I was recently asked by one of my colleagues to give a talk on how machine learning could be implemented in an organization in a way that would provide some quick wins for the organization and that would provide some immediate benefits. This was an interesting challenge as the colleague had mentioned that non-technical folks may not know the power of machine learning or even that they should seriously invest in machine learning technologies in order to remain relevant until they they see some quick wins in their organizations. Here is a list of ideas I came up with in order to implement machine learning in an organization. I will use AI and machine learning synonymously from this point on even though they are not the same.
1 – Help-Desk Augmentation
Ok so I don’t have any real data to back this claim up so please take this with a grain of salt but since I (at most times in my career) have only been one or two levels away from the help-desk support technicians who are facing customers every day, there seems to be some requests that are always repeated. These include things like password resets, accidentally deleted files and requests for information. Despite the fact that most organizations usually have a self-service password reset mechanism in place, users usually end up calling the help desk first. Go ahead and Google “Most Common Help Desk Requests” and you will see what I mean. Let’s suppose that my assertion is true and that most help desk support requests are repeated requests, a very simple and quick win for an IT help desk can be to implement an AI chat bot that can be thought of the Tier 0 on the help desk. Since modern chat-bot frameworks can easily understand natural language, a very quick win might be to utilize AI chat-bot to fulfill the repeated requests. For example, if a user needs to reset a password, the chat-bot can identify the users security credentials and reset the password to a temporary password without a human ever getting involved or simply lookup information in an FAQ or knowledge base and reply to a users query. Implementing a Tier 0 AI chat-bot can be quick-win for any organization with measurable decrease in hold times and increase in ticket closure rates.
2 – Recommendation System
Another quick-win in most organization can be a recommendation system. Most non-technical management staff already understand recommendation systems since they utilize Netflix, YouTube and Pandora on a daily basis. Most customer facing or even internal line-of-business system can benefit from the creation of a recommendation system. For example, one of the ways I recommended AI can be used at a past client was to implement a recommendation for conducting audits. In this case the customer had a custom-built application in which they tracked the audits they conducted in the past, all audit results, artifacts provided by the target institutions, etc in their audit tracking system. A very quick win would have been to create a recommendation system for the audit system that would provide the auditor a list of past audits that have been conducted for similar institution types and their results. This way a new auditor or even a season veteran auditor can see past trends, recent patterns and other key information that would provide valuable insights. Also, any customer facing system where the use case is either shopping, media consumption or even information management can benefit from recommendation systems. There recommendation systems can provide the user recommendation based on what the user has done in the past, what other similar users are doing.
3 – Document Understanding and Semantic Search
According to Wikipedia somewhere between 80 to 90 % of all data in a given organization is unstructured. This includes things like emails, documents, presentations, videos etc. The power of semantic search and documentation understanding can be explained in a very simple example. A sentence like “I am going to visit India” to a computer is just a bunch of character but to AI, this sentence means much more. AI understand that visit is a verb and usually describes a journey or travel where as India is not a string but represents a physical location in the world. Now imagine the insight that are currently hiding in all the unstructured data your organization produces on an annual basis.
A very simple use case for document understanding is what is called semantic search in which an AI can injest information such as a document in unstructred form and then present insight to the user and allow the user to ask questions. For example, a 300 page manual containing instructions on how to install the latest version of a software might be difficult for a human to read but a machine can read the document and the human can ask questions such as “What is the minimum RAM required for me to be able install this software?”. The AI can answer this question. The use cases for this are so vast that they might be hard to list here but a a very simple use case is to gain insight from very large documents. I wish I had an AI assistant that could help me read RFQ and RFC documents that I used to have to do for one of my past jobs.
Here is a video by IBM going through a tehcnical demo of what is possible. This is fascinating stuff.
4 – Machine Learning for Data Analytics
Though structured data and business intelligence usually provide very powerful insights to manage of an organization, sometimes, the organization does know the right question to ask. Other times, data can reveal insight that might be surprising.
An example being that of strawberry pop-tarts and Walmart. Walmart started using predictive machine learning analytics a while ago and a surprising insight they found regarding items consumers purchase when a hurricane approaching was that a hot ticket item was strawberry pop-tarts. You can read the reason and behind this in this article. How is this possible? Well the data does not lie and has no bias in the matter. If pop-tarts really are the best item to stock, then the data will expose that. This is a quick win because surprisingly, many answers machine learning reveals runs counter to human intuition at first until it is exposed by the data and then it finally makes sense.
So, exposing structured data to machine learning algorithms to discover patterns anomalies and clusters in the data can provide surprising insight to management.
That’s it folks. What have been some AI quick wins you have used in your organization? Comment below to let me know.