AI is coming, you ready?

The Shift from reactive to predictive

We have heard repeatedly that the real competitive differentiation of any organisation lies in moving from reactive to proactive mode. Now we are fast moving to a world where predictive is the new normal. Simpler form of Artificial Intelligence has been long in use in for e.g. computer gaming, online sales, social media, manufacturing and Hi-tech to predict outcome based on past patterns e.g. chess moves, product buying behaviour, simple tasks on the shopfloor etc.

But with advent of deep-learning, algorithms can be modeled so that the outcomes further improve the quality of inputs to make predictions more precise without human intervention. This forms a closed and endless loop of continuously improving predictions that feed continuously improving inputs which in turn generate continuously improving predictions and so on. The optimisation achieved in terms of prediction of the next best recommendation, action, strategy, decision is no more incremental - it is exponential and it is explosive! The role of data fed to the learning algorithms becomes more and more vital in moving from reactive to a predictive strategy.

AI is only as good as its data

To avail the full benefits of AI and powers of deep learning, every piece of information and data generated in a value-adding process in any organisation will now suddenly become very valuable as it serves as inputs to AI.

 For organisations to become AI-ready, they need to ensure a few fundamental things about data:

• Identify the value-creating processes, tasks, activities within and outside your organisation.
• Ensure that data is captured from these value-creating processes, tasks, activities in some format be it documents, files, drawings, records, notes, text, speech, videos, chats, emails, transactional data etc. There is no prediction if there is no data; hence some form of data is better than nothing. In legacy industries huge amounts of data are still floating within and across organisations in unstructured format e.g. emails, pdfs, images etc. In some other cases a lot of this valuable data often gets lost in transition between systems, people, tools e.g. important revisions, changes, decisions etc. Data silos pose another challenge for the availability of valuable data. A tremendous source of data is that residing as knowledge in the heads of people that is not available in any “fluid” form.These are all valuable for AI predictions and must be systematically captured. This requires careful evaluation of tools for data capture.
• Careful architectural considerations are required for conversion of critical unstructured data to structured data for complex quantitative & statistical analysis. However AI algorithms learn also from unstructured data e.g. speech, image, chats, text etc. to recognise characters, faces, words, etc. and to recognise patterns of their occurrence which can be very useful for qualitative (and simple quantitative) analysis.
• Also, the amount of data is key. The more the data, the more granular is the learning of an AI algorithm and better the results.
• As we talk about high volumes of data there is need for a cost-effective data capture and data storage strategy. Sensors, digital cameras, speech recorders, robots can nowadays be acquired easily at a relatively low cost to capture data that was previously not captured. For storage, cloud may be an obvious choice in most cases from a scalability, reach, capacity and cost stand-point.
• As data is a crucial source of competitive advantage, robust data security needs to be put in place with respect to who, what, when, how, can data be accessed.
• Data cleansing, data sanitization, data enrichment and data governance efforts for the value-creating processes, tasks and activities need to be prioritised. The output of AI is only as good as its data inputs. Crappy data = crappy predictions = crappy adjustment of input data = high risk business actions.

AI also presents opportunities to minimise or even eliminate waste generated in organisational processes by use of advance automation techniques. Data-in-use from such processes can be captured and taught to robotic algorithms to perform more complex tasks in the future.

AI is not utopia, it's really happening!

There are several practical industrial applications of AI that can improve operational performance. Below are just a few examples but the list is ever-growing-

•  AI to make recommendations: Algorithms can learn patterns of input data as well as the corresponding output data to propose recommendations during data entry. Refined outcome can be proposed upfront without having to run the process until its end. These techniques are increasingly employed (and for a long time) in B2C for e.g. Netflix, Amazon webshop to recommend a movie or a product etc. Companies like Under Armour use IBM's AI technology from Watson to produce in-depth health and fitness insights through digital health and fitness applications. These apps make recommendations to their users for optimal health and fitness.  In B2B similar recommendation techniques are now finding use to enhance user experience of various digital enterprise applications and tools.
• AI to propose and/or trigger next/best action Algorithms can be used to propose next best actions. E.g. Gmail proposes to create a calendar appointment the instance it notices you are agreeing a date for a meet up with a friend. I heard recently that one B2B company is piloting use of chatbot secretaries for e.g. to set up a mutually suitable time for a meeting by contacting participants with polite emails and asking suitable meeting time, then arriving at a mutually convenient time for all including identifying a free meeting room and drafting a polite meeting invitation for all participants.
• AI for automation & control: Algorithms can be used to observe, learn, create and continuously optimise a process or a part of a simpler, repeatable process that can be essentially performed without human intervention e.g. by robots. This has long been in use in some industries but the use is expected to grow widely across many industries to address need for advance automation. A good example is Siemens who help their customers to improve operations of wind turbines by adjusting position of rotors to the direction of winds based on learning and outcome from operational and weather data. Employing AI in combination with Augmented and Virtual reality e.g. for workforce training are creating new use cases for AI e.g. Connectar's MRO.AIR uses artificial intelligence to learn image patterns and AR display to guide technicians step by step in maintenance and repair service tasks.