Write down the decision you want to improve
Start with a specific buying question. Perhaps customers hesitate between a 5 by 8 and an 8 by 10, or repeatedly ask whether a dense pattern will overwhelm a small seating area. That question determines which rugs belong in the pilot and what useful engagement looks like.
Avoid making the pilot a contest for the largest number of launches. A button may attract curiosity without improving product understanding. Your working objective might be to help shoppers compare two relevant sizes, then continue to the correct product variant with fewer unanswered questions.
Choose a fair comparison
Select products with enough existing traffic to observe a pattern, while keeping their merchandising and availability reasonably stable. Record the starting price, promotion, stock status and page changes. A new discount or an out-of-stock bestseller can affect results more than the visualizer.
Decide in advance how you will compare performance. A controlled experiment is stronger than comparing everyone who voluntarily used a preview with everyone who did not. Users who choose to explore a product may already have greater purchase intent. If you cannot run a controlled test, label the findings as observational.
Separate use from business outcomes
Track the steps that your implementation can actually record: visualizer launch, selected rug or variant, room upload, successful preview and return to the product page. Confirm those events fire correctly before the pilot starts. Do not label a launch as a completed room preview.
Then examine product-page progression, add-to-cart activity and completed orders using your existing commerce data. Keep these measures separate from usage. A broken event name should not be mistaken for a drop in customer interest, and an attractive preview should not be counted as a sale.
Watch cost and failure points
Preset room browsing and paid actions are different. Vizbl's published token rules charge separately for each own-room upload and each AR object load. Repeated attempts can therefore affect usage cost even when the shopper never completes a purchase.
Review unsuccessful uploads, unavailable service states and questions from customer support alongside the funnel. If customers repeatedly restart, investigate whether instructions, image quality or an unclear next step are responsible. Removing a practical obstacle may be more useful than making the launch button larger.
Give the findings time to mature
Orders and returns happen on different timelines. Record when pilot purchases were delivered and whether the normal return window has elapsed before drawing conclusions about returns. Use actual return reasons where available, rather than assuming every return was caused by size or appearance.
At the review, decide which products should stay, which need better assets and what needs another test. A small pilot may produce useful qualitative findings without proving a conversion lift. Keep those findings concrete: the sizes customers compared, the questions they asked and the page changes that would help them choose.