Routing A/B Tests

A/B tests let you compare the performance of different PSP connections under real traffic. Instead of guessing which provider is best for a region or payment method, you can run a controlled experiment and let the data decide.

A/B Tests are an advanced feature for optimizing your payment stack. Use them when you have multiple connections for the same region and want to find the highest-performing option.

🎯 When to Use A/B Tests

Use Case	Why A/B Tests Help
Selecting a new provider	Test a new PSP alongside your current one to compare success rates before fully switching.
Optimizing by region	Compare Paystack vs. Hubtel for Ghana mobile money payments.
Fee comparison	Determine if a slightly more expensive provider has better success rates that offset its cost.

⚙️ How It Works

Create an Experiment: Define at least two variants, each pointing to a different connection_id with a weight (traffic allocation).
Set Targeting: Optionally limit the experiment to specific currencies, countries, or payment methods.
Start the Test: Activate the experiment. Reevit's router will split traffic according to your weights.
Analyze Results: View real-time success rates, latency, and costs per variant.
Declare a Winner: Complete the test and apply the winning connection to a permanent routing rule.

🛠️ Creating an A/B Test

Via SDK

import { Reevit } from '@reevit/node';

const test = await reevit.abTests.create({
  name: 'Ghana MoMo Provider Comparison',
  description: 'Hubtel vs Paystack for MTN Mobile Money',
  variants: [
    { connection_id: 'hubtel_gh_live', weight: 50 },
    { connection_id: 'paystack_gh_live', weight: 50 }
  ],
  traffic_percentage: 20, // Route 20% of eligible traffic to this test
  target_countries: ['GH'],
  target_methods: ['momo']
});

📋 Field Reference

Field	Type	Required	Description
`name`	string	✅	A descriptive name for the experiment.
`description`	string		Explain the purpose of the test.
`variants`	array	✅	At least two objects, each with `connection_id` (string) and `weight` (integer percentage).
`traffic_percentage`	integer		Percentage of eligible traffic (1-100) to route through the test. Default: `100`.
`target_currencies`	string[]		Limit test to specific currencies (e.g., `["GHS"]`).
`target_countries`	string[]		Limit test to specific country codes (e.g., `["GH", "NG"]`).
`target_methods`	string[]		Limit test to specific methods (e.g., `["card", "momo"]`).
`start_at`	string		ISO 8601 timestamp to automatically start the test.
`end_at`	string		ISO 8601 timestamp to automatically complete the test.

📊 Test Lifecycle

Status	Description
`draft`	Test is created but not yet active. Traffic is not being routed.
`running`	Test is live. Traffic is being split among variants.
`paused`	Test is temporarily stopped. Traffic reverts to default routing.
`completed`	Test is finished. Results are frozen for analysis.

Managing the Lifecycle

await reevit.abTests.start('abt_xyz123');
await reevit.abTests.pause('abt_xyz123');
await reevit.abTests.complete('abt_xyz123');

📈 Analyzing Results

Fetch the comparison report:

const comparison = await reevit.abTests.getComparison('abt_xyz123');
// {
//   variants: [
//     { connection_id: 'hubtel_gh_live', success_rate: 0.94, avg_latency_ms: 450, total_payments: 1200 },
//     { connection_id: 'paystack_gh_live', success_rate: 0.91, avg_latency_ms: 520, total_payments: 1180 }
//   ],
//   winner: 'hubtel_gh_live',
//   statistical_significance: 0.95
// }

✅ Best Practices

Sufficient Volume: Run tests long enough to gather statistically significant data (usually 1,000+ transactions per variant).
One Variable at a Time: Only compare connections; don't change targeting rules mid-experiment.
Start with Low Traffic: Use traffic_percentage to initially test with 10-20% of traffic, then scale up.
Apply Learnings: After completing a test, create a Routing Rule to make the winning connection your primary.

Smart Routing

Apply your test results as permanent routing rules.

Connections

Manage the PSP accounts you're comparing.

Payments

See how individual payments are attributed to test variants.

Reevit

A/B Testing