Most data driven companies have focused on the structured kind—the rows and columns that live in spreadsheets and SQL databases.
But behind the scenes, an entirely different type of data is proliferating: unstructured data. It's messy, sprawling, and difficult to analyze—but also rich in insight.
Let's explore the different types of unstructured data and why it is hard to make sense of them.
Text
Let's start with text.
Not the neat kind from a form, but the kind that flows from people: emails, chat logs, support tickets, sales call transcripts, and internal wikis. This is where a company's true voice and its customers' lives. Yet it often sits idle in silos or underused systems.
Think of a support team fielding hundreds of tickets per week. Inside those messages are themes: bugs, broken flows, pricing confusion, unmet expectations. How do we surface these themes automatically, flag emerging problems, and even propose fixes?
The value is clear, but extracting insights from text requires you to parse highly variable syntax, spelling, and tone. Not to mention solving this across many languages. Furthermore, long-form context often split across multiple documents or threads, and you also need to think about security issues like masking sensitive or private information embedded in open text.
Then you can unlock powerful use cases on unstructured text. You can detect common pain points, automate ticket routing, summarize long conversations from customer support data. You can build internal chatbots that can answer questions from messy documentation or policy manuals. You can extract competitor mentions, objections, and deal signals from call transcripts.
Voice/audio
Audio is the close cousin of text.
Many text data files are originally transcribed from audio. During this process, information is lost: tone, emotion, urgency—layers of meaning that are often lost if raw audio files isn't used but only their transcripts. Zoom recordings, customer service calls, voice memos, podcasts all fall into this category.
With modern speech recognition and emotion detection tools, we can turn these calls into structured narratives. But speaker diarization (who said what) can be unreliable. Languages become an even bigger problem than parsing text. Background noise, accent variation, and emotion add complexity.
Unlocking this for customer support recordings for example means faster escalations, better training for reps, and a much sharper understanding of how customers feel, not just what they say.
Images and video
Images and video form another massive—and growing—domain of unstructured data. UX teams collect screenshots from users reporting bugs. Insurance adjusters receive photos of damaged cars. Marketers produce banner ads, social posts, and video explainers by the dozen.
A big issue for this type of data is the size, leading to high storage and compute costs for processing. Storage has gotten better in 2010s due to the availability of cheap cloud storage (think S3), but processing power has not kept up.
If you think about it, speech, text, and visual cues must be analyzed together (this is called multi-modal complexity), which means that many machine learning models and technologies requires breaking content into scenes or frames just to be able to process it.
This is why, until recently, making sense of all this visual data required human eyes.
Today, large vision models (the visual equivalent of LLMs) can scan thousands of images or hours of video to find patterns, anomalies, or specific objects.
Optical Character Recognition (OCR) pulls data out of screenshots and scanned documents. Scene detection algorithms break long videos into chapters.
For example, a design team could feed user-submitted screenshots through a model that flags when an error message appears—automating triage. A marketing team might run A/B tests not just on copy, but on the visual composition of ad creatives.
Social and web
Some of the richest unstructured data comes not from internal systems, but from the public web. Think product reviews on G2 or Amazon. Tweets and Reddit threads about your competitors. Blog posts or forum questions about your API.
This content is messy—full of slang, emojis, and sarcasm—but also brutally honest. AI can now make sense of it, clustering sentiment, surfacing trends, and even helping generate responses. Smart companies are combining these signals with their internal data to close the loop between perception and performance.
For example, a product team might notice a sudden spike in Reddit posts mentioning “feature X not working.” Cross-referenced with support tickets, they now have the data to escalate an issue—before it becomes a crisis.
Machine Data
Finally, there's the world of machine-generated unstructured data: server logs, crash reports, telemetry from IoT devices. This isn't written in human language—but it's still “unstructured” in the sense that it doesn't sit in traditional tables or follow a neat schema.
Take log files: they can be massive, repetitive, and full of false positives. But they also contain the early warnings of system failures. Today, AI models trained on embeddings can detect unusual log patterns in real-time—like a cybersecurity analyst on hyperspeed.
Similarly, usage telemetry from apps can be analyzed to understand user journeys, predict drop-off points, or uncover performance bottlenecks. All of it feeds into product, ops, and engineering teams who are trying to make sense of digital behavior at scale.
Why now?
So why is unstructured data so hot right now?
Because generative AI and foundation models are finally giving us the tools to make sense of it—at scale, and with context. Text can be summarized. Audio can be tagged. Images can be searched visually. Logs can be clustered. And all of it can be turned into insight, faster than ever before.
For teams thinking about their data strategy, this is the moment to expand the aperture. It's no longer enough to just analyze rows and columns. The future is in voice, video, messy text, and noisy logs. It's in the unstructured world.
And for those who learn to structure the unstructured—there's massive upside waiting.
