Last updated: September 2026. Written by Josh Hutcheson, OnlineCourseing editor. Definitions and figures re-checked at the source (MIT Sloan, IBM) on 22 September 2026. We are not affiliated with any of the software vendors mentioned. See our review methodology.
By Josh Hutcheson · E-Learning Specialist
Reviewing online learning platforms since 2019. Review methodology
THE SHORT ANSWER
Bottom line: unstructured data is information without a predefined data model, such as emails, documents, images, audio, video and social media posts. It makes up most of the world’s data, and it has become far more usable because AI can now read, transcribe, classify and summarize it at scale.
- Share: 80% to 90% of data is unstructured, according to analyst estimates cited by MIT Sloan.
- The gap: only 18% of organizations in a 2019 Deloitte survey could take advantage of it (MIT Sloan).
- Where it lives: usually in its native format in data lakes or NoSQL databases (IBM).
- How it is analysed: text extraction, natural language processing, computer vision and, increasingly, large language models.
Learn NLP with DeepLearning.AI →
What is unstructured data?
Before you spend money on the wrong online course, read this.
Get the free 2026 Platform Comparison Guide — 12 platforms compared on price, certificates, and refund policies. Instant PDF, plus my honest Tuesday picks.
No spam. Unsubscribe anytime.
Unstructured data is information that does not follow a predefined data model and cannot be neatly stored in the rows and columns of a relational database. A customer table with fixed fields is structured; the emails those customers send are unstructured. IBM describes unstructured data as often stored in its native format in nonrelational databases or data lakes, and notes that machine learning, advanced analytics and natural language processing are used to extract insights from it (IBM).
It is also the majority of data. MIT Sloan cites multiple analyst estimates that 80% to 90% of data is unstructured, including text, video, audio, web server logs and social media, yet only 18% of organizations in a 2019 Deloitte survey said they could take advantage of it (MIT Sloan). That gap is why unstructured data skills are in demand.
Examples of unstructured data
| Type | Everyday examples | What organizations do with it |
|---|---|---|
| Emails and chat messages | Customer requests, internal decisions | Routing, sentiment, compliance review |
| Documents and PDFs | Contracts, reports, policies | Search, clause extraction, summarization |
| Scanned forms and invoices | Paper records, receipts | OCR and data extraction into systems |
| Images | Product photos, medical scans, satellite imagery | Classification, defect detection, mapping |
| Audio | Call recordings, voicemails, meetings | Transcription, quality monitoring |
| Video | Security footage, training videos | Object detection, indexing |
| Social media and reviews | Posts, comments, ratings text | Sentiment and trend analysis |
| Web pages | Competitor sites, news articles | Monitoring, research, AI search indexes |
| Log text | Application and server logs | Troubleshooting, security investigation |
| Medical notes | Clinician free-text records | Coding, cohort discovery (with strict privacy controls) |
Text: emails, documents and chats
Text is the largest category in most organizations: emails, contracts, reports, support tickets, chat logs and meeting notes. Analysing it can route customer requests, flag compliance risks or surface themes in feedback. Customer service teams are heavy users, which is why text analytics features so often in customer service courses.
Images and video
Photos, scans, medical imaging, satellite imagery and video are unstructured because their content is pixels rather than labelled fields. Computer vision turns them into usable information: detecting defects on a production line, reading a scanned form or mapping land use from satellite images. See our guides to image processing courses and, for geospatial imagery, GIS courses.
Audio
Call recordings, voicemails, podcasts and meeting audio become analysable once transcribed. Contact centres use transcripts to monitor quality and spot recurring problems; researchers use them to study interviews at scale.
Social media, reviews and web content
Posts, comments, reviews and news articles show what customers and markets are saying. Sentiment analysis classifies this text as positive, negative or neutral and tracks it over time; our guide to sentiment analysis courses covers the techniques.
Logs and machine text
Application and server logs are often semi-structured: they have timestamps and codes but free-text messages. They are central to troubleshooting and security investigations, which is why log analysis appears in cybersecurity courses and IT support courses.
Unstructured data examples by industry
| Industry | Unstructured data | Typical use |
|---|---|---|
| Healthcare | Clinician notes, medical images, discharge letters | Coding, diagnosis support, research cohorts |
| Financial services | Contracts, call recordings, analyst reports, emails | Compliance monitoring, fraud signals, research |
| Retail and e-commerce | Product reviews, photos, chat transcripts | Sentiment, product tagging, service automation |
| Manufacturing | Inspection images, maintenance logs, manuals | Defect detection, predictive maintenance |
| Legal | Case files, contracts, discovery documents | Clause search, document review |
| Public sector | Correspondence, forms, meeting records | Case handling, records management |
| Media and marketing | Video, images, social posts | Content tagging, audience insight |
Why unstructured data matters more in the AI era
For decades, most analytics ran on structured data because text, images and audio were too expensive to process at scale. Modern AI has changed that. Language models can read and summarize documents, speech recognition can transcribe hours of calls cheaply, and vision models can label images without hand-written rules. That turns the 80% to 90% of data that was previously hard to use into a practical resource.
