Before you invest in AI, assess your AI-readiness

Readying data for AI

AI can be powerful, but only if it has the right fuel.

Getting that quality data out of enterprise content isn’t automatic — it has to be readied. Enterprise content management providers are uniquely positioned to help customers transform their data for the endeavor.

“AI-ready content is probably one of the bigger areas that organizations need to focus on,” Tom Davis, AVP of intelligence for Hyland, said. “There's this misconception that you can just take everything in every repository and database and chuck it into the AI engine, and it's going to learn everything. That’s just not the case.”

First, Davis said, it’s not economically feasible, and second, AI models need to be trained with the right data.

“There’s a garbage-in, garbage-out concept,” David said. “If you throw everything in, you’re not going to see accurate results.”

Making data AI-ready

To capitalize on the power of AI, an enterprise’s data has to be ready for the machine. There’s a data translation that needs to take place; content that was created for human consumption needs to be processed for a computer.

“Consider a document full of text and images,” Tiago Cardoso, AI product manager at Hyland, said. “When retrieving content for an LLM, we need to understand its meaning and context and select only the relevant aspects. For fine-tuning, we separate content based on model inputs without losing meaning or organization.”

Additionally, enterprises need to select the right content to feed the machine. Starting a new model is a big lift, requiring the right data to train, test and fine-tune the system.

“It’s a real science,” Davis said. “There are ways you can overfit or underfit the model, and with too much of the wrong information, answers start declining.”

Once the right content is ready for machines, organizations can start implementing the impactful services AI offers.

Unstructured data enters the chat

Content, in both structured and unstructured formats, holds the important data an enterprise collects. However, research suggests less than 10% of unstructured data is being extrapolated to be used in business processes or decision-making, despite 80% of data sources being unstructured.

Imagine freeing all that inaccessible, unused data using generative AI. With unstructured data predicted to grow, the business implications of being able to fully capitalize on the data you already own are astounding.

Once AI can access and activate those data sources, organizations can draw insights at scale and benefit from the semantic relationships that AI can connect.

For example, more intelligent search results is made possible with AI’s ability to mine and interpret data not just from traditional structured sources, but also from the trickier unstructured documents. Without depending on narrowly defined metadata, an organization can get a fuller picture of the relationships between previously unconnected data points, making it possible to find things based on relationships rather than specific search criteria.

AI-readiness checklist

Hyland’s expertise in both content and AI gives our Intelligence team members a unique view into AI-readiness. They’ve developed a framework for its evaluation using five pillars:

1. Infrastructure

Infrastructure speaks to a technical readiness. To leverage AI and do it securely, organizations need a robust, comprehensive infrastructure to manage data in the cloud, as well as the right tools to do the work. The databases where information is stored need to be secure, compliant and scalable — ready for business booms or declines.

“In the right infrastructure, you’re going to house the components that process your content to make it AI-ready,” Davis said. “When you do that, you create meaningful connections to that content — which is stored securely in a database — and make it readily accessible to AI services.”

This is often difficult for organizations to do on their own, and that’s why Hyland created Hyland Experience, Davis said.

2. AI-ready content

Having AI-ready content is another technical hurdle to clear for being ready for AI. Content needs to be curated for high-quality AI output.

“There are layers to developing the appropriate corpus of data,” Davis said. “Each piece of a document needs to be broken out and sent for training. This creates embeddings that are used downstream in the gen AI space. It’s a process that organizations need to understand and be ready for.”

For example, if an HR department were readying its content for an AI-powered platform, it might identify certain records with bad labels or poor tagging to leave out. Once the quality content is loaded in, the model’s algorithm will run, the team can train the system by ranking output and the AI starts to understand what the content intentions are. Then, it can start improving on its own.

3. Governance

Governance overlaps both states of technical readiness and business readiness. Organizations have a great responsibility when it comes to governing AI. From monitoring data access and detecting malicious incursions to ensuring responsible AI practices throughout the organization, having strict standards enables organizations to implement AI safely and securely.

When incorporating AI into products and daily operations, organizations should develop clear guidelines for product teams and employees to mitigate AI-related risks in different aspects of the business.

An AI council can also help oversee the incorporation and implementation of AI, and ensure that guidelines reflect technological advancements and law changes.

Adhering to an organization's security and compliance standards is essential. Given AI's heavy reliance on data, having robust policies and the right technical tools in place provide a strong foundation for a secure AI implementation.

4. Ethics

“Ethics is something we take very seriously,” Davis said. “You need to have an ethical foundation in place — it’s critical to deliver on responsible AI.”

Davis said ethics is a common point of concern among customers and in RFPs. Honesty, bias and explainability are all facets of this component of business readiness.

“If an AI engine is going to make a decision or a recommendation, customers need to be able to understand how it came to that conclusion and what benchmarks and evaluations are showing these conclusions as accurate,” Davis said. “Being ready from an ethics standpoint means having guardrails in place.”

