Knowing when not to use AI is as important for the exam as knowing when to use it. Machine learning gives you a prediction, a best guess learned from data, and it costs money to build, run and maintain. If the business needs an exact answer, has little data, or would gain less than it spends, a simpler solution is usually better. AWS's own guidance tells teams to check whether ML is the right approach before they build anything.
Predictions vs Exact Answers
A model learns patterns from examples, so its output is probabilistic: usually right, sometimes wrong. That's fine for recommending a product or estimating demand. It isn't fine when the answer is fixed by a known formula or rule, such as calculating sales tax, applying a contract's overtime rate, or checking whether a date has passed.
If you can write the logic down as clear rules, a few lines of ordinary code will be cheaper, faster and always correct. An ML model would only add the chance of error. Rules are also easy to audit: anyone can read them and check that they match the policy.
Cost-Benefit Analysis
AI projects have real costs:
- Data: collecting, cleaning and labeling examples.
- Building: training and testing a model, often several times.
- Running: paying for inference every time the model is used.
- Mistakes: the cost of wrong predictions, and of handling them.
Weigh those against the value the model creates, measured in business terms such as hours saved, revenue gained or losses avoided. If a small business would save a few dollars a day with a model that costs thousands to build, the numbers don't work, however interesting the technology is.
Data and Risk Constraints
Machine learning needs enough relevant, good-quality data. A handful of records, data that doesn't represent real conditions, or data the company isn't allowed to use are all signs that ML isn't ready yet.
Risk matters too, especially in areas such as health, credit and safety. When a wrong answer could seriously harm someone, and nobody can check or explain the model's output, a rules-based process or a human decision may be the responsible choice.
A Quick Checklist
Before choosing AI, ask:
| Question | If the answer is no... |
|---|---|
| Is a prediction acceptable, or is an exact answer required? | Use rules or ordinary code |
| Is there enough relevant, usable data? | Collect data first, or skip ML |
| Will the value exceed the cost to build and run it? | Use a simpler approach |
| Can mistakes be detected and handled? | Keep a human in the decision |
On the Exam
When a scenario stresses a fixed formula, a legal requirement for an exact result, very little data, or a tiny budget, the right answer often avoids ML entirely. Don't assume that a more advanced option, such as deep learning or a foundation model, is better. The exam rewards matching the tool to the problem.