Artificial Intelligence: Bridging the rift between Aspiration & Activity

I was reading an article on Artificial Intelligence at MIT Sloan Review regarding a study that they and The Boston Consulting Group conducted. Some of the interesting results of that study are (you can read it here):

• Almost 85% believe AI will allow their companies to obtain or sustain a competitive advantage
• Only about one in five companies has incorporated AI in some offerings or processes
• Only one in 20 companies has extensively incorporated AI in offerings or processes
• Less than 39% of all companies have an AI strategy in place
• The largest companies — those with at least 100,000 employees — are the most likely to have an AI strategy, but only half have one

These numbers show that though there is a lot of hype about AI & automation, the adoption is still very low. Though it is in-line with any new disruption- expectations exceeding reality & then finally realigning as initial requirements are exceeded, just met, or even unmet, it is logical to assume that innovators & early adopters are less in this area.

The biggest challenge is the lack of proper sanitized data to train the Bots. At an enterprise, for any AI solution to make returns requires rich & diverse sets of data to train the algorithms. In addition, with AI algorithms, the fundamental logic of Garbage-In-Garbage-Out holds, so the quality of data is also very important. Moreover, talent & resource shortages along with regulatory compliance adds to the complexity in adoption of AI.

Though these challenges and governance mechanisms will get sorted out, there are other challenges which needs to be overcome to realize the full benefits of AI. Thinking of AI use case as point solution, rather than an end-to-end solution is one of them. Automating just single task may not help enterprises improve the overall effectiveness & efficiency of their processes. Until the time, business processes operate in silos, as a use-case of AI, the overall potential will not be achieved.

The solution here is to re-engineer the enterprise operations, looking at them from new angles and trying to leverage the organizational cognition. The way to start this journey is to find a common ground between humans and machines, leveraging the advantages of both. I wrote in my previous article (read here) regarding this - humans can take decisions when factors are abstract and have a sense of empathy, rather than through modelling and systematic training; designing the right interfaces between the two entities is very important.

AI can transform the business operations, but first, we must understand what do we want from it, how do we align it with our current structure, and how to best leverage the organizational cognition with it. With this focus, it is easier to start thinking in terms of getting AI adopted in your enterprises.

Build Your Business in the Age of AI

Artificial Intelligence is influencing almost all the business verticals today. For e.g. AI is profoundly used in self-driving cars, financial trading, education & personalized services to consumer. The bulk of data created from social media and connected devices have helped tremendously in building new data models and improving the accuracy of these models. We have also built data models where machines can listen, see & respond. Two important implications can be:

·Now that they can speak, read text format, process & use previous training, it can help humans on myriad topics and at considerable depth

·Now that they can identify spatial objects & optical patterns, they can exit the virtual world & join us in the real world

In short, machines can solve problems where humans are great (parallel processing, for e.g. pattern recognition) and humans are slow (sequential processing, for e.g. logical reasoning).

Using AI for Competitive Advantage

AI in today’s world is helping business to reframe their traditional static competitive advantages (distribution network, capability, market share etc.) into more dynamic one. For e.g.

· Customer: AI has redesigned both the physical stores as well as high-traffic online portals by generating important customer insights. Retailers & e-tailers can run point-of-sale, loyalty, weather, locality, demography etc. data through their AI algorithms to create personalized promotion & marketing offers. They can learn about the customer & provide them with familiar, complementary or even new purchasing options for a complete new WOW experience. These WOW moments can help tailers to up-sell or cross-sell their products at negligible marginal cost.

· Capabilities: Traditionally any enterprise has three-core foundation: People, Process & Technology. By integrating new ways of working, say agile, the time to market product or services quickens as enterprises build prototypes and take them to market.

Add another layer of AI, and the whole gamut of insights it can provide based upon the data, can help you make your product right even in such small prototypes.

Beginning the Journey

Executives are responsible for deciphering activities, processes & tasks, which can give maximum ROI & is scalable. From top level, things may look complex, but as almost all complex activities can be broken down into smaller manageable tasks, executives needs to think fast, act smart. It is always better to analyses the use case of AI from these four parameters.

