Learning When and When Not to Leverage AI in Your Products

You need to go from your house to the Airport. Do you take a Limo or a bike? Of course a Limo? The road is bad and the traffic worse... A Limo is not always the right choice.

Product Managers solve user problems. Sometimes AI is the answer to all your problems. Other times, it is not worth the trouble. 

The question becomes, when and where should we leverage AI in our Products?

My first job as a Product Manager was in an AI based startup whose core competency was image and video based analytics. I was exploring the feasibility and applications in the Security Surveillance space.

What I found surprised me.

One of my visits was to a company helping the Singapore govt with the Surveillance of the country. Singapore has one of the finest infrastructures of the world. And it maintains it beautifully. Littering is a punishable offence. One aspect, hence also becomes ensuring that people don’t throw garbage from the balconies of their highrise buildings.

The few rooms had its walls completely plastered with hundreds of screens. Around 1 person per wall was busily looking at multiple screens at a time trying to detect violations. 24X7 monitoring across thousands of cameras was not an easy task.Was it practical? I would say no, not if done manually. 

So here is how they handled it.

They added pixel monitors on each of the balcony railings within range. Any pixel changes flagged the image and people would set forth to manually analyze them.

There were two main problems. First, this was, of course, not scalable. Second, There were too many false positives. Anyone randomly roaming around in their balcony would trigger the alarm. Needless to say, this was very expensive to implement. That was when I was convinced that an AI could do this better and more effectively. 

Just like this use case, there are many problems that could be solved by AI.

But what are those problems? When do you even dabble with AI to solve your problems.

It is worth a serious consideration because AI is not without its limitations and challenges. AI done wrong often leads to extremely high costs without the added value. Un-Explainability of results and inconsistent responses are other factors often hampering the reliability.

So, what are some guidelines that will help you decide if to go the AI route.

When you visit Amazon app with an intention to buy a product, it is important to Amazon that you make a purchase. With thousands of Products in a single category, how does Amazon shows you the product that you will like? It hence utilizes your behavioral patterns, the characteristic of products, and other parameter to predict the products you are likely to purchase. It can do so without AI as well, but then keeping a track of your changing preferences, purchasing patterns need constant adaptation. AI hence solves this problem beautifully.

Majority of Products  whose bread and butter depend on recommendations leverage AI to satisfy their users. Other examples include OTT like Netflix.

There is a high probability that whatever you say, Alexa will understand it. How does Alexa understand you? How does it interact with you in a human fashion? NLP is the field which explores how machines can understand and respond to human languages. The field of AI is relatively new, exciting and undergoing rapid developments owing to the extensive research.

Finding your older photos based on search keywords has become so easy with Google Photos.

A single word “Beach” reveals all the photos including beaches from your album. How does it happen? When you train a model with millions of pictures of beach, it learns that a beach has water, sand, and maybe coconut trees.

Now, can this be done without AI? Yes, conventional image processing methods would work as well. The accuracy is another matter however.

Classification applies whenever you need to throw things into multiple buckets. You can classify utterances into intents, manufactured bulbs into good quality vs garbage, news based on topics, and so on. Classification is also what keeps your inbox clean by identifying and segregating spam emails.

Clustering helps group similar objects. Everyday you read Google News, there are too many news items around the same topic. But they all appear to you in groups. This helps users consume content better.

Clustering is what helps Google segregate all information on the Web, and Banks identify credit card frauds.

What will be the price of your house in 2023?

There are multiple factors that will affect the price. Some of them are size of house, location, age, brand, economic condition of city, country and so on. If you give the corresponding values for a million houses, and their prices, the model will learn. It will create opinions like size of the house is more important than the brand, or age is less considered than brand.
It can hence predict the price of your house in 2023.

Regression is an important area with applications like predicting when the corona cases will peak, risk associated with particular investments, life expectancy, and so on.

Follow me on Twitter and connect with me on LinkedIn here.