Artificial Intelligence (AI) as a discipline, technology, or a buzzword has been present in our collective mindset for a good part of the last two decades. The importance of AI in enhancing business processes, decision support systems, risk management, and improved customer service has been clearly established in several industry domains. In a recent survey conducted by McKinsey called “The State of AI in 2020”, up to 50% of the participants have reported that their companies have adopted AI in at least one business function with the adoption highest in product or service development and service operations. This emphasizes the relevance of AI in business growth, customer retention, and customer service.

Are businesses equipped properly to leverage AI?

So the question becomes, how well are companies equipped to leverage AI effectively across the various business functions and ensure accurate, timely and relevant intelligence for the functions of business that need it?

Artificial intelligence as an extension of Software Engineering relies on the execution of code on various forms of data. Such code is usually developed using libraries known as an algorithm or a Machine Learning model that is applied to extract the relevant intelligence. Examples of such use cases are Revenue prediction in Sales, Anomaly Detections in Security and Sentiment Analysis in user generated content based on natural language.

The development of use cases using AI algorithms has evolved greatly in the last few years with widely available resources that help software developers and programmers in creating working use cases on the data they possess. 

Here’s a good demonstration of the complete AI Systems Development Lifecycle

Congratulations, you just reached Base Camp!

The development of an AI model based on a common life cycle depicted in the diagram above is a significant achievement for any business. This is even more remarkable when a business is able to successfully implement this process to solve problems iteratively through multiple AI models successfully.

However, the development of an AI model is akin to reaching the base camp in the journey of scaling a summit of Applied AI, or in other words this is the successful completion of step 0! Successful development of an AI model is just the beginning of a long process towards applying the model successfully in a live business function or production systems, as we commonly refer to it in the software industry.

How is the AI being managed?

Once an AI model is developed, the next steps involved are

• Productionizing the model, i.e. deploy into a live environment
• Connect to live data on which the AI needs to be applied
• Provide access to the outcomes to people or business processes
• Secure the access to the AI outcomes through Authentication and Authorization procedures 
• Refine and retrain the AI model based on outcomes observed with live data, establishing a clearly defined feedback loop
• Integrate with software systems for real-time access to intelligence
• Enhance software services and products based on the outcomes derived from integration with the AI model
• Monitor and measure the success of the integration

This process, clearly falling outside the development lifecycle of AI, can be considered as the AI Management process, a distinct and clear need for successful adoption and scaling of AI.

A lot of emphasis has been placed historically on the development of AI use cases as the need of the hour. The resources and talent available now satisfy that need several times over. However, not enough attention or emphasis has been placed on the effective management of AI. This is a general gap in the AI industry overall and not considered as a deficiency for businesses adopting AI. The problem needs to be addressed on a wider scale than in specific instances.

What is an AI Management System?

An AI Management System has the following objectives for effective management of AI in an organization

1. Host – the AI Management system is responsible for

• Hosting the data, 
• Providing the real time and historic analytics comprising Descriptive and Diagnostic insights
• Hosting the AI models developed by following the AI Development Lifecycle
• Continuously training the models on the data hosted in the system
• Extracting the insights from the data based on AI models in an automated or on-demand manner depending on the need

2. Secure–

• Applying Authentication measures to secure access to the insights extracted from the data
• Applying Authorization rules to secure the access to the insights to the appropriate level of the requesting entity

3. Distribute–

• Providing access to the insights to people and systems according to their access level
• Providing integration options such as API integration and Federated Authentication processes such SAML, OAuth to distribute the insights in an on-demand manner

4. Learn–

• Monitor the usage of the insights and the effectiveness of the outcomes in solving business problems
• Provide a feedback loop for people and systems to report their feedback or capture the result of the interaction
• Retrain, refine and redeploy the models to provide more accurate insights based on the feedback gathered

The need and implications

Any business or organization that has an objective to adopt AI into their business functions or scale AI from their current implementation needs to have a well defined process that effectively manages the usage of AI. 

Not having such a clearly defined process can lead to stale insights over time as the models become outdated and are no longer providing the right insights considering the present day scenario.

The other, more significant problem of having an inefficient process of managing AI is the integrity of the insights. Bias in AI has been a common problem from the inception of this discipline and this is usually the result of inefficient training and improper implementation of the feedback loop. For businesses, this can become critical if a business function that is using AI is running based on inaccurate insights and even worse, does not have the mechanism to learn from the gaps and correct them.

The alternatives

Despite the constant buzz around AI, the relative evolution of the systems to manage AI are few and far between. Due to this, businesses that adopt AI handle the need of managing AI by developing such solutions themselves. Many times, this can solve a temporary problem or need but don’t do much to address the much bigger need of having a mature AI Management process.

Should businesses develop these systems themselves? This is a topic we are going to address in an upcoming article, but to summarize this, the need for developing such systems depends largely on the scale of a business’s operations. For a SAAS company offering a technology product or service that can be enhanced by AI, the scale of developing an AI Management system in addition to its own internal systems can be far too demanding for successful execution and effective results. Such companies can clearly benefit from leveraging external systems that can help manage AI successfully.

Next steps

In upcoming articles, we will provide further details on the key elements that need to be considered when choosing an AI Management System or developing one. We would also provide the pros and cons of a Build Vs Buy approach when considering an AIMS.