As data complexity grows and business leaders demand instant, actionable insights, knowledge graphs—especially when combined with generative AI (genAI) —present new opportunities to elevate the value and functionality of cloud data platforms. CTOs and CIOs are increasingly tasked to apply genAI to build AI agents, reduce the time personnel spend identifying and applying useful resources, query all available data sources with natural language interfaces, and more. Before selecting a vendor, it’s critical to ask the right questions. Here are ten you should consider before launching your project.

1. What can knowledge graphs do that my existing cloud data platform cannot do?

Knowledge graph software can connect and contextualize structured and unstructured data from across the enterprise in ways that are beyond the practical capabilities of cloud data platforms by themselves.

Cloud data platforms can bring data together with high degrees of governances and cleanliness, but it will not necessarily be connected across data domains or original source systems. Certain types of enterprise data may not be represented at all in a cloud data platform.

Cloud data platforms are good at aggregating, storing, and cleaning data from multiple sources, but generally do not integrate those sources in ways that makes them easy to use together except to support specifically developed use cases. As new use cases arise, project managers must develop, test, and deploy changes to the platform, which can require major new investments of valuable personnel time.

Unstructured data includes things like email and chat systems, PDF documents, images, text files, and pretty much everything else. Dealing with unstructured data requires advanced software that does more than data cleansing. Simply connecting to unstructured data sources can be challenging; figuring out how to make sense of it is even more difficult without the right tools.

Ask your vendors if their knowledge graph solution requires you to copy and store data in a separate repository or if it utilizes an in-memory/high performance cache.

2. How scalable and performant is the solution?

There are two major aspects of scale worth examining closely: scale in terms of managing the knowledge graph itself and scale in terms of loading/transforming/querying all the data available to the graph.

Enterprise data volumes are massive and growing. Can the software handle billions of nodes and edges? How fast does it load data? Does it utilize parallel computing techniques to improve performance? Does it maintain low latency for complex queries, even as your organization grows even larger?

Make sure the systems you evaluate are designed for enterprise-grade scalability and can support high-performance analytics and complex queries at speed.

3. How does the platform automate knowledge graph construction?

Users should not have to build knowledge graphs using (mostly) manual processes. The tools should take advantage of existing metadata, data models, and governance information to automate building new knowledge graphs. For example, does it utilize the schema from a relational database automatically or does it require manual work to bring the database into the system?

Ask vendors to explain and demonstrate the steps required to build a knowledge graph and ask them to show you how their automation tools work using your own data.

4. How easy are knowledge graphs produced by the software to use?

If your shiny, new (and expensive) knowledge graph-based system is only usable by people with significant data science experience, you will see very limited acceptance. Make sure that executives, product managers, R&D leads, and others can understand how the system works, make queries, and receive meaningful answers without IT support.

Ask vendors to show the process of building and accessing a knowledge graph. Is it easy to understand? Can your technical team learn to manage the software quickly? Will they understand how it works? Can they build and manage a knowledge graph produced by the software with the same ease they manage tables, datasets, and data products in your existing cloud data platform? Can business users access knowledge graphs built with the software using their familiar tools?

5. How does the system handle security, privacy, and compliance?

Knowledge graphs raise security concerns by bringing data together from multiple sources. In every enterprise application and industry, sensitive data must be subject to robust access controls, audit trails, and comply with regulations like GDPR, HIPAA, and more.

Ask vendors what they do to mitigate these concerns and carry over existing security protocols. Request specifics on how the software supports internal governance and compliance requirements. Make sure the system provides the tools/ features needed in your application to ensure data security and regulatory compliance.

6. How does the platform use ontologies?

Ontologies are the semantic building blocks that represent knowledge within domains and facilitate clear, concise data interoperability. Ontology-driven integration describes data in business terms and supports transparency, trust, and actionable insights.

Ask vendors to explain how their systems abstract and describe integrated data using ontologies to make that data accessible to generative AI solutions. Be sure the platform supports relevant industry standards and custom extensions. Will you be able to use ontologies to address current business requirements and new requirements as they emerge?

7. Do knowledge graphs created with the software support natural language queries and responses, and how accurate (really) are the responses?

Senior executives need answers, not dashboards. The ability to translate complex business questions like “What is our gross profit margin for all imported products sold in the home country and how has this changed over the past three years?” into complete, accurate, understandable results is critical.

Retrieval-augmented generation (RAG) is a valuable technique that limits a genAI model’s frame of reference to real and vetted information. Graph RAG (sometimes called GRAG) improves on this by exploiting the contextual information from knowledge graphs to reduce hallucinations. It adds context to prompts that increase the accuracy of the generated response and reduces the likelihood of hallucinations. Graph RAG delivers contextually rich, precise answers that go beyond the capabilities of normal RAG systems and are more efficient and effective at building connections within and between all structured sources and unstructured documents/chunks. Graph RAG grounds all response in your enterprise data to improve the quality of responses make them clearer and actionable for non-technical users.

Ask vendors to explain exactly how their knowledge graph platform supports Graph RAG and request that they demonstrate that capability using your data in a proof of concept.

A successful knowledge graph implementation will enable users to take advantage of all the capabilities offered by genAI through a combination of built-in capabilities and integrations. A properly implemented knowledge graph enables genAI systems to explain its answers and cite sources, which is essential for building trust and transparency in AI-driven decision-making. LinkedIn reported a 78% accuracy improvement in their customer service AI by incorporating enterprise knowledge graphs with RAG systems, simultaneously reducing issue resolution times by 29%.

8. How transparent and explainable are AI-generated answers?

Everyone using the system must be able to trust its output. Does the solution provide clear explanations and reasoning paths for every answer? If you can’t explain and see how the system arrived at an answer, you cannot trust that information.

Ask vendors to show you how their systems ground AI responses in the knowledge graph and its ontologies. Are responses traceable, explainable, and auditable?

9. What is the time to value?

How quickly can your teams start realizing value from your investment in the software and associated implementation costs? Does the system require extensive customization? Ask for case studies and benchmarks, especially for large-scale, multi-source deployments. Can you reasonably expect to begin using the system within a few months? Or will it realistically a year or more?

10. What do the vendor’s support services look like?

How good is the vendor’s implementation team? Are they experienced in building knowledge graphs? Does the vendor provide responsive support, comprehensive documentation, and a community of practice?

When you find a vendor who can give you satisfactory answers to all these questions, add them to your short list and work with them to demonstrate performance and capability using your own data. There’s no substitute for a real-world trial. Contact us to learn more.