It also raises the stakes for data quality. AI assistants that answer questions from company documents are only as good as the documents they retrieve: outdated policies, duplicates and conflicting versions produce confident wrong answers. Organizations getting value from AI usually invest first in organizing, deduplicating and governing their unstructured content.
Structured vs unstructured vs semi-structured data
| Structured | Semi-structured | Unstructured | |
|---|---|---|---|
| Model | Fixed schema (tables, columns) | Tags or keys, flexible schema | No predefined model |
| Examples | Sales records, inventory tables | JSON, XML, email headers, many logs | Emails, PDFs, images, audio, video |
| Typical storage | Relational databases, data warehouses | NoSQL document stores, data lakes | Data lakes, object storage, NoSQL databases |
| How you query it | SQL | Query languages for documents, or parsing | Extraction first, then NLP, vision or search |
| Analysis effort | Low | Medium | High, though AI has lowered it sharply |
In practice most pipelines move data from right to left in this table: extracting fields from documents, transcribing audio or tagging images turns unstructured data into structured or semi-structured data that analysts can query with SQL and visualize in dashboards.
Sources of unstructured data
- Internal communications: email, chat and meeting platforms.
- Document stores: shared drives, content management systems and contract repositories.
- Customer channels: support tickets, call recordings, chatbot conversations, surveys with open-text answers.
- Public web and social media: reviews, forums, news and social posts.
- Operational systems: logs, sensor notes, maintenance reports and field photos.
- Specialist sources: medical records, legal filings, engineering drawings and satellite imagery.
Much of this sits on file shares and servers managed by IT teams, so access control matters as much as analysis. Skills in directory services and virtualization, covered in our guides to Active Directory courses and VMware courses, are part of managing it safely.
Use cases for unstructured data
1. Search and question answering over documents
Search engines and AI assistants let employees ask questions of large document collections. IBM lists retrieval augmented generation, where a language model answers using retrieved passages from an organization’s own documents, among the main uses of unstructured data (IBM).
2. Intelligent document processing
Invoices, claims, applications and contracts are read automatically, key fields are extracted and the results feed business systems. This is often combined with robotic process automation, covered in our guide to RPA courses.
3. Customer insight and sentiment
Reviews, surveys, tickets and calls are analysed for recurring problems, product requests and sentiment trends, informing product and service decisions and marketing automation.
4. Image and video recognition
Quality inspection in manufacturing, medical image analysis, retail shelf monitoring and security video all rely on computer vision models trained on labelled images.
5. Risk, compliance and security
Organizations scan communications and documents for regulated data, policy breaches and fraud signals, and search logs during security investigations.
6. People and operations analytics
Open-text survey responses, exit interviews and maintenance notes reveal issues that structured metrics miss, a growing theme in HR analytics and supply chain work.
How to analyze unstructured data: step by step
- Define the question. Decide what decision the analysis should support, such as reducing complaint volume or speeding up invoice processing.
- Inventory the sources. Identify where the relevant emails, documents, recordings or images live, who owns them and what permissions apply.
- Extract the content. Convert scans with optical character recognition, transcribe audio and pull text from files.
- Clean and standardize. Remove duplicates and boilerplate, fix encoding problems and redact personal data where needed.
- Represent it for analysis. Tag entities and topics, or convert text and images into numerical embeddings.
- Analyse. Apply classification, clustering, sentiment analysis, search or a language model, depending on the question.
- Store results in structured form so they can be joined with other data, reported and monitored.
- Validate with people. Check samples of the output, because extraction and AI models make mistakes.
Techniques for analysing unstructured data
| Technique | What it does | Typical use |
|---|---|---|
| Optical character recognition (OCR) | Turns images of text into machine-readable text | Scanned forms, invoices, archives |
| Natural language processing (NLP) | Tags entities, topics, intent and sentiment in text | Emails, tickets, reviews |
| Speech-to-text | Transcribes audio | Call centres, meetings |
| Computer vision | Classifies and detects objects in images and video | Inspection, medical imaging |
| Embeddings and vector search | Represents meaning numerically so similar items can be found | Semantic search, recommendations |
| Large language models | Summarize, extract and answer questions over text | Document assistants, RAG |
| Clustering and topic modelling | Groups similar items without labels | Discovering themes in feedback |
These techniques draw on natural language processing, deep learning and machine learning skills.
