Let’s analyse the AI space
By Shane McArdle, Director, CAMO Analytics
The pace and desire for advanced analytics and AI is increasing exponentially. This is largely driven by the hype that AI can derive previously hidden insights resulting in increased value for business and society and can continue to do so as it self learns. It is a good time to take a moment to differentiate and clarify the vernacular surrounding the technology. Again and again we hear arguments (which usually end up as very insightful blogs & articles) from the data scientists within the CAMO Software community trying to clarify the many facets of Artificial Intelligence.
For this exercise let’s start looking at three pairs of antonyms:
Business versus technology, top-down versus bottom-up approach and industrial versus business usage of the technology.
Traditionally AI and data analytics has resided within the IT department and it has been up to them to motivate, teach and align governance in IT with upper management thus creating a robust alliance between IT and business.
The whole idea was to advocate the business potential of integrated ERP, databases, automation and more. Heavy duty BI enterprise solutions became the norm and the hype of that time. The buy in stakeholders didn’t know or care about the time (years) and costs (MUSD) it took to actually implement these systems that have resulted in legacy platforms that business can’t now live without, even if better solutions are out there today.
In the last few years the newest promise of AI, machine learning and advanced analytics has been described in terms like Big Data, platform economy and digital business models, and the message has really resonated with disruption eager board members, C-level execs and consultants.
While senior level management of many organizations are on board and want an “AI project”, the job of implementing this technology has again fallen to others in the organization.
There are now many articles popping up, pointing out that despite the hype Big Data transformations has by and large failed to deliver on its promise.
I am not a pessimist but more an optimistic realist and believe the real value of AI and Big Data Analytics can be achieved if the right approach is taken. There is no 1 right solution that fits all, but clear processes and good tools will ensure a better chance at success.
For many AI remains a black box; how does this technology becomes so intelligent? There is probably a basic knowledge that it involves large volumes of data, algorithms and a great capacity for calculations. With the current ignorant (maybe too strong a word) let’s use ill-informed situation, it has become increasingly difficult for data scientists (who understand the details) to flag criticism, skepticism or even use hard facts about artificial intelligence and machine learning that are counter to what people believe. The current trend probably has something to do with this; heavy marketing driven storytelling blurring the reality of what can be achieved. For many the concept of AI is very new but in fact the fundamental components and methods have been around for years. We are just witnessing the slow and steady evolution from reporting to prediction to prescription.
To help (I hope) close the gap between and re-vitalize the dialogue between C-level and the breed of good old-fashioned data scientists let’s review the problem in the terms of business versus technology, top-down versus bottom-up approach and industry versus business usage.
No 1. Business before technology
A survey of more than 500 executives compiled by McKinsey states that 86% of participants reported their organizations were only partially effective at meeting the KPIs they set for their data and analytics initiatives. The technology is promising and has proven very disruptive to business models, job functions and comes with worldwide scalability as a promise. But we need to develop, deploy and implement it with a business problem you want to solve in mind. If we lose sight of business problems we lose the ability to choose the right tool in the digital tool box. “We want to disrupt our business to prevent others from doing it” is not a business problem. It is a strategic goal you can address by nursing an innovative corporate culture and experiment with ideation tracks with diverse participants to learn more and develop ideas. “What can I do to reduce customer churn?” or “How can I improve my processes?” may be a business problem, and you can address these problems and find some key insights using analytics. But you cannot whisper it to a robot or push all this data into a black box algorithm and expect actionable answers instantly.
Rather than jumping in and applying an AI algorithm to a data lake of information, first, identify the problem to be solved and work backwards from there. Trying to digest a lot of (Big) data has proven to fall short. Instead valuable and actionable insights can be achieved from small datasets focused on a defined business problem. Using the right, off the shelf tools such as CAMOs Unscrambler® and Process Pulse® reduces the need for bringing in a team of consultants to program a specific solution simplifying the problem even further.
Once the business problem has been defined and described as best as possible you can then bring the technology to bear. Start with mining the data to find significant variables and creating a set of hypotheses to work with. Each step should lead to actionable, affordable insights and you just keep refining. This is where you need a data scientist to help you develop the models and management support for actions to test it in real life. It is not new, it is not rocket science, it is what data scientists have done for three decades now. The special breed of chemometricians– have been doing this using a multitude of mathematical and statistical methods in chemistry, biochemistry, medicine, biology and chemical engineering – is just one specialised subset tribe of data scientists deeply rooted in a tradition of solving complex industry problems.
No 2. Top-down and bottom-up
So how do we tackle the fact that the promise of the technology has given C-level management the inclination to drive forward in a top-down approach? To be clear, I believe it is a positive and very necessary phenomena to have this digital push from the board room. From the practical point of view, I have seen how expectations from the top have been several years ahead of what the technology can deliver, and it has created frustration and conflict between the visionaries and the do’ers (a noun adopted by the people whose job it has been to implement these type of programs). This climate is not conducive to successfully implementing a Big Data strategy.
I would suggest a dual approach of top-down and bottom-up at the same time. In most industries we already have data scientists working to optimize and develop the current set-up. They work with data close to the on-going and un-disrupted traditional production and R&D. It is this group that have the tools, the skills and the experience to close the gap between the vision and the reality and develop the next, realistic step forward. This resource is extremely valuable in an innovation track looking for new business models and new potential. The CDO, CMO or CEO in charge of digital transformation should not dismiss these competencies – they know what they are talking about although they will probably not be delivering the most compelling slides about cloud enabled, robot based digital business models for the board room!
No 3. Industrial versus business AI
The tool box of analytics and AI is big and still growing. Companies need to understand the differences and use the right tool for the right problem – the hammer for the nail and the screwdriver for the screw.
Playing a closed rule game like chess or GO is a completely different story than a self-driving a car. Netflix predicting that I would like Game of Thrones after I watched all three Lord of Rings movies is a whole different ball game than mapping a cancer treatment to a patient. The implications of getting the first one wrong is not worth mentioning while getting the latter one wrong may lead to death and bankruptcy. A similar analogy can be used for using marketing analytics to pick the right color for new jacket range compared to classifying and doing a root cause analysis of defective products in real time in a manufacturing environment. Getting the first one wrong will not have a comparable cost impact as the second.
The industrial need for scientific certainty has been the premise for the data scientists working methodically with raw materials, processes, manufacturing and industrial implementation. For the self-driving car to become a reality it needs to strive for these same zero tolerances for failure as more regulated industries. Maybe the culture surrounding AI needs to be more aligned with the working chemometrician or statistician and by adopting some of their inherent traits we will be able to deliver on the AI promises and mitigate the risk at the same time.