What is supervised learning?
Quick answer
Supervised learning is a way of training AI on examples that come with the correct answer attached: emails labeled spam or not, receipts labeled by category, loans labeled repaid or defaulted. The model learns the pattern and then predicts the label for new examples. Unsupervised learning, by contrast, finds groups in data with no labels.
Last updated
Take the free module →Module 12 of the free AI for Accounting & Bookkeeping Firms course, about 12 minutes. No account needed.
Updated · By Robert Breen
Why it matters for a small business
Before large language models, most business AI was supervised learning: spam filters, fraud flags, demand forecasts. It still runs behind the scenes in many tools, and it needs a lot of carefully labeled examples.
Language models changed the starting point. You can often get a decent classifier just by describing the categories in a prompt, or showing a few examples, with no training at all. Supervised training is still worth it for very high volume or when the prompt approach isn't accurate enough.
If a vendor says its product is trained on labeled data, ask whose data, how many examples and how they measure accuracy on new cases. Those three answers tell you more than the word AI on the box.
In a real lesson: AI Receipt Extractor: Receipts to Google Sheets with n8n
In the AI Receipt Extractor, the model decides the Corner Mart receipt is groceries and writes that to the Category column, with no labeled training set. It follows the instructions and what it learned in general training.
A traditional supervised approach would need hundreds of receipts already labeled by a person before it could categorize a new one. That is the shift: you describe the categories and check results, and labeled examples become your test set rather than your training set.
Try this lesson free or read the step-by-step guide.
Common confusions
Supervised learning vs few-shot prompting
Supervised learning changes the model by training it on many labeled examples. Few-shot prompting shows a handful of examples in the prompt and changes nothing permanently.
Supervised vs unsupervised learning
Supervised learning learns from answers you provide. Unsupervised learning finds patterns, like customer groups, without any answers.
Tips
- Try a clear prompt with category definitions before training anything.
- Keep a labeled sample to test the AI's choices.
Related terms
More AI basics terms
Where you use it: free lessons
- AI Receipt Extractor: Receipts to Google Sheets with n8n (n8n, 12 min)
- Give Examples: Few-Shot Prompts That Copy Your Style (ChatGPT, 6 min)
Frequently asked questions
- Is ChatGPT trained with supervised learning?
- Partly. After pretraining, models are usually trained on example answers written by people, then on human preferences.
- Do I need labeled data to use AI?
- Not to start. You need a small labeled set to check the AI's accuracy.
- What is an example of supervised learning at work?
- A spam filter trained on emails people marked as spam, or a model that predicts late payers from past invoices labeled paid on time or late.