# Cognitive skill

> Source: Courseiva IT Certification Glossary — https://courseiva.com/glossary/cognitive-skill

## Quick definition

Cognitive skills are the mental abilities that artificial intelligence uses to learn from data, understand context, and solve problems. They include things like recognizing faces in photos, understanding spoken commands, and reading text. These skills allow AI to behave more like a human brain, making it useful for real-world tasks.

## Simple meaning

Think of cognitive skills as the toolbox your brain uses to make sense of the world. When you see a red light at an intersection, your brain instantly knows to stop. That is a cognitive skill working hand-in-hand with your senses. Now, imagine you are teaching a computer to do the same thing. The computer doesn't have eyes or a brain, so you give it sensors like cameras and microphones, and you load it with special programs, cognitive skills, that let it recognize the red light and decide to stop. In the world of artificial intelligence, cognitive skills are the programs that let computers see, hear, read, speak, and make choices. They break down complex human abilities into tiny, logical steps that a machine can follow. For example, a cognitive skill for reading would involve the computer first looking at a picture of a word, then breaking that picture into individual letters, matching each letter to its known shape, and finally putting those letters together to form a word. Just like you learned to read by practicing and memorizing, a computer learns its cognitive skills by being fed thousands and thousands of examples until it becomes really good at the task. These skills are the building blocks of smart applications like voice assistants, face-unlock features on your phone, and spam filters in your email. Without cognitive skills, AI would just be a calculator that can only follow strict, simple instructions. With them, it can handle messy, unpredictable real-world information.

## Technical definition

In the context of Microsoft Azure and the AI-900 exam, a cognitive skill refers to a predefined or custom operation within an AI pipeline that extracts information from unstructured data. These skills are commonly used in Azure Cognitive Search enrichment pipelines to transform raw data into searchable, structured information. Azure provides a set of built-in cognitive skills that leverage pre-trained machine learning models from Azure Cognitive Services. These skills operate on documents, images, and text to perform tasks such as optical character recognition (OCR), entity recognition, key phrase extraction, sentiment analysis, language detection, and image analysis. Each skill takes an input, processes it through a machine learning model hosted in the cloud, and produces an output that can be stored in a knowledge store or used for indexing. For example, the OCR skill scans an image or PDF and extracts printed or handwritten text, making it available for full-text search. The entity recognition skill identifies people, places, organizations, and other named entities in text, allowing for faceted navigation in search results. The key phrase extraction skill pulls out the most important points from a document, helping users quickly understand content. The sentiment analysis skill assigns a positive, negative, or neutral score to text, useful for customer feedback analysis. In addition to built-in skills, developers can create custom skills using Azure Functions or other code to handle specialized requirements not covered by the predefined set. These custom skills can call external APIs or perform proprietary logic. The entire set of skills is orchestrated in a skillset, which is attached to an indexer pipeline in Azure Cognitive Search. The indexer runs the skillset sequentially, feeding outputs of one skill as inputs to the next. The output of the enrichment process is a knowledge store, which can be projected into tables or blob storage for further analysis. The AI-900 exam expects candidates to understand the difference between built-in and custom skills, the flow of data through a skillset, and how Azure Cognitive Search uses these skills to enable powerful search experiences over unstructured data. Key considerations include data limits, cost of calling Cognitive Services, and the importance of data privacy when processing sensitive information.

## Real-life example

Imagine you are a librarian at a huge university. Every year, hundreds of students submit handwritten thesis papers. Your job is to read each paper, note the author's name, the main topic, the key findings, and whether the conclusion is positive or negative about their research. You then create a card for each thesis with all this information so that other students can search for it later. This is exactly what a cognitive skill does in the digital world. Now, instead of a person, the system uses a set of cognitive skills. The first skill is like your eyes, it reads the handwriting using OCR. The second skill is like your brain recognizing names and places, it uses entity recognition to pull out the author's name and university. The third skill is like you highlighting the most important sentences, it performs key phrase extraction to get the main research findings. The fourth skill is like you deciding if the student was happy or unhappy with the results, it runs sentiment analysis. Finally, all this extracted information is stored in a library card catalog, which in the computer world is the search index. Now, when someone searches for 'climate change thesis from 2023,' the system finds the right card instantly. In your daily life, every time you use Google Photos to search for 'dog' and it correctly shows you all your dog photos, that is a cognitive skill working. When your email automatically sorts a message into your spam folder, that is another cognitive skill. These digital skills save us enormous time by automating the thinking part of processing information.

