UCP Checkout Speed: Benchmarks and Conversion Data

The Question Everyone Asks

Developers building UCP integrations ask the same question early: how fast is it? Not in theory. In production, with a real agent, a real merchant endpoint, and a real payment provider in the chain.

The answer is not as simple as "faster" or "slower" than standard web checkout. It depends on where you measure, what you are comparing, and which part of the checkout flow you care about.

This post breaks down what we know about UCP checkout performance based on published benchmark data and merchant reports from the first months of production UCP traffic.

What to Measure

Standard web checkout has multiple phases from a user's perspective: page load, form fill, payment entry, submit, confirmation. The bottleneck is almost always human: filling in shipping and payment details.

UCP checkout is different. There is no human in the loop during execution. The agent does everything. The phases are:

1. Manifest discovery: fetch /.well-known/ucp from the merchant domain
2. Catalog query: query product availability and pricing
3. Identity token retrieval: fetch a short-lived identity token from the user's identity provider
4. Checkout initiation: POST to the merchant's UCP checkout endpoint
5. Payment token exchange: the merchant's payment handler resolves the token
6. Order confirmation: receive the order ID and confirmation data

The agent controls steps 1, 2, 3, and 4. The merchant and payment handler control steps 5 and 6. Most latency problems appear in steps 4 and 5.

Published Benchmark Data

Based on conformance testing data and early merchant reports from Q1 and Q2 2026, here is what typical UCP checkout looks like end to end:

These are server-to-server timings. The agent and merchant are both cloud-hosted. No browser rendering, no network from a consumer device.

For comparison, the median time for a human to complete a web checkout on a site with saved payment details (autofill enabled) is 45 to 90 seconds. UCP agent checkout is roughly 30 to 60 times faster for the payment execution step.

Where Latency Comes From

The biggest variable is payment token exchange. This step involves a round trip from the merchant's checkout endpoint to the payment handler (Stripe, Adyen, etc.) to the card network and back. Three network hops. Cold start latency is the biggest driver of P95 spikes.

Manifest discovery is consistently fast for merchants hosted on major platforms. Shopify-hosted UCP endpoints typically respond in under 150ms. Self-hosted implementations vary more, with cold start behavior (serverless functions, containerized APIs) responsible for most P95 outliers.

Identity token retrieval depends on the user's identity provider infrastructure. Providers running on distributed cloud infrastructure are consistently under 200ms. Providers with single-region deployments show geographic latency for agents and merchants on different continents.

What This Means for Conversion

The conversion metric that matters for agentic commerce is not page load time. It is session completion rate: the proportion of UCP checkout sessions that reach order confirmation.

Early merchant data shows session completion rates between 87 and 94 percent for successful UCP implementations. The failure modes are:

Payment token expiry. UCP payment tokens are short-lived, typically 5 to 15 minutes. If an agent queues a checkout session and does not execute it quickly, the token expires and the session fails. Agents need to be designed to use tokens promptly after retrieval.

Inventory race conditions. A product can be in stock when the catalog query runs and out of stock when checkout initiates. This is rare but needs a retry path in agent code. Well-designed agents query inventory close to checkout initiation, not minutes earlier during the research phase.

Network timeouts. P95 latency approaching 4 seconds means that 5 percent of checkout sessions take longer than most agents' default timeout settings. Set checkout timeouts at 10 seconds minimum. Implement retry with exponential backoff for the payment exchange step specifically.

Improving Your Numbers

If your UCP checkout is slower than the medians above, three places to look:

Cold start latency. If your UCP endpoints run on serverless functions (Lambda, Cloud Functions, Cloud Run), cold starts are your biggest P95 driver. Keep functions warm with scheduled pings, or move latency-critical endpoints to a containerized service with persistent instances.

Payment handler region. Stripe and Adyen both run payment processing in multiple regions. Configure your merchant account to use the region closest to your average agent and shopper geography. A cross-continental payment round trip adds 200 to 400ms.

Identity provider caching. If your identity provider verifies tokens on every request by hitting a database, add a short-lived cache (Redis, Upstash) in front of token verification. Tokens are cryptographically signed and can be safely cached for their validity window minus a few seconds of buffer.

The Bottom Line

UCP checkout at median performance completes in about 1.4 seconds of server-to-server time. That is fast enough that session completion rate, not raw speed, is the metric worth optimizing.

Get your session completion rate above 90 percent first. Then look at P95 latency if merchant analytics show agent sessions timing out or being abandoned.

The merchants and payment handlers with the best-performing UCP implementations are tracked in the UCPList directory. Conformance scores, which correlate with production reliability, are visible on each listing where published.