AI Based Digital Transformation in Large Enterprises — Considerations Before Kickoff

Who Should read this ?

If you are a medium or big size company and you are ready to start a Digital Transformation Project then this article might give you some useful information about various roles and responsibilities that are required for the project.

Introduction

The mid to late 1990s saw the initiation and implementation of many ERP projects to integrate horizontal business functions. Likewise, there is now momentum to initiate large-scale projects for Digital Transformation. Leading consulting companies define the following technologies as the building blocks of Digital Transformation:

• Artificial Intelligence (AI)
• Block Chain
• Internet of Things (IoT)
• Virtual Reality (VR)
• Augmented Reality (AR)
• 3D Printing
• Robotics

Many enterprises are embarking on an AI-first road map to lead the Digital Transformation. In this article, I will discuss some important points to consider before kicking off an AI initiative.

Note: In this article both Machine Learning and Deep Learning initiatives are considered AI.

Before getting into the Project set up, a brief discussion of the key terms is in order.

What is Artificial Intelligence?

Though AI is defined in many ways, the most widely accepted one being “the field of computer science dedicated to solving cognitive problems commonly associated with human intelligence, such as learning, problem solving, and pattern recognition”, to me it is the idea that machines can possess intelligence.

The heart of an Artificial Intelligence based system is its model. A model is nothing but a program that improves its knowledge through a learning process by making observations about its environment. This type of learning-based model is grouped under Supervised Learning. There are other models which come under the category of Unsupervised Learning Models.

Why We Need to Build AI Models internally?

An AI model developed on external data to solve a problem in a different organization might not produce satisfactory results for your organization, but this cookie-cutter approach is practiced by many IT vendors. In some specific & very narrow scenarios this cookie-cutter approach may work but is not transferable for a vast majority of the cases. Hence, we are better off developing new models from scratch or do Transfer Learning on top of the pre-trained models. [Transfer Learning is the ability of an AI to learn from different tasks and apply its pre-learned knowledge to a completely new task. The pre-training was done using the data which is not part of the organization].

What is Digital Transformation?

To me, Digital Transformation is re-engineering the current business processes with information flow to plugin new technologies across various business functions without directly impacting employee head count.

I will focus on a possible approach to augment business processes with AI based transformation.

Essential Components for Success

The success of AI depends on increased computing power and the development of complex algorithms to leverage exponentially growing data generated in transactional systems by humans and machines.

With this basic information about Artificial Intelligence, I will walk you through the project setup. This could potentially be a Discovery and Evaluation phase for larger projects.

First Things First

The first and most important thing for any organization is to form an AI CORE team filled with a mix of adequately qualified technology and business experts. This approach is suitable for enterprises that are non-digital native [Companies that have not yet adopted Machine Learning (ML) or where small-scale machine learning algorithms like email spam detectors are currently being used].

Teams and Responsibilities

These teams are experts in their respective domains and work together to achieve project goals.

Data Science Team

• Define the medium and long-term Road Map for AI adoption in the organization
• Identify the short and medium-term research areas to address the organization’s business challenges

For Example

o Computer Vision(CV) vs Natural Language Processing (NLP),

o Convolutional Neural Network(CNN) Vs Recurrent Neural network(RNN)

• Gather business scope to size the hardware and other infrastructure components
• Identify tools and frameworks for the Model development and deployment life cycle
• Build and Test the Models and Deploy it for Serving

I personally do not recommend starting with Unsupervised training models (like Generative Models and Reinforcement Models) as they are still in their early stages and will take time to be mature enough for introduction into mainstream business processes.

Infrastructure Team

• Decide the AI Model serving landscape (Cloud or Edge)
• Work with the Data Scientist and Application teams to define various frameworks and tools for the new landscape
• Work with the Data Scientist and Business expert teams for initial and scalable data growth for Model Training
• Define environments (Development, Training, Testing and Serving) to support the AI development life cycle for both Desktop and Mobile applications (though mobile based applications may not be required right now, it is still recommended to consider this for scalability purposes)

Business Expertise Team

• Gather knowledge about the capabilities of AI and its suitability for the organization
• Identify business cases where AI implementation will be beneficial
• Define Acceptability Metrics for AI models
• Identify gaps in data available from transnational systems and rectify them
• Prepare Training and Test Data for AI models

Application Team

• Identify gaps in data available from transactional systems and rectify them
• Develop tools to help business teams to label the data for Supervised Learning
• Below are a few scenarios where the development of tools would help

- For the structured data available in existing databases, the business team might add classification categories such as Success Vs Failure or Class 1, Class 2, etc. for each record

- For unstructured data in the Natural Language Processing category, such as customer complaints and Twitter feeds, the business team might want to add labels for sentiment analysis classifications (Positive Vs Negative)

- For Computer Vision Models based on the type of data (medical images, photos, video streams, and satellite imagery) appropriate tools needs to be developed for labeling purposes (also called as ground truth).

• Develop skills in Data Cleaning, Augmentation, and Data Preparation Techniques/Programming
• Enhance business applications with newly deployed AI models, including new mobile apps

Security Team

• Develop standards, guidelines, and checks for possible cyber security threats

AI Model Development Life Cycle

In the diagram below, I have proposed an AI model development cycle and its integration with business processes.

Conclusion

Once the teams are formed and adequately equipped from a resource perspective (for example, infrastructure and new skill augmentation), the project schedule with roll out phases could be defined based on the scope and availability of resources. I intentionally did not touch on budgeting as it varies widely based on many factors, which are beyond the scope of this article.

Implementation scenario on BIG data Infrastructure

It appears simple if I think of only utilizing just deep learning framework such as TensorFlow or MxNet or PyTorch to start an AI project. However, it may become complex , if the organization already invested in a BIG data infrastructure enabled with ML capabilities (for example Apache Spark or SAP HANA etc.).

In the BIG Data Scenario, there are many ways to embed Deep Learning with the BIG Data Infrastructure.

For example, With Apache Spark, the common workflows can be done with TensorFlow on Spark or Deep learning Pipelines.

For HANA, at this stage, it is only possible to perform scoring on models that have already been created in TensorFlow and are being served by TensorFlow serving. As for training a TensorFlow model from HANA — that is not currently possible.

A Strategical view, Visionary analysis and Key stockholders’ input are essential to finalize the Infrastructure.

This strategy is only my opinion of course, for other people it might be easier to do things in a different way.

Thank you for reading this article! Hope you found it interesting. Hit the clap button if you did! If you have any questions, you could hit me up on social media or send me an email ([email protected]).