## Why it matters

Cognitive skills matter in IT because they unlock the value hidden in unstructured data, the kind of data that makes up over 80% of all digital information. Emails, PDFs, images, social media posts, videos, and audio recordings do not fit neatly into database tables. Without cognitive skills, businesses would have to hire armies of people to manually read, tag, and categorize this information. That is slow, expensive, and error-prone. By implementing cognitive skills, organizations can automatically extract insights from mountains of data at a speed and scale impossible for humans. For example, a healthcare company can use cognitive skills to scan thousands of medical reports and instantly identify patients with specific conditions. A legal firm can use them to review contracts and highlight clauses that might be risky. An e-commerce site can use image analysis skills to automatically tag products with attributes like color, style, and brand. In addition to efficiency, cognitive skills enable new user experiences. Search engines powered by these skills can understand the intent behind a query, not just match keywords. For instance, a search for 'affordable Italian restaurants near me with outdoor seating' can return relevant results because the system understands the meaning of each phrase. For IT professionals, knowing how to configure and deploy cognitive skills means they can build smarter applications that interact with users in natural ways. It also means they can reduce manual work, improve accuracy, and deliver faster insights. As AI becomes more embedded in business processes, the demand for professionals who can implement cognitive skill pipelines will only grow. Finally, cognitive skills are a core component of the Azure AI stack, and a strong understanding of them is essential for passing the AI-900 exam and for advancing to more advanced Azure AI certifications.

## Why it matters in exams

For the AI-900 exam, cognitive skills are a core objective under the Knowledge Mining module. Microsoft explicitly expects candidates to understand the concept of cognitive skills, the types of built-in skills available, and how they fit into the overall Azure Cognitive Search enrichment pipeline. The exam tests your ability to distinguish between cognitive skills and other AI workloads like Computer Vision or Natural Language Processing, which are separate services but provide the models that power the skills. You need to know that cognitive skills are used within a skillset, which is part of an indexer, which is part of Azure Cognitive Search. The exam will present scenarios where you have to decide which built-in skill to use for a given data extraction task. For example, if the question says 'you need to extract printed text from a scanned PDF,' the correct answer is the OCR skill. If the question says 'you need to identify the names of people mentioned in customer reviews,' the correct answer is the entity recognition skill. You should also know the difference between built-in and custom skills. A common question type asks: 'When would you create a custom cognitive skill?' The answer is when your data processing needs are not covered by any of the built-in skills, or when you need to call an external API that does a specialized task. The exam also covers the concept of the knowledge store, the output destination where enriched data is stored. You need to know that a knowledge store can project data into tables or blob storage. Expect multiple-choice questions that ask which skill is appropriate for a specific business requirement. Also, be prepared for 'select all that apply' questions about the steps in a typical enrichment pipeline. The AI-900 exam often includes case studies where you must design a solution for knowledge mining. For instance, a company wants to make its legal documents searchable by case name, judge, and verdict. You must choose the correct combination of skills (entity recognition for case name and judge, key phrase extraction for verdict summary) and understand the data flow. Being solid on cognitive skills will help you answer at least 3-5 questions correctly on the exam, which is significant for passing.

