Product Data and Knowledge Strategies to Enable an AI-Powered Enterprise Webinar Recap

In our latest webinar, Ashok Kartham, Founder and CEO of Circuitry.ai, was joined by Seth Earley, CEO of Earley Information Sciences and author of the award-winning book AI Strategy: " The AI-Powered Enterprise: Harness the Power of Ontologies to Make Your Business Smarter, Faster, and More Profitable." We discussed how companies the importance of organized data, knowledge architectures, and ontologies for successful AI initiatives. 

Keep reading for key takeaways from the webinar or click here to watch the recording on-demand. 

The importance of personalization at scale 

The ultimate goal of AI is to deliver the right information to users in the context they need.  This is achieved by personalization.  

Personalization is more than just a buzzword; it’s a critical component of modern business strategy. Personalization at scale means understanding the user and the signals they give and being able to provide them with the products they want when they need them. Think of it as reading their digital body language, seeing what they do online, and responding accordingly. 

 A big part of meaningful personalization includes creating high-fidelity journey models that represent a user’s intents, expectations, and needs in data terms. This model allows you to translate the user’s needs into information your systems can use and respond to. 

Organizational maturity  

Achieving success with AI-powered personalization requires maturity in several areas:  

• Product data operations: Having high-quality, well-structured product data is essential. This involves consistent data practices and governance to ensure accuracy and reliability. 

• User experience: Developing a detailed journey model helps to understand and anticipate user needs at different stages. 

• Content operations: involve systematic processes for creating, managing, and distributing content that will engage the user.  

• Enterprise knowledge: Effective knowledge management practices ensure that valuable insights and information are captured, organized, and accessible across the organization. 

Companies need to evaluate their current capabilities in these areas and identify gaps that need to be addressed. Organizational maturity doesn’t happen overnight. It requires continuous effort and investment. However, the payoff—improved operational efficiency and customer satisfaction—makes it worth the endeavor. 

The role of ontology  

Your systems need to be able to speak to each other. Your knowledge management, product information management, and customer management tools must understand the same organizing principles. Ontology provides that. Ontology is the framework for organizing product information, user data, and content. Ontology involves defining and categorizing the various entities and their relationships within a domain. This categorization helps organize data in a meaningful way, making it easier to retrieve and use for different applications. Ontologies help eliminate data silos by ensuring all relevant data is accessible and interpretable by different systems. Without a well-structured ontology, data silos can cripple even the most advanced AI initiatives.  

Knowledge architecture 

Knowledge architecture is critical for leveraging generative AI. A good knowledge architecture organizes your data and content with metadata and ontologies, helping AI understand and respond accurately to user questions.  

 By structuring information into smaller, manageable pieces and tagging it with the appropriate metadata, businesses can improve personalization and decision-making. This alignment of systems and processes ensures better data quality, making information easier to find and use. A strong knowledge architecture makes AI solutions more relevant and valuable. 

How can you successfully implement AI at an enterprise level? 

At Circuitry.ai, we help manufacturers of complex products improve sales and service outcomes. Here are our three success factors for becoming an AI-powered enterprise: 

• Choosing the right use cases: Focus on areas where AI can deliver measurable improvements. For instance, in sales, AI can help recommend personalized products and configurations based on customer needs. In service, AI can assist in diagnosing and resolving technical issues. 

• Product data and knowledge strategies: By centralizing and leveraging comprehensive product data, organizations can train AI models to provide more accurate and relevant responses. This includes using product graphs and metadata to understand the relationships between different data points and improve the reasoning capabilities of AI systems. 

• Leveraging Enterprise AI as a Service model: This approach reduces the cost and risk associated with AI implementation while accelerating the time to value. Organizations can support various stakeholders across the product lifecycle, from marketing and sales to field service and customer support. 

AI and knowledge management assessment 

Not sure where your company stands? Earley offers a two-week assessment to see if your company is ready for AI in knowledge management. The assesment helps educate your team, define goals, check the quality of your knowledge base, and set up metrics and governance.  

Click here to get started on your assessment. 

