Shopify Unveils SimGym, an AI Tool for Synthetic A/B Testing

Shopify's new AI tool for synthetic A/B testing, SimGym, gives merchants a revolutionary way to test store changes with virtual shoppers. This eliminates the need to wait for real visitor data, allowing even low-traffic stores to quickly identify UX issues and optimize conversions. By removing the high-traffic requirement of traditional A/B testing, SimGym democratizes the powerful optimization cycles once reserved for high-volume brands, helping all stores find winning layouts before real shoppers ever visit.

From Winter '26 launch to 400,000 sessions a day

Shopify's SimGym is an AI-powered simulation tool that tests store designs using thousands of synthetic shoppers. It delivers rapid, low-cost feedback on user experience and conversion potential, allowing merchants to identify winning layouts and fix friction points before launching to live traffic, effectively democratizing A/B testing.

Debuting as part of the Winter '26 Edition on February 11, 2026, SimGym was immediately recognized as a "material launch" that significantly enhances Shopify's AI infrastructure Scouts by Yutori. A subsequent Shopify Engineering post detailed its impressive scale: a capacity of 400,000 simulated sessions daily, achieved with a 5.2× speedup on Blackwell B200 GPUs.

Why 2000 robots walk into a shop: Simulated A/B testing (2026) matters for small merchants

Traditional A/B testing is often unfeasible for the majority of Shopify stores that receive fewer than 1,000 visits per week, as they lack the volume for statistically valid results. SimGym overcomes this by deploying hundreds of AI agents, trained on billions of real transaction data points. These synthetic shoppers mimic human behavior - browsing collections, viewing products, and attempting checkouts - to reveal critical friction points before any real customer is affected.

What a five-minute run produces

A single simulation, costing around $10, provides merchants with a comprehensive report in minutes. Key deliverables include:
- Quantitative metrics comparing add-to-cart rates and average order value between variants.
- Qualitative feedback identifying friction in navigation, search, and checkout processes.
- A ranked list of theme variants scored by the success of synthetic shoppers.

Shopify advises using SimGym as a powerful directional filter to eliminate weak designs, with final validation of winning variants conducted through live traffic tests.

Hardware muscle courtesy of NVIDIA

The tool's remarkable speed and affordability are powered by cutting-edge NVIDIA hardware, with model serving optimized by the NVIDIA-acquired CentML. Running on Blackwell GPUs allows SimGym to process 57,000 tokens per second per chip. This efficiency keeps the cost for merchants in the single digits per run and enables the rapid four-minute test results.

Early reception: intrigue outweighs skepticism

Initial reactions to SimGym have been overwhelmingly positive, with industry watchers dubbing it a "flight simulator for stores." Commentators praised its ability to accelerate UX iteration and provide a low-risk sandbox for testing bold design changes. While no significant negative feedback has surfaced, the consensus advises merchants to use the simulated results as strong guidance rather than absolute truth.

Competitive landscape

In the specific field of synthetic shopper testing for e-commerce UX, SimGym currently stands alone. Competing solutions are either focused on different domains, like NVIDIA's warehouse-oriented Physical AI toolkit, or rely on less scalable methods like generic AI models and manual user testing. SimGym's ability to run 400,000 simulated sessions daily sets it apart.

What to watch next

The future of SimGym depends on validating a strong correlation between its simulated results and live A/B test data from Shopify Rollouts. If the connection proves reliable, synthetic testing is poised to become a standard, platform-wide pre-launch procedure for all theme changes, drastically lowering the cost and risk of experimentation for millions of merchants.

---

What is Shopify SimGym and how does it work?

SimGym is an AI Research Preview from Shopify and NVIDIA that simulates hundreds of AI-powered shoppers visiting your store. These agents interact with your site in a real browser, performing actions like clicking, adding to cart, and checking out. By selecting a store variant to test, merchants receive a detailed report on performance and UX friction in under five minutes.

Why is SimGym especially useful for low-traffic stores?

Stores with fewer than 1,000 weekly visitors typically cannot run effective A/B tests due to insufficient data. SimGym solves this by using AI shoppers trained on billions of historical transactions. This allows emerging merchants to test everything from navigation to pricing and fix issues before they impact real customers, achieving results that are impossible with traditional tools.

How much does a SimGym run cost and what do I get back?

A single SimGym run costs approximately $10 and delivers results within minutes. The output includes a comparison of conversion metrics, a friction report detailing the top UX issues found by the AI shoppers, and a recommendation for the theme variant most likely to perform best with live traffic. The low cost is made possible by efficient processing on NVIDIA Blackwell GPUs.

Is SimGym a replacement for live A/B testing?

No, SimGym is not a replacement for live A/B testing. Shopify positions it as a preliminary screening tool to quickly and cheaply eliminate underperforming designs. The recommended workflow is to first use SimGym to identify a promising variant, and then use a live A/B test tool like Shopify Rollouts to validate that choice with real customer traffic.

Where can I try SimGym today?

SimGym was introduced in the Shopify Winter '26 Edition and is available as an AI Research Preview within the apps section of the Shopify admin panel. Merchants can run a simulation by selecting a theme variant and clicking "Simulate shoppers," with no need for a separate NVIDIA account or any code modifications.