The 4 Product Patterns in AI

A primer for identifying AI opportunities at your company

Get a free early copy of my new book

You’re wondering “what do I need to know about AI?” but you don’t have time for courses & blogs. You need practical, revenue-generating ideas — not research.

How to Stand Up and AI Practice gives specific, no-bs advice for product & corporate innovation teams. Get it here on LinkedIn.

Wondering, “what can we do with AI”?

I’ve been slowly building a boutique AI consulting practice this year. My #1 challenge has been answering this question: “What can we do with AI?”

CTOs, corporate innovation teams, product managers all have this question. I begin answering by learning about the operations, customers, and data. Then I try to apply on of these 4 AI product patterns.

Why 4 AI product patterns

The pace of AI is overwhelming — even AI researchers cannot keep up with the volume of discoveries hitting arXiv daily. Add in startup press releases, Google research, VC blogs, and business press and your average product team can’t make sense it.

Fortunately for busy product people most of this research isn’t ready for the enterprise. You just need to a basic understanding of 4 AI product patterns to begin discovering revenue-generating AI opportunities.

Presented in order of product readiness.

Pattern 1 — Collaborative filtering

When you visit services like Netflix you see movie recommendations.

Collaborative filters can be used to make product recommendations.

Because you watched Toys you may like … Twins … Crazy on the Outside … Street Fighter … The Chosen One …

How does Netflix know what you like? The company tracks movies you’ve recommended and identifies other people like you. The suggestions are based on the movies similar people like. This is known as a collaborative filter.

Collaborative filtering is useful any time you have lots of data about customers and want to predict what an individual customer will do or buy. Examples:

• Online ads. Facebook and Google ads are the killer app for AI
• Product recommendations
• Marketing promotions
• Loyalty/rewards programs
• Fraud detection
• Cyber security

Collaborative filters have been around a long time and you don’t need AI to build them. AI allows you to build incredibly smart ones which automatically discover associations people could not.

Take the Netflix example. What do Toys and Street Fighter have in common? After all, Toys is an animated kids movie and Street Fighter is an action movie.

Answer: we don’t know. The AI is building more complex associations which consider the movie and the viewer.

Pattern 2 — Computer vision

Computer vision is a programming technique of analyzing digital images and making decisions. As of 2014 it didn’t work very well — that’s why you haven’t seen mainstream facial recognition solutions despite decades of research and false promises.

AI is creating exponential improvements in computer vision such that many previously complex tasks are now trivially easy.

3 years ago state-of-the art computer vision algorithms could achieve ~80% accuracy in determining whether an image contains a dog or a cat. Today I can get >97% accuracy by downloading proven algorithms.

Simple image classification problems have now reached human-level accuracy.

Computer vision is useful whenever images can be used to make decisions. For example:

• Facial recognition. Citibank just launched a facial recognition tool for its iPhone mobile banking app.
• Insurance fraud detection. Insurance companies are asking customers to take pictures of automobiles involved in accidents to validate the customer’s claims.
• Product classifications. Given a product image computer vision can be used to classify it as a shirt, soccer ball, etc.
• Automated product description generation.

Pattern 3 — Natural language processing (NLP)

Natural language processing is programming technique of analyzing text and making decisions. Language is more complex and less structured than images, so AI advances in NLP are a few years behind computer vision.

But in 2017 this is all changing very fast as companies like Google, Amazon, Baidu, and IBM make fundamental breakthroughs every few weeks. I have shocked a few clients with the power of existing NLP technology.

Source: Google

Discerning human meaning or intent from language is still a very, very hard problem — even people can’t do it very well. Let Amazon, Apple, Google, and Facebook work on these problems.

Your best opportunities for NLP are in leveraging APIs from companies like Google or Amazon Lex to supplement your existing products or build new ones. Do you use audio anywhere in your organization? Do you have employees making decisions based on what they hear from customers? Are employees reading written text and making decisions?

These are all opportunities for NLP. Some examples:

• Automated cognitive assessments
• Identifying fraudulent claims or anomalies in text.
• Reviewing product descriptions for accuracy.

Here’s a tip — go buy an Amazon Alexa if you don’t already have one. You’ll quickly get a sense for what this technology can (and can’t) do.

Pattern 4 — Next-in-sequence predictions

The final pattern isn’t a topic for AI researchers. I have identified product opportunities where people are predicting complex scenarios after a series of events. For example, your doctor makes a prescription after recording symptoms, running tests, talking to you, and testing various prescriptions.

I call these next-in-sequence predictions. Next-in-sequence is usually implemented in conjunction with other fundamental AI techniques like computer vision and NLP.

Examples are:

• Autocompleting words or sentences, especially on mobile devices.
• Prompting customers to purchase a product before they express interest.
• Customer service.
• Time-based anomaly detection.

Look for situations where people are trying to make a decision after observing a series of events and you will begin identifying next-in-sequence opportunities.

Emerging patterns

I try to identify practical patterns based on what technology is getting deployed. Here are other trends I’m watching:

• Reinforcement learning techniques in robotics and games.
• Unsupervised learning techniques through generative adversarial networks

Product teams are tracking the research in these areas closely for potential opportunities.

Wondering how to get started with AI? Get an advanced free copy of my new book on LinkedIn.