## How it appears in exam questions

In the AI-900 exam, questions about cognitive skills appear in several distinct patterns. The first and most common pattern is the scenario-based question. For example, the question might describe a company that receives thousands of handwritten customer feedback forms every month. They want to automatically extract the customer's name, the date, the main complaint, and whether the feedback is positive or negative. The question will ask which combination of cognitive skills should be included in the skillset. The correct answer would be: OCR skill to read the handwriting, entity recognition to extract the customer name, key phrase extraction to identify the main complaint, and sentiment analysis to determine positive or negative tone. Another pattern is the definition question. The exam may ask directly: 'What is the purpose of a cognitive skill in Azure Cognitive Search?' The answer is to extract or enrich information from unstructured data during the indexing process. A third pattern is the comparison question. For instance, the exam might present two scenarios: one where you need to extract text from an image, and another where you need to detect objects in an image. The question will ask which skill applies to the first and which to the second. The answer is OCR for text extraction and image analysis for object detection. A fourth pattern is the pipeline question. The exam could show a diagram or describe steps: 'Data source, indexer, skillset, index, knowledge store.' You might be asked what component contains the cognitive skills, the skillset. Or what order the skills run, sequentially in the order defined in the skillset. A fifth pattern is the custom skill question. The exam might ask: 'When would you use a custom cognitive skill instead of a built-in one?' The correct response is when you need to perform proprietary calculations, when you need to call an external API, or when no built-in skill meets your requirement. A sixth pattern is the cost and performance question. For example, 'What happens when you exceed the free tier limits for Cognitive Services when running a skillset?' The answer is that the skill calls will fail, and enrichment will stop. You may also see questions about data residency or privacy, for instance, if your data cannot leave a certain region, you need to ensure your Cognitive Services instance is in the same region. Finally, there are 'select all that apply' questions where multiple skills are listed, and you must pick which ones are used for text enrichment (entity recognition, key phrase extraction, sentiment analysis, language detection) versus image enrichment (OCR, image analysis). Understanding these question patterns helps you prepare effectively.

## Example scenario

A large insurance company, SafeGuard Insurance, receives thousands of claim forms every day. Each claim form is a scanned PDF with handwritten notes from the claimant and a photo of the damaged vehicle. The company wants to automatically extract the claimant's name, policy number, date of loss, and the type of damage (e.g., 'front bumper dent'). They also want to know if the claimant's description sounds urgent or angry. You are tasked with building a solution using Azure Cognitive Search. First, you set up an Azure Cognitive Search service and create a data source pointing to the blob storage where the scanned PDFs are uploaded. Next, you create a skillset with several cognitive skills. The first skill is OCR, which reads the handwritten text from the scanned PDF. Because the handwriting might be messy, the OCR skill uses a pre-trained model to convert the image into machine-readable text. The output of this skill is a large block of text containing all the handwritten content. The second skill is an entity recognition skill configured to look for names and numbers. This skill scans the extracted text and identifies 'John Smith' as a person and 'POL-2024-7890' as a number, the policy number. The third skill is a key phrase extraction skill that pulls out important phrases like 'front bumper dent' and 'delayed response.' The fourth skill is a sentiment analysis skill that reads the tone of the claim description and assigns a score. A score close to 0 is negative, and a score close to 1 is positive. You then attach this skillset to an indexer. The indexer reads each PDF, passes it through the skillset, and stores the enriched results in an index and a knowledge store. The knowledge store allows analysts to query the data using SQL or Power BI. At the end, when a claims adjuster searches for 'John Smith bumper dent,' the system instantly returns the relevant form with all extracted fields. This saves the adjuster hours of manual data entry every day.

## Common mistakes

- **Mistake:** Thinking that cognitive skills are standalone AI services like Azure Computer Vision or Azure Language Service.
  - Why it is wrong: Cognitive skills are not standalone services. They are individual operations that run inside a skillset within Azure Cognitive Search. They use pre-trained models from Azure Cognitive Services, but they are components of a larger enrichment pipeline.
  - Fix: Remember that cognitive skills live inside a skillset, which is part of an indexer pipeline in Azure Cognitive Search. They are not separate services you call directly.
- **Mistake:** Believing that you can use cognitive skills on structured data like SQL tables without any transformation.
  - Why it is wrong: Cognitive skills are designed to extract information from unstructured data such as images, PDFs, and text documents. Structured data in tables does not need cognitive skills because it is already organized. Using them on structured data would be unnecessary and wasteful.
  - Fix: Only apply cognitive skills when your source data is unstructured. For structured data, use regular indexing without enrichment.
- **Mistake:** Assuming that all cognitive skills work on images and text equally.
  - Why it is wrong: Some skills are specifically for images, like OCR and image analysis. Others are for text, like entity recognition and sentiment analysis. You cannot use an image analysis skill on a plain text document, and you cannot use a key phrase extraction skill on a photo.
  - Fix: Match the skill to the data type. Use image skills for image files and text skills for text content. In a PDF with both, use OCR first to extract text, then run text skills on the extracted text.
- **Mistake:** Thinking that you must create custom skills for everything instead of using built-in skills.
  - Why it is wrong: Azure provides a comprehensive set of built-in cognitive skills that cover most common scenarios. Creating custom skills for tasks that already have a built-in skill adds complexity, cost, and maintenance burden.
  - Fix: Always check the list of built-in skills first. Only create a custom skill when your specific requirement is not met by any built-in skill.

