Context
Search…
Best Practices
Recommendations to get the most out of Context Experiments

Build Around Feature Gates

The more values you get from Context, the more experiments and changes your team can make without code changes. That means marketing and product can have more independence from developers, and developers can spend their time on bigger projects.
Implementing a Banner like this requires developer resources for every change.
1
const Banner = () => (
2
<div>
3
<p>
4
$5 off when you use code YW2HH3`
5
</p>
6
</div>
7
);
Copied!
Pulling values from Context allows you to change and experiment with the banner's copy, discount amount, and discount code without needing to use developer resources.
1
import { useFeature } from "@context-experiments/react";
2
3
const Banner = () => {
4
const [, { discountAmount, copy, discountCode }] = useFeature("my-banner-j2dl3");
5
6
return (
7
<div>
8
<p>
9
`$${discountAmount} ${copy} ${discountCode}`
10
</p>
11
</div>
12
);
13
};
Copied!

Use Hypotheses to Monitor your Critical Events

Teams will often manually monitor critical events after a new feature launches by checking analytics dashboards. If you add these events and properties as hypotheses, we will include them in our statistical models to show you how they change, including if there is no change. You can do this by setting them as a hypotheses with 0% expected difference between groups.
Large changes will generally show quickly. If there is a small change or no change, it will take longer, but you can know with certainty that your most important metrics aren't being negatively effected.
If you have a background in traditional, aka frequentist, statistics accepting the null hypothesis should set off alarm bells. Our backend uses Bayesian Estimation to do things that aren't possible using t-tests, like showing with certainty that there is no change between groups. For more on this see the article Bayesian Estimation Beats T-tests or reach out to us.

Capture Experiment Meta Data to Do Your Own Analysis

Context exposes the active experiment and variant id for every user. If you capture this data in your existing data pipeline (e.g. via Segment identify events) you can use that data to explore all your events, grouped by experiment and variant seen. In other words you can manually explore how all your events changed between variant groups using your existing tools. See the FAQ for more information on implementation.
Last modified 8mo ago