Enriching your data

1. Enriching your data

Cleaning data gets you to a usable baseline. Enrichment is where you add the analytical value — new fields, derived metrics, and signals that didn't exist in the raw input. This is where AI starts to feel less like an assistant and more like a collaborator.

2. AI as advisor and doer

AI plays two distinct roles in enrichment. As a doer, it executes the work — writing the calculation, classifying records, extracting fields. As an advisor, it tells you what's worth enriching in the first place. Most analysts use AI heavily as a doer. The advisor role is the one that's underused.

3. Enriching structured data

Take structured data — the fields you already have in your tables. The advisor role is most useful here: ask AI what new fields would be worth deriving.

4. Enriching structured data

It'll come back with options grounded in your data and the question you're trying to answer — recency, frequency, monetary value, churn risk, segment classifications, derived ratios, etc. It will tirelessly generate ideas based on your needs. Then have it write the calculation. Both halves of this process matter: the suggestion and the execution.

5. Build it where you work

And that calculation doesn't have to be a one-off that lives only in a chat window. Ask it to explain how to build the enrichment in your BI tool — a Tableau calculated field, a Power BI measure, a dbt model — and you get something reusable, governed, and embedded in your stack.

6. The unstructured unlock

The bigger shift is what AI does with unstructured data — text, transcripts, images. Work that used to require specialist NLP or computer vision tools is now a prompt away. You can pull signals from text, audio, or images without leaving your AI assistant.

7. Examples

For example: Transcribe support calls and classify them by caller intent. Score review text by sentiment. Group social media posts to surface emerging trends. Scan product photos for competitor logos. Each of these used to be a dedicated machine learning project — custom models, training data, weeks of work. AI hasn't made specialist models obsolete. But for many cases, a prompt now does what used to take a project — and it makes exploration much faster.

8. A worked example

Here's one concrete case. We have a dataset here of customer reviews for a leather goods manufacturer. We ask the AI to categorize each review by sentiment: positive, negative, and neutral. The output is a new structured dataset we can analyze alongside the rest of our data. Free text turned into a categorical field, in seconds.

9. Spot-check the judgment calls

One important caveat. AI makes subjective decisions when it enriches — what counts as positive sentiment, where one category ends and another begins. Those judgments need to match how you and your business define them. Provide as much detail before it acts. And after it’s done the work, sample what it produces, read enough rows to know the calls hold up, and refine the prompt where they don't. It's your name that will be on the results.

10. Let's practice!

Let’s go enrich some data!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.