## Exam trap

{"trap":"The exam might present a scenario where a company wants to extract text from a scanned document, and among the answer choices there is 'Computer Vision API' and 'OCR cognitive skill.' Learners might pick 'Computer Vision API' because they remember it can read text.","why_learners_choose_it":"Learners confuse the standalone Computer Vision API service with the OCR cognitive skill. They both use the same underlying model, but the question is about a cognitive skill in a search pipeline, not a direct API call.","how_to_avoid_it":"Read the question carefully. If the question mentions Azure Cognitive Search, an indexer, or a skillset, the correct answer is the OCR cognitive skill, not the standalone Computer Vision API. The Computer Vision API is a separate service, not a cognitive skill."}

## Commonly confused with

- **Cognitive skill vs Azure Cognitive Services:** Azure Cognitive Services is a suite of pre-built APIs for vision, speech, language, and decision-making. Cognitive skills are specific operations that use these services but are embedded within Azure Cognitive Search enrichment pipelines. Cognitive Services are standalone endpoints; cognitive skills are pipeline components. (Example: You call Azure Cognitive Services directly to analyze an image. But you add an OCR cognitive skill to an Azure Cognitive Search skillset to extract text from documents during indexing.)
- **Cognitive skill vs Data extraction:** Data extraction is a broad term for pulling data from any source, like a database or an API. Cognitive skills are a specific method of extracting and enriching data using AI models, usually from unstructured sources like images and text files. Not all data extraction uses cognitive skills. (Example: Extracting a customer name from a SQL table is data extraction without cognitive skills. Extracting a customer name from a scanned handwritten note is data extraction using a cognitive skill (OCR and entity recognition).)
- **Cognitive skill vs Machine learning model training:** Training a machine learning model involves feeding labeled data to a model so it learns patterns. Cognitive skills use pre-trained models that are already trained. You do not train them; you just configure them and use them in your pipeline. (Example: Training a custom model to recognize your company's logo is machine learning training. Using the built-in image analysis cognitive skill to detect that a picture contains a logo is using a cognitive skill.)

## Step-by-step breakdown

1. **Identify the data source** — First, determine where your unstructured data lives. It could be in Azure Blob Storage, Azure Data Lake Storage, or a file share. The data source tells Azure Cognitive Search where to find the files to be enriched.
2. **Create the indexer** — The indexer is the automation engine that connects the data source to the skillset and the index. It runs on a schedule or on demand, pulling new or changed documents and feeding them through the enrichment pipeline.
3. **Define the skillset** — A skillset is a collection of one or more cognitive skills arranged in a specific order. You define which built-in or custom skills to use, their inputs, and their outputs. The order matters because the output of one skill can be the input to the next.
4. **Add cognitive skills** — Add the specific skills you need. For example, an OCR skill to read images, followed by an entity recognition skill to find names and dates. Each skill has its own configuration, such as the language or the fields to extract.
5. **Configure output field mappings** — Map the enriched data from the skillset to fields in the search index. For instance, you may map the output of the entity recognition skill to a field called 'people_names' so that it can be searched or filtered later.
6. **Create the knowledge store projection** — Optionally, you can project the enriched data into a knowledge store, which can be tables in Azure Storage or Azure SQL Database. This allows you to perform further analysis with tools like Power BI without running the enrichment again.
7. **Run the indexer and verify** — Execute the indexer to process the documents. Check the logs for any errors, such as throttling from Cognitive Services or malformed documents. Once successful, the enriched data is available in the search index and knowledge store.

