NLP: Working with Language
Natural language processing (NLP) focuses on understanding and transforming language.
Typical NLP tasks include:
<ul>
<li>classification</li>
<li>extraction</li>
<li>translation</li>
<li>search</li>
<li>summarization</li>
<li>question answering</li>
</ul>
If the core input or output is language, NLP is usually part of the solution.

Which feature is primarily an NLP problem?

NLP is especially useful when the product needs to understand, classify, or generate language.

Computer Vision: Working with Visual Inputs
Computer vision helps systems interpret images and video.
Typical tasks include:
<ul>
<li>OCR</li>
<li>object detection</li>
<li>image classification</li>
<li>visual search</li>
<li>screenshot understanding</li>
<li>scene analysis</li>
</ul>

Which e-commerce feature is a clear computer-vision use case?

OCR and screenshot understanding are computer-vision tasks.

Generative AI: Creating or Transforming Content
Generative AI is useful when the product needs to draft or create something new:
<ul>
<li>text</li>
<li>images</li>
<li>audio</li>
<li>video</li>
<li>code</li>
</ul>
For PMs, a key question is whether the AI is producing a final artifact or a draft that a user will edit.

Which workflow is a strong fit for generative AI?

Generative AI often works best when users can review, edit, or constrain the output.

Multimodal Products Combine Modalities
Many modern AI experiences are multimodal. They combine text, images, audio, video, and tool results in one workflow.
Examples include:
<ul>
<li>uploading a screenshot and asking what went wrong</li>
<li>asking questions about a chart image and spreadsheet text</li>
<li>summarizing a recorded meeting with slides and transcript together</li>
</ul>

Which experience is most clearly multimodal?

Many useful AI products combine modalities instead of relying on only one input type.

Choose the Modality Based on the Job
The right technique depends on:
<ul>
<li>the user's input format</li>
<li>the needed output</li>
<li>acceptable error tolerance</li>
<li>latency and cost constraints</li>
<li>privacy requirements</li>
</ul>
Do not start with the flashiest model. Start with the job to be done.

A clinician dictates a note while also attaching an image for interpretation support. Which combination best matches the task?

The right AI technique depends on the user workflow, not whichever model category is currently the most fashionable.

The PM Decision Rule
Strong PMs can explain why one AI technique fits the problem better than another.
That means they can say things like:
<ul>
<li>"This is mostly a retrieval and summarization problem."</li>
<li>"This needs vision because the input is visual."</li>
<li>"This is a generation task, but humans still need approval."</li>
</ul>
That level of clarity leads to better product choices.

Your team wants to add AI "everywhere." What is the strongest PM response?

A strong PM can explain why one AI technique fits the workflow better than another.