· Customer needs: Finding out the problem statement, which you want AI to solve from your customers, their implicit or explicit unmet needs.

· Technology: Use technology to capture data, signals & pain points from different sources and exploit the technological & computational advances to work towards a resolution. Find out ways, using which you can optimize your processes and services

· Data: After finding needs & exploiting technological advances to working towards a resolution, it is imperative, that executives do a sanity check. Using external data sources and feeding your algorithm can help you better your chances of finding out a potent business solution

· Divide & Rule: This step is mostly inclined towards aligning yourself to breaking down the complex, tested & proven resolution into simple tasks, which can then be followed and scaled.

These four steps, though may be intuitive, may have their own hurdles in terms of capabilities of enterprises and the required investment in hardware.

However, eventually, AI belongs to those enterprises & people that must put it to use. Any business (using AI) will be successful only if humans & machines work together!

The AI Ambidexterity: Combining Exploration and Exploitation

Executives often feel a tremendous amount of pressure to reduce their costs by achieving operational efficiencies. At the same time, there is also continuous pressure to implement new ways of working: automation, innovation etc. The solution to such pressure internal (cut costs) & external (available innovation in market) requires being ambidextrous while building your short-term & long-term strategy.

By ambidextrous, I mean a structured road map with defined timelines, RACI matrix etc., which is aligned with the business vision to exploit operational efficiencies and scaling those (efficiencies), while exploring new technologies which can give you competitive advantage in the market. On a side note, I have penned down my thoughts on why it is better to invest in innovation & not in technology. You can read it here

Importance of Ambidexterity

Ambidexterity in planning & strategy becomes necessary when the external environment is diverse, competition is fierce, government regulations are unclear & there is disruptions in technologies (PEST analysis) – requiring enterprises to be flexible all the time and change the way they operate in delta time.

Building Ambidexterity

The plan & layout to integrate ambidexterity into your enterprise is one of the most crucial activities for your sustenance. It effectively boils down to how many parameters (both internal & external) can you study and how effectively can you map them into Termination or Transition.

By Termination, I mean identifying activities, processes, tasks that can be eliminated (which then maps to innovation). By Transition, I mean redesigning, refinement & renovating these activities, processes or tasks (which maps to scaling operational efficiencies).

Note of Caution

Ambidexterity, though, allows you to be flexible and nimble in your short-term or long-term strategies, it is very difficult to master.

The struggle to adopt ambidexterity is going to rise (as the analysis of PEST framework leads you with very uncertain results). The division of your enterprise into Termination & Transition is the stepping-stone towards facing the dynamism of the business environment. Acting now can be useful; else, your enterprise may risk being overtaken by an ambidextrous competitor.

AI & RPA led Renovation: Three Critical Activities for Success of your Enterprise

Most CEOs have understood the importance of automation led renovation in their organization. Using data driven insights, they would like to see 15-20% EBITDA gains in sales, marketing, R&D, manufacturing, & supply chain.

However, becoming data-driven is becoming a challenge for all the CEOs. They want it, but the initiatives & actions to embed and assimilate data in operational excellence at an enterprise level mostly fails. This is because they

• Plan to see the benefits of embracing data too soon – without proper sanitization, cleansing & curation
• Try to reinvent the core IT systems – that can be strenuous on the budget & can take time
• Have no sustainability plan in mind

One thing to keep in mind when planning such large transition & transformation plan is that the current market is very dynamic and the rules of the game are changing very frequently, so what they need is an agile model, focused upon results, and action items, which are manageable. They need a roadmap that can give them security about the future, not kill them during the process.

The most important key driver for such initiatives to succeed is – cost effectiveness, manageability & sustainability.

Three simple ways that mitigates the risk of the future and the outcomes of the renovation are:

#1: Start with Proof of Concepts (POCs) & Pilots

Identify routine activities, low hanging fruits as the first cases for POCs & Pilots. These POCs & pilots should be finished off in either weeks & quarter, and no longer. The experience from these initiatives will give an overview of how ready is your environment to proceed with automation of other simple/complex use-cases and rolling out these automated use-cases to the complete enterprise.

In this step, it is also important to start building plans for talent acquisition, enterprise learning and capacity enhancements.

