AI and the Fear of Missing Out!
We all want to do something with AI and Machine Learning. But how do we overcome the AI FOMO (Fear of missing out) fever? In this short article, we will try to “place” AI and its branches in the enterprise. We hope you find value in this article.
AI at a high level has 3 main branches that are relevant for Business (there may be others).
- Machine Learning
- Natural Language Processing
These are very wide topics; and they are relevant to business. But where do we apply them? Finance, HR, Operations, IT, the Executive Branch?…The answer is simple; All of the above. The next question is: How do we apply them?
Well, most businesses fundamentally have processes; Manufacturing Processes and/or Businesses Processes. Businesses also have managers and executives who make management decisions. More and more executives are making these decisions based on data that is generated by the business and the operations.
At a high level AI and its branches can be used for the following:
- Prediction & Forecasting Problems
- Process Automation Problems
Prediction problems are usually addressed by advanced business intelligence and analytic groups within enterprises. Machine learning and data science expand the horizons of current BI systems by bringing in the ability to analyze vast amounts of internal and external structured and unstructured data, and by applying a learning concept that is inherent to machine learning. In a nut shell, by applying machine learning and data science you can probably increase the accuracy of the prediction and forecasts that you are able achieve today using existing BI technologies. This sounds simple, but this is a huge competitive advantage to many businesses. For example, loss prediction and mitigation in insurance and fraud prediction and prevention in banking. As you can imagine, the list goes on and on. However, the skill sets required to solve such prediction problems are highly technical and not easily available.
Process automation problems are usually addressed by all sorts of enterprise software. ERP, CRM, BPM, etc. enterprise software today is pretty good at workflow and rules based process problems, but is still reliant on humans to make inputs into the systems and interpret the output of the systems. This is where manual business processes start to pile up. The field of robotics (Robotic Process Automation) is one such solution to these problems but RPA inherently can only automate tasks in between various systems without any judgement. In other words, RPA to a certain extent is “dumb” automation.
This is where the evolution of Smart Process Automation comes in, using technologies such as NLP data extraction, text analytics, sentiment analysis, etc. If you are already evaluating such technologies, it’s important that you separate out the actual AI implementation problems from AI training problems.
Because the skill sets required for these problems are different. AI training requires a much lower level of skills as compared to AI implementation. AI Training is essentially a labeling of data to describe that data.
Let’s clarify the AI implementation and training problems further.
OCR is one such area where there is plenty of existing OCR platforms which focus on solving data extraction problems using machine learning. Most of these softwares allow the users to teach the system their document structures. So, this job does not require highly technical skill sets.
There is plenty of Chatbot software out there which use some sort of an AI engine from either IBM, Microsoft, Google or Amazon, to figure out user intent from a chat question and then match that to an answer. Once again, in most cases this is a training problem where the Chatbot needs to be trained on a set of questions and configured as per the business’s requirements. This does not require much of a technical skill set either but more linguistic and business analysis skills.
This is where you are trying to solve a problem which is not already solved by an existing off-the-shelf software out there. For e.g. you want to create a system which automatically analyzes the chat or blog content on your website. This problem requires an in-depth problem definition, analyzing existing data sets, algorithm selection and implementation. Such problems require your teams to have data science and machine learning skills which require much higher-level skill sets than that of folks who are just training the AI systems.