AI in content management is a natural fit for automating many tasks, such as search, image and video object recognition, and metadata tagging, but organizations should apply AI carefully to produce the best results.
An enterprise requires efficient, intuitive content organization and access, and AI drives efficiency and intuition. There are times, however, when content management in AI may fall short. AI can make predictions that are disappointing -- or, worse, misleading. The partnership between AI and content management is so new that it's difficult to prevent these issues.
Here are three areas where AI in content management can fail and how smart admins can solve these common issues.
Perhaps the most obvious application of AI in content management is personalization. It's extremely useful for online apps to recognize their users and cater to their interests. Personalization can save time for both the user and app and result in repeat customer visits in e-commerce workflows.
Machine learning can augment customer profiles and purchase history by detecting patterns in preference shifts over time. It can also customize the content that a website or app delivers to returning users to avoid repetition. This is useful because it trims away the time required for a curiosity browse and avoids burnout from regular users. The best content management systems (CMSes) don't just offer content the user is likely to enjoy; they should know what the user wants or needs before the user does, based on general customer behavior with the personal details of a customer's history overlaid.
This can be challenging because the audience of a CMS can vary greatly. An enterprise CMS, for example, is likely to interact with employees at all levels. Online news and shopping sites, however, may only have a handful of annual interactions per user. An AI system needs plenty of opportunity to learn, and a CMS may not provide enough opportunities.
To prevent this issue, content managers can provide extra work for the AI system. When the AI behind the CMS doesn't have enough data from the individual user to make strong predictive choices, it needs an extra layer of input. Content managers should also require the AI system to determine other characteristics of demographically similar users. When the AI system garners additional information about each user from other sources, it can create a stronger profile of the user. And if the CMS makes the wrong predictions, the machine learning system can tweak each individual user's profile and history accordingly.
Tend the shelves
AI can also help to organize and index content. Content managers must meaningfully index data across a variety of relevant dimensions to arrange the fastest, most flexible access to content. Often, this requires content managers to superficially keyword items as they add them to the system.
AI can automate this process through text clustering. Text clustering mines the content itself for keywords and phrases found in other documents and generates clustered content profiles based on the content and document type.
The problem with self-organizing content driven by AI, however, is that there is no inherent way for the AI system to know that its clustering is on target. Employees that receive the content should give feedback to the AI system. They should provide details by answering specific questions, such as: "How will we use this content?" Keywords and phrases from this extra bit of metadata can complete the text clustering circuit and boost accuracy.
Separate AI tools can create metadata from image and video recognition algorithms and tag multimedia content far faster than human curation.
Communication is key
One of the most innovative applications of AI in content management is voice interaction. A CMS serviced by chatbots can particularly benefit from this.
In a typical scenario, a support desk sits atop a vast archive of fixes and procedures to address customer issues spread over a large array of products. A chatbot takes a user's request or complaint over the phone to select the proper content to offer the user. In this scenario, the user's emotional tone matters as much as the words they say. The degree to which the user is upset can affect the selection. An urgent tone of voice, for example, should elicit a quick patch rather than a slow overhaul to meet the user's needs.
The same concept applies to users' text requests; in both cases, the AI system can misread the users' emotions. This problem self-corrects over time, but it's certainly doable to speed up the process. Many companies record support calls so they can train the AI system with varied training sets and supervised learning. Human support personnel should score the AI's content selections. Time spent training is key for organizations that want to customize AI tools to their customers' common questions. Organizations should also retrain the AI engine as product lines evolve, as well as the customers' questions.
AI can perform many impressive tasks -- but not out of the box. This shouldn't be a deal breaker, nor require expensive and time-consuming customization. To make good AI content management great, organizations should commit to that extra mile of further training, additional forethought in process design and new layers of analytics.