#2: Communicating the larger picture

This activity can start while the first step is still underway.

Though the enterprise readiness for a complete automation is still distant & depends upon the results of the POCs/Pilots, today or tomorrow, the enterprise has to align itself with the automation roadmap. Hence, it is pertinent that some initial hypothesis & governance are ready for building the portfolio of automation: identification of departments, prioritizing units amongst them, finding out use-cases in those units. These activities needs to be accepted by the concerned teams, therefore, communicating and getting their buy-in is important. This also helps to understand and plan for the ground level challenges which leadership have to ponder over for an enterprise wide automation-acceptance journey.

#3: Structuring for Sustainability

This step comes into picture when the other two activities are in place – enterprise is ready to rollout automation and necessary funding & buy-ins from relevant stakeholders is in place. Now is the time to start planning for the culture change in the enterprise. Employees working in silos have to come together (because now routine works that were performed in silos are over & more collaborative, knowledge driven work is expected); way-of-work has to change & the leaders have to make necessary organizational level change to sustain the automation: maybe change management program to inculcate new mindset, behaviors etc.         

There are other important processes as well which needs to be undertaken – building operating model, procuring infrastructure, deciding on legacy systems (phasing out).

Executives are motivated to use automation to reduce costs, improve efficiencies & performance of the system. Adopting automation at small scale is good, but devising & formulating an enterprise wide roadmap & strategy is the key for sustaining the business in today’s world. Their renovation journey to adopt automation will be successful by being agile, structured, disciplined and pragmatic.

What Should an Executive Know about Machine Learning? "Unsupervised Learning"

This article is in continuation of my previous articles on Machine Learning & Supervised Learning.

In this post, I am going to share my views on Unsupervised Learning. I have tried to capture the basics here

The basic fact in unsupervised learning is that that the data model performs prediction/actions/inferences by learning from input training data, which in itself does not have any output/results defined. Meaning there is no particular solution/target/output or even error to evaluate an outcome/prediction.

Unsupervised Learning can be further divided into two categories:

Clustering

It means grouping of items into subsets (or cluster) so that the observations & inferences coming from the same clusters are similar. It also implies that the behavior one subset will be different from another subset.

Applications of clustering:

1. You run an e-commerce firm (with large volume of data on customers & their buying patterns) and want to find groups of customers with similar behavior for chronographic watches. Clustering is what you do

2. You are an insurance company and want to segregate group of policyholders with high average claims.

Dimensionality Reduction (DR)

A straightforward method for feature selection and feature extraction, this method reduces the features to process, so that the performance improves and the technique becomes computationally more efficient.

For example, consider a situation where you want to classify buyers of watches from non-buyers of watches based upon their demography. The dimension of this data can be very large (age, education, race, sex etc.). Therefore, if one start applying classification upon all these dimensions, then the system may take very long to process the records. A computationally easier way can be to use DR to find a subset of data that can represent the original data in a non-redundant way; and hence, both cases will lead to the same result.

In addition, it is common experience that projecting higher dimensions data into 2D leads to better visualization of the data set.

Summary:

· In unsupervised learning, we do not know the outcomes

·  It can be of two types: Clustering (grouping) & Dimensionality Reduction (50,000 features become 10)

Hope it helps in your next sales pitch to convey these concepts better!

What Should an Executive Know about Machine Learning? "Supervised Learning"

This post is in continuation of my earlier post on Machine Learning & the three buckets in which it can be understood. You can read it here

In this post, I am going to express my opinion on Supervised Learning.

In supervised learning, the output variable is known, and this output variable is used in the training.

There are three steps for building a supervised model: Building model, Training model & Testing model. Let us understand these three with the help of an example.

Step #1: Building model

Suppose you have joined a coaching class to learn machine learning. Hence, in this case, you become the model.

Step #2: Training model

Your faculty will be teaching you. She will also use various teaching aids during this process. This is the training process. Here, we try to train the model using historical & recent data. The basis of this process is to identify either patterns or dependencies in the data.

Step #3: Testing model

Now is the time when you (model) has to appear for the exam. Obviously, the teacher will not use the same data to test you on which she has trained you, hence, the exam paper will have similar patterns on which you have to respond, but not the same.