## Practical mini-lesson

In practice, implementing cognitive skills in Azure requires a solid understanding of both the service structure and data handling. Let's walk through a typical professional scenario. Suppose you work for a logistics company that processes thousands of shipping invoices every day. Each invoice is a PDF containing a mix of printed text, a barcode, and a signature. Your goal is to extract the invoice number, date, total amount, and the sender's name, and then make this data searchable. You start by uploading the PDFs to Azure Blob Storage, organized in date-based folders. In Azure Cognitive Search, you create a data source pointing to that container. Next, you build a skillset. You add an OCR skill first to extract all text from the PDF, including the barcode number which is usually printed in a specific font. The OCR skill converts the PDF into a JSON structure containing lines of text with bounding boxes. You then add an entity recognition skill configured to find dates and monetary values. This skill scans the OCR output and pulls out '10/15/2024' as a date and '$1,250.00' as a monetary amount. You also add a custom skill that you wrote in Python, hosted as an Azure Function, that parses the invoice number using a regular expression because the format is unique to your company. The custom skill takes the OCR output as input and returns 'INV-2024-12345'. You then map these outputs to fields in the index: invoice_number, invoice_date, total_amount, and sender_name. Finally, you create a knowledge store projection that stores the enriched data in an Azure SQL table for reporting. Professionals need to be aware of several gotchas. First, the OCR skill has a file size limit of 4 MB per image, and for PDFs it must be under 32 MB after base64 encoding. If your invoices are larger, you need to split them into pages. Second, each skill call to Cognitive Services incurs a cost, so you should only include skills you truly need. Third, if you use custom skills, they must return results within a few seconds, or the indexer will time out. Fourth, always test your skillset on a small set of documents before running it on millions. A single misconfigured skill can waste hours of processing time and money. Finally, data privacy is critical. If your invoices contain personal data, ensure your Cognitive Services instance is in the same geographic region as your data to comply with regulations.

## Memory tip

Cognitive skills enrich search data, like adding spices to a raw dish. They transform bland files into flavorful insights.

## FAQ

**Can I use cognitive skills on data that is already structured, like a CSV file?**

Cognitive skills are designed for unstructured data like images, PDFs, and plain text. Using them on structured data like CSV files is unnecessary because the data is already organized. You can index structured data directly without enrichment.

**How many cognitive skills can I include in a single skillset?**

There is no hard limit on the number of skills, but each skill adds processing time and cost. Azure Cognitive Search imposes a maximum of 30 skills per skillset. In practice, you should only include the skills you absolutely need.

**Do I need to train the models behind cognitive skills?**

No. The built-in cognitive skills use pre-trained models from Azure Cognitive Services. They are ready to use out of the box. You only need to train a model if you create a custom skill that relies on your own machine learning model.

**What happens if a cognitive skill fails during indexing?**

If a skill fails, the indexer can be configured to continue with the remaining skills or to fail the entire document. By default, the indexer will fail the document but continue processing others. You can check the indexer execution history for error details.

**Can I use cognitive skills with data stored on-premises?**

Yes, but you need to use an Azure Cognitive Search indexer that connects to an on-premises data source via a data gateway. The cognitive skills still run in Azure, but the data is securely transferred via the gateway.

**What is the difference between a cognitive skill and a custom skill?**

A built-in cognitive skill is pre-defined by Microsoft and uses standard models like OCR or entity recognition. A custom skill is code you write yourself, usually as an Azure Function, to perform a specialized task that the built-in skills do not cover.

## Summary

Cognitive skills are a fundamental concept in Azure Knowledge Mining, especially for the AI-900 exam. They are small, focused operations that extract, enrich, and transform unstructured data into structured, searchable information. By using pre-trained AI models from Azure Cognitive Services, these skills allow you to perform tasks like reading handwritten text, identifying names and places, extracting key phrases, and analyzing sentiment, all within the indexing pipeline of Azure Cognitive Search. Understanding how to configure skillsets, choose between built-in and custom skills, and map outputs to search indexes is essential for any IT professional working with AI. In the AI-900 exam, you will encounter scenario-based questions that test your ability to match the correct cognitive skill to a business problem, as well as questions about the enrichment pipeline and knowledge stores. The biggest key to success is to remember that cognitive skills are components of a skillset, which is part of an indexer, which is part of Azure Cognitive Search. They are not standalone AI services. By mastering this relationship and the specific use cases for each built-in skill, you will be well-prepared for the exam and for real-world knowledge mining projects. Cognitive skills are the bridge that turns raw, messy data into valuable, actionable insights.

---

Practice questions and the full interactive page: https://courseiva.com/glossary/cognitive-skill