Predictive vs. generative AI
A bit of a trick question — both are prediction. Here's the real difference, with familiar Salesforce examples.
You'll hear two phrases thrown around constantly: predictive AI and generative AI. They're often presented as two completely different things. Here's the slightly sneaky truth:
Both of them are doing prediction. That's the trick question.
Generative AI feels new and magical, while "predictive" sounds older and more boring — but under the hood, a large language model generating an email is predicting the next token, over and over. It's prediction all the way down. So if they're both prediction, what's the actual difference? It comes down to scope and what they predict.
Traditional predictive AI: narrow and task-specific
Traditional predictive AI is built and trained to do one specific job, and to output a specific kind of answer — usually a number, a category, or a probability. You've almost certainly seen these inside Salesforce already:
- Lead scoring — predicting how likely a lead is to convert.
- Churn prediction — predicting which customers are likely to leave.
- Sales forecasting — predicting next quarter's numbers.
- Case routing — predicting which team a support case should go to.
Each of these is a sharp, narrow tool. A churn model can't write you an email, and an email model can't score your leads. They're trained for one task and they're very good at exactly that task.
Generative AI: broad, and predicts content
Generative AI — the large language models we've been discussing — predicts content: text, and increasingly images, audio, and code. Instead of one narrow job, it's a general-purpose engine you can point at an open-ended range of tasks: summarize this, draft that, explain this, rewrite that.
The thing that makes it feel different isn't that it stopped predicting — it's that what it predicts (open-ended content) and the range of things you can ask it (almost anything) are both enormously wider.
The punchline
| Traditional predictive AI | Generative AI | |
|---|---|---|
| What it predicts | A number, category, or score | Content (text, images, code…) |
| Scope | One narrow task | Broad, open-ended |
| Example | Lead scoring, churn | Drafting an email, summarizing a case |
| Under the hood | Prediction | Also prediction |
Both are prediction. The difference is scope and output — not whether they're "guessing." Keeping that straight will save you from a lot of confused conversations.
📝 Practice / Homework
Think about your own role. Name one task where a narrow predictive tool would fit best (something with a clear numeric or yes/no answer), and one task where a generative tool would fit best (something open-ended). Bring both examples — we'll sort them as a group.