Generally, to test the prediction or accuracy of the model, we test it on the untrained data. Usually, the ratio of training data to test data is 70/30.

If your exam score falls below a configured value, then, re-training happens.

Let me correlate it with IT & Business use-cases:

Case #1: IT | Proactive Maintenance of Infrastructure

You take 100,000 tickets from your ITSM tool, build a data model (70,000 tickets) & test it with the rest of the 30,000 tickets. If the accuracy of the data model is >85% (e.g.), then you roll the model for proactive maintenance of your infrastructure (servers, routers etc.).

Case #2: Business | Detect Fraud Transactions of Credit Cards

You gather data on the fraud transactions. Again you split the data into 70:30 ratio. In this case, let’s assume that the model has 75% accuracy (which may be good for rolling it out live). So, this model, when encounters pattern abc in the new transactions, it can predict the probability of fraud in that transaction. Hence, now, you can take necessary actions

Both the cases which I mentioned falls under the category of Classification problems under our initial Supervised Learning. There is another category, called Regression (same thing which you did during your MBA days using SPSS, & hence you all have learnt some machine learning!), which is the second category under Supervised Learning.

Technically speaking, Regression is independent of any framework: machine learning or any classical statistical methods.

Regression refer to a model to predict some numbers, like real values. This is different from Classification, which predicts discrete variables (fraud, mangoes etc.)

Summary: 

● In the supervised learning, we know the outcomes

● The Three steps process: Build, Train, Test

● It can be of two types: Classification (discreet classes) & Regression (real values)

Hope it helps in your next sales pitch to convey these concepts better!

Special thanks to Aditi Aggarwal for helping me with the content & Debapriya for the motivation!

What Should an Executive Know about Machine Learning? "Reinforcement Learning"

The age of advanced data analytics & cognitive computing has enabled the use of complex algorithms across a wide range of business problems, industries, verticals etc.

These advancements have led to usage of innumerable jargon on Machine Learning, data modelling, AI and RPA. Given, we understand that these machine learning models & complex algorithms solve abstract business problems (of course, with probability), improve performance, & create a differentiation positioning for their user. The challenge is that these algorithms, mostly, act like a black box & too much dependency upon them for decision making, without knowing their boundary conditions, can lead to innumerable risks.

So, understanding Machine Learning in simple English can help us all to make a better choice on our dependency upon such algorithms.

Machine Learning basically can be categorized under three buckets:

Reinforcement Learning

This learning enables software or your data model to self-determine the ideal behavior, ceteris paribus (i.e. other things equal), so that the performance of the software or the data model is maximized.

This model works as an interaction between two elements – the learning agent (your software/ data model) and the environment.

The reinforcement signal is sent to the learning agent as an award for every correct action which the learning agent takes. This mechanism repeats and the continuous rewards improves the agents’ learning of its environment. So, if you are thinking it’s more like a trial & error way of learning, then you are right!

Case in point #1

You are devising a data model for a robot, which will help tourists navigate a historical place.

While the robot is in learning stage, every time the robot takes 10 steps on the left, it is hit by a wall. So, now, after multiple iterations, it knows that 10 steps is dangerous. Hence, the next time, instead of taking 10 steps, it will understand that the passage on the left is 4 (for e.g.) steps away.

Case in point #2

Your robot is designed to store/retrieve products for optimized space utilization in a warehouse.

Suppose the warehouse is divided into blocks. Now the robot is able to place 100 items in block A, but for every 101(st) parcels, the entry to that block is jammed. So, the rewards system from the environment (in this case), entry to the blocks, will enhance its learning of the layout of the warehouse.

The above two examples are very crude way to understand how these algorithm works.

Supervised Learning & Unsupervised Learning

These two are very often heard in business meetings, sold in sales pitches of software vendors etc. To make it simpler, I have written a complete article on both these terminologies. You can read it here

In my next post, I shall try to break down these three buckets into simpler sub-buckets, categorizing many more jargon into these three buckets.

Understanding basics of ML, its components & use cases can lead you to be an innovator & a disruptor in your industry. Start small but start learning!