accuracy of ML models.

AI is the most critical and promising driver for quality innovation today. By applying AI within quality and test activities, organizations can build truly intelligent test and validation systems.

Today, AI is used in a variety of ways for testing, including in smart analytics to assist with quality management decisions, test strategies and automatic test optimization, ML technologies for automatic test adaptation to application changes, Natural Language Processing (NLP) to generate test scenarios from user stories, and AI-based techniques for creating synthetic test data. The ultimate promise of AI is that within a few years, a large portion of testing will be generated and done entirely autonomously by self-adaptive and self-healing test bots capable of adapting to fundamental changes in applications. We will also see more intelligent solutions for provisioning of test data and test environments.

One area that still in its infancy is the validation of AI itself. As AI has advanced and become a de facto industry standard for performing and automating routine tasks, the barrier to widespread AI development and adoption is no longer the technology itself and the skillset required. Today, it’s the ‘human’ aspect of AI – including AI ethics, morality, transparency and governance. Corporations must now learn to identify and mitigate potential risks (such as discriminatory models) to avoid the generation of biased, unfair ML models. To mitigate this, it is critical to establish a quality framework to guide the development of ML solutions.

To handle the validation of AI successfully, Sogeti recently developed our QAIF (Quality for AI Framework) – a cohesive, generic quality framework that can be tailored to any AI solution. This is designed to assist product managers and business owners to identify and mitigate risks associated with each stage of the AI lifecycle. The framework is governed by the EU ethics principles: Fairness, Transparency, Accountability, Traceability and Robustness. With this model we can add quality gates to each AI development phase (Business Understanding, Data Understanding, Data Preparation, AI Model Development, AI Model Evaluation and AI Model Deployment) to ensure regular quality control checks.

Sogeti has also built practical tools for data scientists to utilize alongside QAIF, and can now offer AI Quality as a Service, infusing fairness and transparency into any AI model automatically. This should be a requirement, not a luxury, for businesses embracing AI technologies. These initiatives are supported by new quality approaches, like our specific quality frameworks for testing AI.

Elsewhere, AI is being used alongside Robotic Process Automation (RPA) to create the Autonomous Quality Enablement Bots mentioned above, and smart, predictive dashboards that provide stakeholders with appropriate information on quality status and recommended quality measures in production situations.