The world’s leading publication for data science, AI, and ML professionals.

AI powered business decision making

Addressing the dilemma in choosing the right business problems to be solved by AI..

source:pixabay
source:pixabay

Why does it matter?

It may take intelligence to solve problems but it requires wisdom to figure out which problems are worth solving. The introduction of AI has provided a paradigm shift in the computing world. Many forward thinking businesses have started to reinvent themselves and have started using AI deep into their solution building while designing systems. The hype created around AI has forced several organizations to re-plan their tech strategy and contemplate on what relevant problems to solve with it to add value to their business. Marketing content is being re-created and sales pitch decks are being re-designed. Building AI is expensive and requires collaboration amongst all teams involved with the application. The journey must start with an intent to solve a relevant business problem through AI. Trying to add AI by any means in the application just to join the AI revolution is as risky as speeding without a seat belt on. It most certainly will lead to big time failure and huge loss of time and effort. Funnily enough in some cases, a simple rule based configuration is passed off as an "AI powered feature" to gather attention or improve sales. In the midst of all this eccentricity, it is prudent for businesses to know when to use AI and more importantly when not to.

Digging a little deeper

Multiple parallels can be drawn on what exactly AI is but it is certainly much different from traditional programming. Rather than coding the rules an AI model derives insights from the data on its own through training and without explicitly being programmed. For instance:

Rule based (non-AI) system notification: "Add money to your wallet" if balance is less than 1000 INR.

AI based system notification: "Add xyz INR to your wallet as you may need it tomorrow".

The recommended amount is based on past spending patterns to ensure the user never runs out of money in wallet. Another example could be to turn on the notifications for emails to appear on the screen that may be of high importance to the user rather than showing it for every email.

The later one provides a more personalized experience and is so much more convenient for the user. This is how AI adds value to even the most simple business use cases. One can imagine how difficult it would be to develop an understanding of the spending patterns of each user in the application or to maintain rules to send out notifications.

AI has solved multiple problems related to computer vision really well. For instance, what is it that makes a "dog" look so and how is that different from a "cat". Explicitly coding this would require defining just too many rules in the system which leads to chaos. AI techniques can be used to train or learn what features of a "dog" make it look so by showing it a bunch of photos of "dogs". Once properly trained, AI software is able to identify a "dog" in a picture without having explicitly been programmed to do so because it has developed an understanding of how a "dog" looks like. It becomes an even more complex problem if one has to identify the type or breed of "dog" in the picture as well. AI can help there too and with quality data and good training, it may become exceedingly efficient at it. This can be applied to support functionalities like face recognition for attendance capturing or for alternate user authentication method. This can be applied to classification or categorization of the presented documents automatically. In general terms this may also be referred to as object classification.

AI is most definitely an overkill to take simple decisions that can be taken with simple conditional construct (if-else) in a software program e.g. "Deliver the order if the payment is done" or "add late fee charges, if the bill payment is delayed". AI doesn’t add any value to these obvious decisions. However, AI can help in predicting the chances of a consumer defaulting on a bill payment basis past behavior. Reminders can be sent well in advance to avoid late fee charges.

One should reconsider using AI in scenarios where the expected output needs to be definitive in nature. For instance, performing a look up (search basis a key) on a table. It will always return the same output given a certain input search key e.g. Fetch salary of an employee basis employee ID. AI is not going to make it more accurate or faster. However, AI can be used to perform a smarter search on the input search key by ignoring spelling mistakes or synonyms based search or basis the meaning of the input sentence(s). This can be used in "conversational chatbots" to understand the input query of a user and look for a relevant answer from a list of FAQs in a repository.

Descriptive vs Cognitive dilemma

Charts or visualizations depicting top 10 customers based on the orders placed or top 10 defaulters etc. are only descriptive business use cases and not AI. AI can be used to predict the chances of a customer defaulting or chances of gaining a spot in the top 10 on the basis of recent change in trend. This adds huge business value and a good understanding of the customers’ interaction with the system. This is the difference between descriptive and cognitive business use cases.

Summary: Putting it all together

AI is quickly becoming a key to solving a myriad of problems across multiple industries. It is expected to solve complex optimization problems and get better at it over time with access to more quality data for training. Simply put, AI is meant to solve complex business problems by identifying patterns in the data using algorithms and help decision making process faster. **** There can be many more examples where AI can give reasonable results in the area of risk estimation, trend forecasting scoring engines, customer segmentation etc. Classification of objects from images, image understanding or captioning, face recognition, text understanding, sentiment analysis and recommendation engine. Generative models can be used to create art, music, images, videos or text in various styles. These are all examples of problems that AI addresses really well.

However, there must be a strategy to first figure out a business problem relevant to be solved with AI e.g. Should recommendation engine be used for cross-selling or upselling to improve sales? Or should it be used to recommend prioritization of requests of certain customers to be processed at the back-end? It must also be followed by a data strategy. Any AI model needs a lot of quality training data devoid of any anomalies to give meaningful results. Iterative process needs to be followed to train an AI model well. Hence, adopting AI is a long journey and it requires constant maintenance and upkeep.

It is important to note that rote memorization of the facts is not the purpose of AI, even a parrot can do that without developing any understanding of the words it utters. AI brings a new set of ideas to solve complex problems where understanding of the past is generalized to ascertain the future for quick decision making. Perhaps, that’s what intelligence means..


Related Articles