ISSN 2644-2876

DECEMBER - 24 - 2019

APACCIOOUTLOOK.COM Connecting the Enterprise IT Community

AI
E D I T I O N

A DIGITAL
COMPANION
FOR PEOPLE
WITH
DISABILITIES
FRASIL
$15
CXO NSIGHTS

AI FOR ROI –
HOW TO
CHOOSE
THE
RIGHT AI
SOLUTION?
AN INTERVIEW WITH ABHINAV SINGHAL, CHIEF
STRATEGY OFFICER, ASIA PACIFIC, THYSSENKRUPP

A
bhinav Singhal is the Chief Strategy & Innovation 69% of the use cases advanced AI methods such as deep
Officer for all thyssenkrupp companies in the neural networks helped in improving performance of
Asia Pacific region. His technical expertise is already established methods and for remaining 15% could
responsible for creating and shaping growth provide only limited benefit.
opportunities for organizations. He is actively This is also consistent with our experience that the
involved in leading digitalization initiatives for the group greatest potential for AI is to create value in use cases
and is a board member for thyssenkrupp Innovations, where already established analytical techniques (such as
which is responsible for incubating new technology- regression or classification) can be used, but where neural
enabled solutions for customers in the Asia Pacific. Before network techniques can generate additional insights or
joining thyssenkrupp, Abhinav led the strategy team at broaden the application base. It is a common pitfall for
Dell Technologies, where he was involved in defining companies to get caught up in all the hype surrounding
Dell’s digitalization strategy. He has also worked with the advanced AI developments and miss the rationale of
McKinsey & Company for several years and in his last adopting AI in the first place. Oneway, CIOs can assess
role was an Associate Partner, where he advised energy the suitability of an AI solution is it to break it down into
and industrial businesses on their strategy and operations, simpler elements and ask five basic questions.
in markets across the world.
Companies looking to adopt AI today are bombarded 1. What is the core business problem to
with technology companies and start-ups selling advanced be solved?
machine learning based solutions built on exciting use Over two third of the expected value from using AI is
cases. However, before kickstarting newer pilots and in either revenue generation use cases (e.g., product
investing in these advanced solutions it is useful to step recommendation, customer service management, pricing
back and reflect on the overall intent of using AI for & promotion) or operational improvement (e.g., predictive
the organization and the traditional suite of analytical maintenance, yield optimization, supply chain). While
techniques and resources available. consumer led industries such as retail and high tech tend
A recent study by McKinsey & Company compiling to see more potential from marketing and sales related AI
hundreds of AI use cases and applications across multiple applications, manufacturing and other heavy industries
industries found out that just in 16% of use cases advanced see more benefit in using AI for operational excellence.
machine learning based AI solutions were only applicable The remaining value pool is distributed across the support
and traditional analytical methods were not effective. In functions for example, task automation, people analytics,
risk assessment etc. Given the wide range of applicability 4. What are the ‘training’ requirements?
of AI techniques, it is important to pin down the source Large labelled training data sets are required in most
of value creation for the company and then determine the applications to make effective use of advanced techniques
pain points or opportunities where it makes most sense to such as neural networks, and in some cases, millions of them
invest in AI deployment. to perform at human levels. As training data set increases,
performance of traditional techniques tends to flatten and
2. Which analytical method or technique advanced AI techniques tends to increase. However, if a
is best applicable? threshold of data volume is not reached, AI may not add
Most of the business problems can be classified into few value to traditional analytics techniques making it also
standard types and have a corresponding set of established an important consideration. While promising new ML
techniques to solve them. Some common examples (not techniques such as reinforcement learning, generative
exhaustive) include: adversarial networks, one-shot learning are trying to
• Classification (e.g., categorizing products of acceptable overcome the requirements of supervised learning, which
quality)–Logistic regression, Discriminant analysis, Naïve requires humans to label and categorize the underlying
Bayes, Support Vector Machines, CNNs data. Companies still need to assess their ability to collect,
• Estimation (e.g., forecasting sales demand or predictive integrate, govern and process data at scale before deciding
analytics)–Linear Regression, Feed Forward Neural on the right AI solution.
Networks
• Clustering (e.g., segmenting customers or employee)–K 5. What is the optimum system (or hardware)
means, Gaussian mixture, Affinity propagation configuration?
• Optimization (e.g., capacity or route optimization)– Depending on the end use case and its requirements, data
Genetic Algorithms, Differential Evolution, Markov can be generated from a variety of sources including sensors,
decision process IoT devices & machinery or social networks. This could
• Recommendation (e.g., next product to buy)–Collaborative result in massive amount of fast, structured, semi-structured
filtering, Content filtering, Hybrid or unstructured data, time series or real-time which can be
The most prevalent problem types are classification, stored in original granularity or aggregated or pre-analysed.
estimation, and clustering, suggesting that developing the The choices are immense. It’s important to maintain focus
capabilities in associated techniques could be a good starting on the ultimate requirements of the end application and
point. Also, in most cases, for a specific problem both the prioritise accordingly. Similarly, the AI application can
traditional techniques or advanced deep neural network- reside in the cloud or at the edge to minimise latency.
based techniques could be applicable and its important to Cloud typically offers a flexible and scalable environment
assess the trade-off between them before selecting. at relatively low-cost and without huge initial investments.
Many providers also offer APIs for computer vision, speech
3. What types of data-sets are needed? recognition, and natural language processing (NLP) or other
Data sets also play an important role while choosing cognitive domains on their cloud platforms which are pre-
between traditional and advanced techniques. They can vary trained and preconfigured for a certain task and serve as
from being structured or time series based to text, audio, gateways to AI applications. All these configuration options
image or video based. Typically, neural AI techniques excel need to be considered before finalising the optimum system
at analysing image, video, and audio data types because infrastructure to enable the preferred AI solution.
of their complex, multi-dimensional nature compared to Ultimately, for CIOs to create value from their AI
traditional techniques. However, sometimes even more initiatives, it is important to develop an understanding about
value can be extracted from mining insights from traditional which business use cases have the most value creation
structured and time series data types rather than going for potential, as well as which AI or any other analytical
audio visual data sets. Building a deep understanding of use techniques is most suited to capture that value and can
cases and how they are associated with particular problem be deployed at scale across the company. The decision of
types, analytical techniques, and data types can help guide choosing the right AI solution is often driven less by the
companies regarding where to invest in the technical sophistication of the technique but more by the available
capabilities and the associated data that can provide the skills, capabilities, IT infrastructure and data at the end for
greatest impact. organization wide implementation.