How to Measure the Success of AI Photos and Videos After Publishing in E-commerce

Cover image: production by DELFI for Falabella.

Useful measurement starts after publishing. Before that, everything is still hypothesis, enthusiasm, or fear. To know whether AI photos and videos are truly working, it helps to look at business metrics, operational metrics, and visual quality at the same time. If you only look at subjective taste, you miss the real picture.

What is worth tracking

• click-through from PLP to PDP
• add-to-cart rate
• conversion by product family
• publishing speed by collection
• cost per approved asset
• correction rate by batch
• visual consistency across SKUs and campaigns

It also helps to compare by category. A system may perform extremely well in basics and still need more adjustment in tailoring. Or it may accelerate catalog output while video still needs refinement. That detailed reading helps decide where to scale faster and where to ask for tighter control.

DELFI supports this stage well because it does not only create visuals. It structures the workflow so the outcome becomes measurable. Its concierge service reduces internal time, and its brand-specific training in fabrics, fit, and style supports consistent review. That affects something very concrete: more assets ready to publish, less rework, and a real ability to scale catalog production in agreed batches. Measuring well is not about chasing one magical number. It is about knowing whether the visual system makes it easier to sell better and publish faster.

What you measure and its denominator

• Conversion: orders divided by sessions in the observed group.
• Revenue per session: revenue attributed to the group divided by its sessions, using the same currency and revenue definition.
• Add-to-cart rate: sessions adding to cart divided by group sessions, if that is the agreed definition.
• Relative change: new result divided by baseline result, minus one; do not calculate it when the baseline is zero.

Record the exact definition for both groups. More clicks or approved files can be useful, but they are not equivalent to more sales.

How to document a proven result

Keep the data source, period, country, SKUs, group sizes, visual change and conditions for price, stock, traffic and promotions. Distinguish controlled tests from before-and-after comparisons. A confidential source does not authorize publishing figures or client attribution without permission. DELFI’s track record does not guarantee the same outcome for another brand.

Post-publication record

• Category, country, device and period:
• SKUs and visual change:
• Actual publication date:
• Commercial metric and denominator:
• Changes in price, stock, traffic or promotion:
• Corrections and internal hours:

If several conditions change, describe an association rather than proof of causality.

Reading a published result: the Topper case

DELFI publishes metrics reported by Topper: −32% in visual production cost and +7% in ecommerce conversion for an equivalent content scope. The summary does not specify period, sample size or whether +7% means relative change or percentage points. It is a precedent with limited scope, not a reproducible A/B test or a guarantee for your brand.

Document those details before publishing to make your own measurement comparable. Another client’s result does not fill gaps in your experiment.

Continue reading

• Visual A/B testing in fashion: what to test with AI photos
• Generic AI vs a Premium Fashion Solution: How to Justify On-Brand Production

Your next collection

DELFI directs AI virtual talent toward content that sells, backed by a proven sales track record. A collection’s own result needs data and a valid comparison.

What product do you want to promote, and what does its shopper need to see? Use that answer to agree the talent, usage and production scope.