AI Powered Applications

Consider how you’d like your devices and applications that you use to become smarter. They would anticipate your needs, curate your experience, understand the context that you’re working in, and recognize the sequence of the tasks that you would like to accomplish.

Artificial Intelligence (AI) is going to bring intelligence into our applications, which means that there will be a massive paradigm shift in how we conceive solutions and implement applications. I am going to explain this by comparing a traditional engineering project with an AI power project.

Let me provide an abbreviated, but standard approach for building applications. In a traditional project, we gather information to establish the business goals and the problem domain. As we create this documentation, we begin to develop a functional specification, an architecture, and eventually we build the system. The solution will have business logic that defines the experience for the end user and the rules for how the solution operates.

Business logic is typically defined as a set of static rules and are not often changed or updated because the current paradigm for building applications includes the imperative of standardization and consistency. What is smart about these solutions is that they provide output reports and data that then become the basis for human intervention. The word Knowledge Workers was coined in the 90s to identify people within a company that conduct reporting, analysis, and recommendations to improve the business. The feedback loop of continuous improvement operates outside the application and in the hands of the knowledge workers, domain experts and management to act upon.

The new model for building AI powered applications will embed the continuous feedback loop into the application. The application will be trained with a Machine Learning (ML) model that will make the necessary adjustments to improve efficacy. In some cases, those changes will be entirely ML driven and in other cases the application will provide recommendations to the knowledge workers and management for implementation. The point being is that there will always be some kind of human intervention before AI fundamentally changes a business process or application.

Let’s walk through the model for building an AI powered application compared to a traditional one. The initial phase of gathering information, ratifying the business goals and problem domain would remain the same, however it would culminate in a definition of a mathematical model of the business problem. The mathematical model is meant to capture the key points of variability and leverage within a business process. Every variable in the model and the data set(s) that support those variables represent opportunities for ML improvements. The next step is to build a general-purpose learning model build upon the available data sets, often this will be based on the historical data within a company. All the data from the line of business application, all the systems that interact with the application and any third-party data sets that are relevant. We would set up a general-purpose learning model and be training it on the historical combination of inputs and outputs against the achievement of business goals. We are teaching the system to become familiar with the patterns of success and to look for areas of optimization. This is the basis for embedding the feedback loop into the application. We’d then define the business logic and determine which aspects of the logic can be modified based on the ML solution identifying patterns that are most likely to result in a positive business outcome. With this in mind, the rest of the project is pretty standard, we define a functional specification, an architecture, and a build plan.

The best way to think of what is trying to be accomplished is creating dynamic business logic that is powered by a ML solution that optimizes patterns within the application or experience that result in a business improvement.

In a real-world marketing scenario, we know that the order in which a subject is learned is directly related to how people develop knowledge. Think of school, they teach you one step at a time, building a foundation and graduating you to the next class. Now let’s take a considered sales cycle (you need to do a bunch of research about a product to make a decision.) Each customer segment is going to need content sequenced to them differently to account for different levels of knowledge. Each customer segment may want to experience content in different formats, YouTube video vs. whitepaper for example, and the context for each customer segment will be different, “in a hurry”, “late at night”, “on an iPad, on a phone”, “ready to buy now” … etc. The consumer experience needs to be adjusted to suit the individual circumstances for each customer segment. With that in mind, what we would be doing is allowing the ML model to make decisions about how to provide you an experience that is most likely to result in a purchase. It will choose the content, style and user experience that is most effective for each audience. We will swap the static information architecture and then choose your own adventure digital experience into a Machine Learning curated experience that gets better and better at understanding the needs of customer segments over time.

I select a marketing example because we are all familiar with patterns of personalization, but don’t be mistaken into thinking this is just a better way to solve marketing problems. It is a better way to solve all business problems. There is not a company or market on the planet that will not be affected by AI. The interesting part of this technology change is that because the AI solutions learn and get better over time. There will be an early mover advantage that may be competitively insurmountable. Companies will be competing on the timing and quality of AI solutions within the business. Executed properly, profit margins and growth rates will rise precipitously, allowing the early adopters to accumulate capital for expansion and industry consolidation. We are at the very beginning of this new era; it’s going to be great to see what lies ahead.