Tools for working with unstructured data
Tools fall into a few categories. The examples below are well-known options, not recommendations for any specific project, and we have no commercial relationship with these vendors:
- Storage: data lakes and object storage on the major clouds, and NoSQL databases for documents.
- Document extraction: cloud services such as Amazon Textract, Google Cloud Document AI and Azure Document Intelligence, which Microsoft now presents as part of its Azure Content Understanding tools.
- Text processing libraries: open-source options such as spaCy for Python-based natural language processing.
- Search: engines such as Elasticsearch for indexing and searching large volumes of text and logs, increasingly combined with vector search.
- Language models: general-purpose AI models used for summarization, extraction and question answering over documents.
For the wider toolkit, see our guide to data science tools and our ranking of data analysis courses.
Challenges and risks
- Privacy and compliance. Emails, documents and recordings often contain personal data, so collection, access and retention must follow data protection law and internal policy.
- Volume and cost. Storing and processing large archives of files, audio and video is expensive; decide what is worth keeping.
- Quality. Poor scans, background noise and inconsistent formats reduce extraction accuracy.
- Accuracy of AI output. Models can misread documents or produce confident but wrong summaries; keep human review for decisions that matter.
- Skills. Teams need a mix of data engineering, NLP, computer vision and domain knowledge.
- Governance. Without clear ownership and access controls, unstructured data sprawls across drives and tools.
Careers and skills
Unstructured data work sits at the intersection of data engineering, data science and AI. Useful skills include Python, SQL, natural language processing, computer vision, working with language models and cloud data platforms. For the programming side, see our guide to data science programming languages; for the business side, business analytics courses and digital transformation courses.
Courses to learn unstructured data analysis
- DeepLearning.AI – Natural Language Processing Specialization (Coursera). A four-course program on the core techniques for working with text, from classification and vector spaces to sequence and attention models.
- IBM – Introduction to Computer Vision and Image Processing (Coursera). A practical introduction to working with images, from image processing basics to classification.
- Udacity – Natural Language Processing Nanodegree. A project-based program on building NLP applications, for learners who already know Python.
- IBM – Data Science Professional Certificate (Coursera). A broader foundation in Python, SQL and machine learning for beginners.
Coursera removed its free audit option for most courses in 2025, so check the price or trial terms before enrolling. For more options, see our guides to coding courses and small business courses if you want to apply these ideas in a smaller organization.
Frequently asked questions
What is unstructured data?
Unstructured data is information that does not fit a predefined data model or the rows and columns of a relational database, such as emails, documents, images, audio, video and social media posts. It is usually stored in its native format in data lakes or NoSQL databases and analysed with techniques such as natural language processing and machine learning.
What are examples of unstructured data?
Common examples are emails and chat messages, word-processing documents and PDFs, scanned forms, photos and medical images, audio recordings and call-centre transcripts, video, social media posts and reviews, web pages, and server log text.
What is the difference between structured and unstructured data?
Structured data follows a fixed schema, such as a customer table with defined columns, and is easy to query with SQL. Unstructured data has no predefined model, comes in many formats and needs extra processing, such as text extraction or image recognition, before it can be analysed.
How much of the world’s data is unstructured?
Most of it. MIT Sloan cites multiple analyst estimates that 80% to 90% of data is unstructured, including text, video, audio, web server logs and social media.
What is semi-structured data?
Semi-structured data has some organizing tags or markers but no rigid table schema. Examples are JSON and XML files, email headers and many log formats. It sits between structured and unstructured data and is often the first step when turning unstructured information into something queryable.
What are the main sources of unstructured data?
Inside organizations: email, chat and meeting platforms, shared drives and document systems, support tickets and call recordings, and application logs. Outside: social media, product reviews, news, forums and other web content. Specialist sources include medical records, legal filings and satellite imagery.
Is a PDF structured or unstructured data?
Usually unstructured. A PDF stores text and images laid out for reading, not in fields a database can query, so the content has to be extracted first, with optical character recognition if the PDF is a scan. Forms with fixed fields sit closer to semi-structured once their values are extracted.
How is unstructured data analysed?
Typically by extracting content (for example with optical character recognition for scanned documents), converting it into features or embeddings, and then applying natural language processing, computer vision or machine learning. Large language models and retrieval augmented generation are now widely used to search and summarize large document collections.
The verdict
Unstructured data is the bulk of what organizations hold: emails, documents, images, recordings and more. For years it was hard to use, but extraction tools, natural language processing, computer vision and language models have made it far more accessible. The organizations that benefit start from a clear question, extract and clean the content carefully, keep personal data protected and check AI output before acting on it.