Audience Q&A

 How do you power the AI enterprise to combine the data correctly, focused on the company’s terms, logic, and rules of operation? How do you correlate and define the data points across the data lake? 

SE: it’s always based on use cases. What are the objectives of the organization, of the departments and leaders, and what is strategically important? It’s not a matter of making the correlation and understanding the data across the data lake or data warehouse; it’s a matter of applying that data to solve a specific problem. This is where we get into identifying use cases that have measurable baselines. You need to get into the measurement mindset. Why are we doing this? What are we going to get?  

I’ve had people say, “We just need to build a knowledge graph,” and I said, “So what?” What is that going to do for you? Unless you’re solving a business problem, it becomes a distraction. Yes, you do need to get your data house in order. There is some rationale for building ontologies and knowledge graphs. But, when you start those projects, you really need to think about the end goal to show people what you spent money and company resources on. In some situations, you can say, “I don’t have a good business case or a process I can measure, but I just want to experiment, build some capabilities, and try this out.” And that’s fine as long as you know you’re doing that. But it is best to have that use case, baseline, and instrumentation in mind so you can say, “Look at what we achieved; this is really valuable.”  

AK: Use cases play a critical role. In our case, we see that because right now, AI is good at performing specific tasks, it’s not so good at performing every generic task. So, if we take a vertical industry, the domain, processes, and use case, we can focus on that. But that doesn't mean you need to have silos, either, because that's what we were talking about earlier, there are certain pools of data. Today, we talked about the data related to the products and how they can support some related use cases. One way to succeed is to select the right use case, focus on a specific domain and process areas, and also look at where the collection of data that supports these various use cases comes from.  

How important is organized taxonomy and structured content for successful AI implementation 

SE: You want to take advantage of the standards out there for efficiency, but it's important to organize based on the objectives of your customers and objectives of your business areas and functions. Structured data is important. We did a study recently where we built componentized content for a pharmaceutical company for their portfolio review. This included all the drugs they were investigating, their investments, licensing compounds, purchasing companies, and clinical trial results across their entire portfolio. This information was highly regulated, sensitive, and marketing moving, so we couldn't publish it to an LLM in the cloud; we had to use a localized model. So, we componentized these large 40-page narrative documents. LLMs have to break up content, so it’s better to break it up in a semantically meaningful way, meaning chunks that answer specific questions.  

We did it in two ways. First, we did it by ingesting the componentized content. Then, we tested the accuracy of its answers. We told the LLM that if it didn’t have the answer from the data source, say “I don’t know,” which virtually eliminated hallucination.

Then, we had a standard set of 60 gold use cases where we asked questions that a portfolio manager might ask the system. And we knew the answers to the questions. We found that by just ingesting the content, without knowledge architecture, without structuring the content, and without metadata, we could answer 53% of the questions. 

 When we added the metadata, added structure to the content, and added the knowledge architecture, we answered 83% of the questions. Adding these items is critical and makes a huge difference. It gives the LLM additional signals in the form of embeddings. You’re embedding that content into the vector space and providing additional signals in the form of metadata. That can significantly improve the ability of LLMs.

AK: I completely agree with Seth, especially for the manufacturing industry. Recently, a study was published in CIO magazine that said manufacturers are slowing down Gen AI initiatives due to concerns over accuracy. When you ask certain questions, the same question can retrieve an answer for different products. Let’s say you’re asking for specifications for a particular engine; it needs to answer that because it could be different depending on the type of engine or what supplier it comes from. 

 One way to increase the accuracy and relevancy of the answers is to have the right context and to provide it; you need metadata and structured content. You can get started in some aspects of it, but you need to incrementally populate the metadata and structure to improve the results. We also believe the relationships between entities are important. 

Ready for the next steps?

Sign up for our live demo session and see the power of Service AIdvisor in action. See how Service AIdvisor can improve productivity by up to 35%, increase profits, and elevate the customer experience. 

We're a proud sponsor of Field Service East 2024. Visit us at Booth 103 to learn how Service AIdvisor, powered by our Decision Intelligence platform, can improve the quality, velocity, and consistency of your service decisions.