Hyland’s AI standards include transparency, data ownership, honesty, verifiable results, privacy and security, and governance.

AI-ready businesses can support quality AI outputs with ethical data, as well as monitor for things like bias. AI models also need to be able to defend against situations in which users might try to use disingenuous prompts to receive information they shouldn’t have access to.

The implications are very real for many industries, notably financial services, insurance and higher ed. From historical redlining practices in lending to fraudulent insurance claims and student evaluations, the stakes are high, and the data that feeds an AI model must be protected against bias and tainted data.

5. Skills

With AI capabilities popping up across new and familiar technologies in every industry, organizations can’t fully realize their AI ambitions without the right people to take them to the finish line. The competition for AI skills talent is fierce and has created a talent gap ranging from engineering and data scientists to business users who need leverageable AI know-how. Organizations are eager to bring on highly trained faces, but AI experts point to upskilling as another route.

“We believe everyone in the organization needs to be leveled up from a knowledge perspective on AI,” Davis said. “We're going to see businesses focus more on budgets, as well as the effort and time it takes to level up their employees, so they are able to take advantage of AI and use it to get an ROI.”

Snapshot: The AI-ready enterprise

Once an enterprise achieves AI-readiness, the exciting work of building an AI-powered workplace launches. New processes — new possibilities, even — are on the table. At a high level, employees and customers should benefit from greater efficiency and visibility, including:

• Quick delivery of accurate business information
• Accelerated, quality decision-making
• Ability to outsource time-consuming intelligent work to AI
• Support for and amplification of the workforce
• More rewarding customer experiences

Additionally, Davis notes these areas as particularly talked about in future-state visions.

Comprehensive intelligent search

“A more intelligent search is one of the things people most want out of AI,” Davis said. “They want to be able to do their search with a natural language prompt, in a conversational way. To ask for some information and get the right answer back, even if the data is in multiple places, to get recommendations, guidance or even actionable insight to work from.”

AI-powered platforms organize content in a more human-like way by going beyond the narrow data labels and filters of legacy systems. Systems like Hyland Experience can intuit more insightful relationships among data points, no matter where the content is housed. Additionally, generative AI can jump in to take the search to the next level by providing insight and answers.

Intelligent search in practice: Imagine being tasked with the review of 1,000 legal contracts, searching for language around the topic of land ownership. Instead of searching by keywords or metadata tags (or worse, getting out a highlighter after a trip to the printer), the AI-empowered worker can direct the system: “What are the key terms and conditions related to land ownership transfer in our standard property contracts?” Within seconds, results are retrieved and the employee can review and take the most valuable information.

“AI delivers the ability to use an intelligent virtual assistant that can find great information, make recommendations, provide guidance, summarize things,” Davis said. “People want to do this simply, using natural language, and they want to look across multiple repositories to find information in other systems.”

Smarter automation

Because of Hyland’s native automation capabilities, customers are dialed in on the potential to unlock the rest of their data and create relationships that can drive new business processes.

“Folks see the value of using AI as autonomous agents behind the scenes to help move processes forward,” Davis said. “AI can play a part in all of that, the processing and the automation piece. Probably most importantly, I think people really understand that a lot of the work they do could be automated with the right information.”

And it's not just pure automation of processes. AI can amplify and augment the people who work in those processes to help them be faster and more effective. AI models can even understand how a process works and recommend process flow changes based on what it’s learned.

Intelligence beyond metadata

With AI-ready content, the entire information life cycle gets an upgrade. The relationships built between data points and understood by AI creates opportunities for enhanced content management, processes, search and governance. For example, AI enriches your workstream via:

• Content management: A higher volume and deeper value of information can be organized, recognized, extracted and activated from unstructured content, leading to more comprehensive content management and better decision-making.
• Processes: More complex processes can be automated with AI’s ability to make sense of connected and relevant data from unstructured content, in addition to its ability to use human-like intelligence to propel workflows.

AI in different roles

Leveraging AI effectively across an enterprise takes time, education and innovation. Many organizations ramp up their AI use as they gain confidence, competence and creativity. Davis shared three scenarios of AI execution:

Level 1: AI supporting humans
Consider an existing process in which an employee looks at everything and makes the decision. AI can step in to support by providing answers to questions about content. Now, instead of an employee reading thousands of documents, AI can summarize the content and provide the worker the information they need to make a quick, well-informed decision.

Level 2: AI automates processes under human review
In this scenario, the process is set up for the AI model to do the leg work and present its findings to a highly skilled employee for review. For example, a process may have five decisions that can be automated by AI. As the AI model works through those decisions, it can go back for review, but it eventually works through the process. The model’s result goes to the highly trained employee for review. This augmentation of intelligent work and skilled review drives efficiency and improves the quality of work humans spend time on.

Level 3: 100% AI-driven
When models reach a 99% accuracy rate, it’s considered fully operational. Of course, the need for governance and quality assurance is still there, but at this stage, the AI is a fully automated part of the team.