# Query Learning

A **Query Learning** fixes how Kaiya should interpret a full question pattern (aggregation, grouping, filters, time windows, ranking). It must include an **Intent** so Kaiya can reuse the learning for similar phrasings.

1. Navigate **Kaiya → Open Side Bar → Manage Learnings → Query Learnings**. The following page will be displayed. 

<figure><img src="https://content.gitbook.com/content/8GaK1h3pmgbR63x0ftET/blobs/7Y9Pi1thoivsTAh6EtZL/image.png" alt=""><figcaption></figcaption></figure>

2. You can sort and order the Learnings as follows.

<figure><img src="https://content.gitbook.com/content/8GaK1h3pmgbR63x0ftET/blobs/9Ckdv3iKhXYPlPVBkgzR/image.png" alt="" width="287"><figcaption></figcaption></figure>

3. The following will be displayed for each Learning:

* **Query:** literal phrase/alias for which the Learning has been created
* **Intent/mapping:** the resolved meaning for the phrase
* Creator name, created time. and the timestamp of last update.
* Click on the three-do kebab menu to edit or delete a Learning.

{% hint style="warning" %}
Query Learnings are org-wide patterns. You’ll notice there’s no **Learning Level** or **Selected Business View** field in the Query Learning dialog; only admins can create them and they apply across the org. BV is resolved at run time from the user’s active context. If you need per-user or per-BV semantics, use Phrase Learnings.
{% endhint %}

4. Click on **Create Learning** button and the following window will be displayed.&#x20;

<figure><img src="https://content.gitbook.com/content/8GaK1h3pmgbR63x0ftET/blobs/tG2rz1vS2904n7nEbUpK/image.png" alt="" width="563"><figcaption></figcaption></figure>

5. **Query:** Enter a representative natural-language question users actually ask. Avoid single-word aliases like supplier → use **Phrase Learning** for that.
6. **Query Type:** Choose whether you want to add an analytical query or a compound query.

* **Analytical Query -** Use for a single analysis pattern (one result set) with clear aggregation/grouping/time/filter/ranking. Examples: Top-N per group; threshold + grouping; single MoM/YTD compare.
* **Compound Query -** Use for questions that combine multiple analytical operations into one Learning (multi-step or multi-output logic you want treated as a unit). Examples: *“Show MoM revenue by channel and highlight segments where TRx > 150.”, “Compare actual vs plan vs forecast by region with a ranked variance list.”*

7. **Intent:** Briefly describe in natural language what the query is intended to achieve. This helps Kaiya understand and apply the Learning accurately. Include:
   * **Aggregation(s):** sum/avg/count/calculation names
   * **Grouping:** the exact “group by” fields
   * **Filters/thresholds:** value conditions, segments
   * **Time logic:** window and/or Period A vs Period B
   * **Ranking/limits:** order by, Top-N, partitions

{% hint style="warning" %}
**Why Intent is mandatory?**

Different phrasings (“top three per region”, “best 3 by region”) map to the same structured action because the **Intent** is the anchor. With **Intent**, Kaiya doesn’t rely on exact tokens; it “knows” what to build even if synonyms or order change, reducing follow-up questions.
{% endhint %}

8. **Custom Mappings:** Turn on to bind tokens from the query to exact Business View fields, eliminating column ambiguity. Ideal for situations when multiple synonyms or overlapping names exist. Click on **Add Mapping** and the following window appears.

<figure><img src="https://content.gitbook.com/content/8GaK1h3pmgbR63x0ftET/blobs/VtnEBwTxATfbefmdfTaF/image.png" alt="" width="262"><figcaption></figcaption></figure>

9. **Mapped Phrase:** The exact token from the query you’re pinning. Kaiya will resolve this token using the mapping you define.
10. **Type:** What kind of thing the phrase represents.

* **Aggregation:** Picks the metric and how to roll it up. Use when the word should always resolve to a specific metric/aggregation.
* **Dimension:** Sets the *“group by”* field to break down results.
* **Resolution:** Sets the time grain used for grouping without adding a date filter. Use for phrases like *“weekly revenue”.*
* **Filter:** Applies a non-time condition on a column
* **Time filter:** Constrains the time window or comparison periods

11. **Business View:** The specific BV that contains the column you want to bind to.
12. **Column:** The exact field in that BV to map to (pick the modeled measure with its aggregation, or the precise dimension/flag).
13. Click on **Apply** to save the Learning, or click on **Cancel** to dismiss.
14. Finally, click on **Save Changes** to create the Learning. Or click on **Cancel** to dismiss.
15. If you want to delete a Learning, click on **Delete Learning** button. All the above fields are applicable when editing a Learning.