AI has taken the business world by storm. Seemingly overnight, a multitude of apps and platforms have begun boasting of their AI-drive capabilities. Even dishwashers are using AI. The change has occurred so abruptly that many businesses have found themselves behind, lacking an AI strategy, and vulnerable to falling behind more proactive competitors.
It has left managers and executives across the globe wondering, how should we be using AI? What needs do we have that could be addressed with these novel techniques?
Of course, the answer to those questions is complex. There is no simple recipe for AI excellence. How the technology can help depends a lot on your industry, company, and your team’s responsibilities.
To pinpoint exactly what AI can do for you, following the six steps below. The first three help you determine whether AI is a good fit; the second three help you determine which AI techniques would be appropriate.
Step 1: Determine if you are trying to process information
AI can’t do everything. Sadly, we know of no apps that can get your teen to clean her room or walk the dog at 6 am in a snowstorm. What it’s primarily good at is processing large volumes of information and generating insights. It can quickly and effortlessly handle tasks that would be overly onerous or tedious for humans. For instance, AI can find trends in your business’s transaction records, allowing you to detect fraud, manage inventory, segment your customers, or personalize marketing.
Step 2: Figure out what kinds of output you need
In other words, what problem are you trying to solve? Clearly identifying the outputs you seek will let you determine the type of AI solution you need and how you'll use the insights it generates. There’s a wide range of possibilities. If you work in a bank, you would be able to use AI to figure out which customers are likely to default on their loans, based on trends the algorithm can identify by analyzing enormous quantities of customer data.
Step 3: Identify how the outputs will be used
Evaluating how the outputs will be used enables you to understand the risks and benefits of the application. In a previous article, we mentioned that an airline paid a fine after its chatbot misled a customer about bereavement fares. In using AI to power its chatbot, an airline can realize cost efficiencies from replacing human labor and improve its customer experience with faster, 24/7 chatbot access. But the airline assumes the risk from the mistakes the AI algorithm makes. That may be an acceptable risk for the airline, but if the outputs were to be used to make healthcare or financial decisions, the implications could be far more serious.
Step 4: Define the modality of the information
This can get technical, but knowing the modality (or type) of data you seek insights from can allow you to determine the type of AI model you can use. There are different models for images, unsorted text, and tabular data (the latter is akin to what you would find in an Excel file or database). The modality also helps pinpoint other technical requirements, such as how the data needs to be prepared before AI can process it. Audio data may need to be transformed into spectrograms, and image data could require resizing and normalization.
Step 5: Specify the outputs you are trying to create
Just as the inputs will influence how you design your AI solution, so will the outputs. Your goal may be to predict maintenance needs, to summarize large quantities of text, or to make recommendations for products that a certain customer may want, based on previous purchases. Each of these use cases would influence the type of AI model you use.
Step 6: Determine the feedback you will provide after an output
Some AI models learn on their own; others use feedback to improve. If a user acts on a certain type of video recommendation but ignores another, the AI model underpinning the recommendation will learn from this feedback. This is called supervised learning. It is particularly useful in applications that involve classification. Some AI solutions, known as unsupervised models, can learn entirely on their own. An unsupervised model can detect unusual network activity that could signal a cyberattack.
Determining how AI can help your business may be intimidating at first, but like most tasks, breaking it down into manageable components can go a long way toward helping you find appropriate solutions. By addressing these six steps, you’ll be better able to discuss the possibilities with your IT team, ensuring that your business can harness the benefits from this fast-moving technology.
