> ## Documentation Index
> Fetch the complete documentation index at: https://docs.thetinybell.com/llms.txt
> Use this file to discover all available pages before exploring further.

# A/B testing

> Test different offers and designs to find what converts best.

## How A/B testing works

When you have multiple active notifications of the same type in a property, TinyBell automatically splits traffic between them. Each visitor sees only one variant, and TinyBell tracks which one gets more clicks and conversions.

## Setting up a test

1. Create your first notification (Variant A).
2. Clone it or create a second notification (Variant B).
3. Change one element in Variant B (e.g. the headline, offer, or button text).
4. Enable both notifications.
5. TinyBell splits traffic 50/50 automatically.

<Warning>
  Change only one element per test. If you change the headline AND the image AND the offer, you won't know which change made the difference.
</Warning>

## What to test

| Element          | Example A              | Example B                |
| ---------------- | ---------------------- | ------------------------ |
| **Headline**     | "Book Direct and Save" | "Skip the OTA, Save 15%" |
| **Offer amount** | 10% discount           | 15% discount             |
| **CTA button**   | "Book Now"             | "See Rates"              |
| **Trigger**      | Page load              | Exit intent              |
| **Image**        | Room photo             | Pool photo               |
| **Urgency**      | No urgency             | "Only 3 rooms left"      |

## Reading results

Go to **Statistics** to compare performance:

* **Impressions**: How many times each variant was shown.
* **Clicks**: How many visitors clicked the CTA.
* **CTR**: Click-through rate (clicks / impressions).

<Tip>
  Wait for at least 500 impressions per variant before drawing conclusions. Small sample sizes can be misleading.
</Tip>

## Best practices

* Run one test at a time per property.
* Let tests run for at least 7 days to account for weekday/weekend differences.
* Once you have a winner, disable the losing variant and create a new test against the winner.
* Keep testing. There is always room to improve conversion rates.