Capitalizing on the Most Beneficial Use Cases of AI
Over 90% of companies agree that artificial intelligence (AI) is one of the most important technologies for 2019, according to research firm Vanson Bourne. So, how can AI be implemented to deliver more intelligent experiences and drive new efficiency gains? The research shows that most organizations consider customer communications, predicting customer behavior, and CRM as being among their top three most beneficial use cases of AI.
Here are some first-hand tips for practically applying these use cases within your organization:
One of the greatest applications of AI we’re seeing for customer communications is personal virtual assistant technology (either the devices themselves or their backend systems) for customer service purposes as a new channel of communication. For example, you could take the machine learning and natural language processing (NPL) technology behind Alexa (Lex) and use it to build customer service automation platforms similar to traditional voice IVR. Amazon, for example, leverages Lex to provide conversational interfaces to user’s applications powered by the same deep learning technologies as Alexa.
Outside of personal virtual assistants, organizations can apply machine learning around customer conversations—whether text- or voice-based—to better understand the types of experiences customers are having in a specific communication channel also known as mining for intent. This can help brands pinpoint the top things customers are saying or top issues being reached out about. Machine learning technology—as part of the backend systems of AI-enabled devices—can rapidly provide analysis around communication and how customers are interacting to identify important patterns and surface problems more immediately for faster resolution.
Predicting Customer Behavior
One of the greatest challenges of adopting AI is that it generally requires a three-step process. The first, and arguably most difficult, step involves accurately pulling in data sets. What is the appropriate relevant data that will help build the model to achieve the desired outcome being pursued? Once you have identified the relevant data sets, it must be determined how the model will be built so that the data yields the desired outcome (or not). This can be verified by how the model reflects the trends and patterns leading to the desired outcomes (or not). At this point (or sometimes done just prior to this) it is important to determine if there is bias in the data that is being reflected in the model that is resulting in outcomes that do not really reflect a more full range of potential input. You need to avoid the garbage in garbage out phenomena.
Once you have effectively completed this three-step process you can start driving what we at Avaya call “next best action” (a.k.a. deciding what action to take for a particular customer based on a particular issue). For example, you might determine that because a customer’s bill is over by a certain amount of dollars every month or has reached out to customer support a certain amount of times per month that they are at risk for retention. Informed by your AI-based analyses, you can choose to automatically route these customers to your retention-risk team.
After this comes measurement for continual improvement. For example, you might discover that you save eight out of every 10 customers sent to your retention-risk team. This is solid data that can be used to effectively enact new and improved strategies.
CRM technology remains fragmented, with the average organization operating around three or four platforms. There’s a massive amount of consolidation that needs to happen to overcome the greatest challenge of applying AI in this space: ensuring an accurate system of record. If data is questionable or inaccurate, you’ll need to go through a data sanitizing exercise to ensure everything is correct and up-to-date. It’s critical that companies do this, as CRM is one of, if not the greatest source of data for AI systems.
As mentioned previously, it is important to recognize and then correct for bias inherent in the system. The choice of what data is collected, the methods of collecting, the correlations that are assumed in the organization of the data all can lead to bias or hidden relationships that can drive the resulting model. However, that hidden relationship may drive the model inappropriately resulting in an unsatisfactory result. Any model development that does not undergo a bias detection (and correction) step likely will result in an unsatisfactory result.
Ready to get started? Schedule a Discovery Workshop with Avaya Professional Services. We’ll help you identify the key areas of your organization where AI can have the greatest impact upfront. Learn more.