What is structured and unstructured data in AI?

Understand the difference and why it matters with this simple guide.

Glenn Jaume

Product Manager at Coda

What is structured and unstructured data in AI?

By Glenn Jaume

Share this article

Share on TwitterShare on LinkedInShare on Facebook
AI · 4 min read
As our use of AI expands beyond simple tasks like asking questions and generating funny images, you might be hearing more about “structured” and “unstructured” data. The importance of structured data in AI has particularly come into focus with the rise of the use of enterprise AI tools for day-to-day work tasks. But, as with many things in this relatively new space, it can be hard to find a clear explanation of what this actually means, and why it matters. So here’s a simple guide to help.

What is structured and unstructured data?

 In the context of AI models, structured and unstructured data can refer to both the information it can access and the format of the responses it delivers (i.e., a text response, an image, a table, etc.) At its most simple:
  • Structured data is information that is highly organized and formatted in a way that is easily stored and searchable in relational databases (e.g., in rows and columns). Think of a spreadsheet of financial records or a table of customer information from your CRM.
  • Unstructured data does not have a pre-defined format and cannot easily fit into a traditional database of columns and rows. Think written content like docs, messages, and emails, or media like videos and images.
Both structured and unstructured data have uses and benefits; structured data tends to be easier and quicker for an AI to search and process accurately, while unstructured data provides a richness and versatility to what an AI can do.

Is unstructured or structured data better in AI?

 The simple answer is that both structured and unstructured data have their place in AI, and are suited to different use cases:
  • Structured data is typically best for AI tasks like classification and bulk categorization, data analysis, or data retrieval (e.g., showing me all my current sales opportunities over $10k).
  • Unstructured data is more suited to AI tasks like natural language processing (e.g., sentiment analysis), synthesizing or finding answers from large volumes of text, describing visual media, or transcribing voice to text.
Most people using consumer AI like ChatGPT will be more familiar with unstructured responses, usually text or images. This works well when you’re using AI almost like a search engine to get answers or to generate content. But when it comes to using AI at work, unstructured data alone isn’t enough. Think about your own business for a moment. Where is information and data stored? I’m guessing it’s across a mixture of unstructured sources, like docs and Slack messages, and many structured sources like CRMs, spreadsheets, databases, and apps like Jira or Asana. To be most valuable and accurate, enterprise AI needs to be able to access and analyze all of those data sources. And it needs to be able to respond in the format that is most appropriate to the question. For example, if you’re asking for help writing a blog post, a text response is probably just fine. But if you’re asking for last month’s closed-won data for your exec update, a table of data is going to be much more useful—not only because it is more verifiable than a summarized list, but also because it can be directly inserted into your document or dashboard.

3 examples of structured and unstructured data in AI.

 When using enterprise AI, you will likely need both structured and unstructured data for different use cases—sometimes it will be one or the other, and sometimes both for a single task. Let’s take a look at what that could look like in reality.

Writing a product proposal.

 You’re a product manager creating a write-up for a proposed roadmap feature. You want to gather relevant internal context and data that exists about this feature to help make your case.
  • Structured data: You want to include evidence for why you should prioritize building this feature now, so you ask AI what recent deals you’ve lost due to this feature not existing. The AI returns a table of lost deals from Salesforce, filtered to include those mentioning this feature. You insert this directly into your write-up.
  • Unstructured data: You ask AI if this feature has been explored before. In response, it looks through your internal docs, Slack channels, and emails, and returns a summary of previous explorations and the reasons they didn’t continue. You add this context to your write-up to provide some background.

Sending a project update.

 You’re a project lead and want to send an update on progress in the last week.
  • Structured data: You also want to include an overview of progress, so ask AI for project tasks that are completed, blocked, and on track. The AI returns a filtered table of tasks from Asana, so you can insert it into the update without having to link out or screenshot your tasks.
  • Unstructured data: You ask AI to summarize that week’s stakeholder meeting. After analyzing your meeting notes, it provides a brief summary of the main decisions and action items, which you insert into your update.

Planning your next sprint.

 You’re an engineering leader working on a new feature and are planning what needs to be prioritized in the next sprint.
  • Structured data: You also want to see progress on this project so far, so ask AI for an update. You receive back a view of all the related issues from Jira, including how complete they are, so you get an overview without switching between tools.
  • Unstructured data: You want to check the agreed scope for the feature, so you ask AI to show you the product requirements. It retrieves them from your internal documents and summarizes them for you.

The best AIs do both.

 These examples show that both structured and unstructured data have their uses, and the power of AI is multiplied when it can leverage both in its training and responses. We built Coda Brain, our turnkey enterprise AI platform, to understand and respond to both structured and unstructured data for maximum value. We use approaches like RAG, citations, human-in-the-loop, and more to ensure that it provides accurate, verifiable answers you can trust (because hallucinations simply aren’t welcome at work!). If you’d like to learn more about Coda Brain, join the private preview.
Join the private preview

Related posts

Explore more stories about AI.
AI Hallucinations

How we’ve built accuracy and confidence into Coda Brain.

Building your own AI bot?

Building an AI bot fit for work isn’t as simple as it might seem.

What is RAG?

How RAG works and why it’s key to enterprise AI.