A workload is a category of problem that AI solves. The first skill AI-900 measures is recognizing the workload behind a business scenario, because once you know the workload you can choose the technique and the Azure service. Exam questions usually describe a need in plain business language and ask which workload it is.
Prediction and forecasting use historical data to estimate an unknown or future value: next month's sales, the price a house will sell for, or whether a customer will cancel a subscription. These are classic machine learning tasks (regression for numbers, classification for categories). Anomaly detection learns what normal looks like and flags unusual events, such as a fraudulent card transaction, a sudden spike in server errors or a sensor reading that suggests a machine is about to fail.
Computer vision interprets images and video: classifying photos, detecting and locating objects, reading text in images (optical character recognition, OCR) and analyzing faces. Natural language processing (NLP) works with written language: detecting the language, judging sentiment, extracting key phrases and entities, summarizing and translating. Speech is closely related: converting speech to text and text to speech. Conversational AI combines language and speech to build bots and assistants that hold a dialog.
Document processing (often called document intelligence) extracts structured data from forms and documents such as invoices, receipts and ID cards: not just the text, but which value is the total and which is the due date. Knowledge mining goes further, extracting information from large volumes of unstructured content and building a searchable index, so people can find insights across thousands of files.
Generative AI creates new content in response to a prompt: text, code, images and more. It powers copilots and chat assistants that draft, summarize and answer questions, and agents that can also take actions. Unlike the other workloads, which analyze existing data, generative AI produces something that did not exist before, which brings its own risks such as fabricated answers.
Real solutions often combine workloads. A claims app might use OCR to read a form, entity recognition to find the policy number, anomaly detection to flag suspicious claims and a generative model to draft a letter. On the exam, focus on the main need described in the question.
Key terms
- Workload
- A category of problem that AI addresses, such as computer vision, NLP or anomaly detection.
- Anomaly detection
- Identifying data points or events that differ significantly from the normal pattern.
- Knowledge mining
- Extracting information from large amounts of unstructured content and making it searchable.
- Generative AI
- AI that creates new content, such as text, code or images, from a prompt.
An airline uses forecasting to predict passenger numbers per flight, anomaly detection to spot unusual engine sensor data, computer vision to check baggage images, NLP to analyze complaint emails, and a generative AI assistant to help agents draft rebooking messages.
Check yourself
A factory wants alerts when vibration readings on a motor look unusual. Which workload is this?
Anomaly detection, because it flags readings that differ from the normal pattern.
What distinguishes generative AI from the other workloads?
It creates new content from a prompt, whereas the others analyze or extract information from existing data.