AI Governance Framework and the Role of Primary Research

The adoption of technology fueled by Artificial Intelligence (AI) has been on the rise across organizations, industries, and functions. AI predictions are now at the forefront of multiple functional use cases in decision-making, be it in customer experience & centricity, talent engagement, sales & marketing, product development, or service operations.

Addressing Key Challenges in the Field of AI

AI development is largely designed around historic transactional & high-volume data through continuous processes where predictions are made, and responses collected over time are fed back to improve AI learning. This process leads to multiple challenges:

1. It creates a vicious cycle of a closed-loop system and can lead to AI model decay and predictive biases or unfairness.

According to Urs Gasser and Virgilio Almeida at Harvard University, “AI-based systems are ‘black boxes,’ resulting in massive information asymmetries between the developers of such systems and consumers and policymakers.” By the very nature of software development and trying to build more and more powerful AI models, much of the rules, standards, and governance that go into the model is hidden from consumers and policymakers. The complexity or simplicity of an AI model depends on the people developing it, leading to a certain amount of mystery with how different models compare and what biases or inbuilt prejudices are specific to each model.

2. The time lag in the overall process for an AI model to react to changing trends based on changing ground truth is long and not suitable for a changing marketplace. Lags are across the model deployment lifecycle too – in data collection frequency, availability of 3^rd party data, against changing market at a fast pace.

In the efforts to build more powerful models that can handle “big data”, developers run the risk of performance and time limitations for how long it takes the model to process and learn from a dataset. Just giving a model the means to digest huge amounts of data doesn’t mean that the model can automatically do it efficiently or sustainably. When it comes to business data, customer needs and motivations change on a regular basis, in irregular ways. Because of this, models have a hard time keeping up with the pace of developing and evolving customer data.

3. The data feeding AI systems are not always complete, e.g., in a customer or product space, the dataset might be missing data of non-customers and their needs, thus not giving a holistic view.

Many technology products will use Artificial Intelligence/Machine Language (AI/ML) interchangeably, but they serve different purposes in how they help break down data and build predictive and prescriptive models. Making sure an AI model can fit a business’s needs is important to test comprehensively before investing, in case the model runs into any problems with a given set of data or use case.

When it comes to differences between AI models, and which makes the best fit for a given business, it’s important to consider the dataset on hand and how the AI model may approach or tackle analyzing it. When it comes to business data, insights can be based on internal as well as accessible external data, giving a holistic picture of a business in its competitive field. By focusing solely on internal data, an AI model may present options for internal innovation, but divorced of outside data and information, those options may not be the best when considered without competitor analysis. But if a business solely wants to focus on its own internal efficiency and productivity, focusing on internal data may be the best approach. Working on a case-by-case basis to make sure a technology solution fits a business’s needs is important for getting the best return on investment for a technology platform or analytics solution.

AI Governance Framework as a Solution

An emerging space to address these challenges focuses on building an AI Governance framework. While multiple organizations have tried to address this through establishing audits, model monitoring & Machine Learning Operations (MLOps), most of these solutions would not holistically address all 3 factors that lead to sub-optimal AI predictions.

Google, in fact, published a whitepaper on the topic of perspectives within the field of AI Governance, which explains that the key areas for focus when it comes to defining governance over AI models are: “Explainability” standards, fairness appraisal, safety considerations, human-AI collaboration, and liability frameworks. All these together comprise the breadth of the discussion around governance for AI and help frame the key attributes for AI models that should be examined, recorded, and standardized. These five attributes serve as five avenues for governments and civil society to support legal and ethical boundaries for AI development and use so that this powerful technology doesn’t go beyond any globally accepted limits of uses or applications.

Primary Research Data – a step towards AI Governance

Looking deeper at the customer-centric space, another dimension for calibrating AI models would be to integrate with forward-looking data focused on a sample-based holistic view of customer preferences. This can then be overlayed on the transactional data. That way, businesses can digest historical and current customer data and use that to build forward-looking data models to predict needs and motivations. By not only addressing current customer needs but also projecting future needs, businesses have a better chance of maintaining and building their customer base through actionable insights from such predictive models.

A systematic process of measuring customer preferences that are stable for 3-5 years and the changing perceptions on a periodic basis to calibrate AI models can bring in the necessary correction required in addition to the current controls in place. A similar approach to AI Governance can be applied across talent engagement, sales & marketing, and other business functions. Understanding how current and potential employees experience a business, HR teams have a better chance of hiring the people they want and retaining their workforce for the long term. On the sales & marketing front, understanding what tactics and positioning currently work for a business only works when building immediate strategy. By planning and looking to the future, businesses can find the whitespace in a competitive field and position themselves for future and continued growth. It’s the basics of customer experience at play: by meeting and exceeding customer expectations, businesses have a better chance of reducing churn and building stronger loyal relationships with customers.

Another area of utilizing Primary Research data would be in Model baselining. Where heuristic business intuition-driven approach is compared against AI model predictive accuracy.

Having OSG’s Powerful AI-Driven Technology on Your Side

OSG’s ASEMAP™ technology addresses this need for a powerful and comprehensive AI Governance framework by bringing a new dimension to customer-centricity. Our method and technology do the hard work of digesting past and current customer data and getting to the heart of how and why the customers make their decisions. Understanding not only the basic business data of what purchases are being made, but going beyond to why those purchases are made, makes all the difference. ASEMAP^TM can make that jump from the discrete transactional data to gathering and overlaying behavioral and cognitive data straight from your current and potential customers. Reach out to our experts to learn more about how our data analytics and technology can work for your business, no matter your industry!

ASEMAP™ serves as the foundational IP for our technology products, bringing the power of behavioral and cognitive analytics to different aspects of the business world:

• OSG o360™ focuses on tracking the omnichannel customer journey, to understand opportunities for targeted marketing, product positioning, and nudging initiatives to push customers toward purchase. In this space, AI models help predict future customer behaviors based on their past and current choices, so that businesses can not only meet current needs but also anticipate and exceed future needs.
• PatientX360™ centers on the healthcare journey, gathering key data from relevant stakeholders, including but not limited to patients, physicians, pharmacists, healthcare staff, medical researchers, medical device manufacturers, and clinical trial participants and organizers. For the healthcare realm, AI is helpful for understanding how best to deliver care and designing nudges and engagement strategies based on each patients’ specific health plan and needs.
• RetailX360™ tracks and analyzes data for retail businesses, to help gauge metrics that businesses need to make purchasing, stocking, and future product decisions. Retail businesses need to be able to accurately predict future shopping needs to make intelligent current choices in inventory and advertising, which AI models and frameworks can help support.