Product Recommendation Emails: How to Design Every Cell of the Product Grid (Image, Price, Stars, and the Data That Decides)

Product recommendation emails are the highest-revenue email blocks in retail and ecommerce, yet the individual cells within the product grid are often the least-designed elements in the entire message. Most brands treat the grid as a functional placeholder: algorithm picks the product, HTML renders the card, and that’s it. But the data says otherwise. When we look at real performance variation, the same recommendation logic can produce a 15x difference in revenue per thousand (RPM) depending on how each product cell is composed. The algorithm is table stakes. The cell is the frontier.

In Q4 2025, Zembula benchmarked 22 variants of a single abandoned cart Smart Block across our retail customers. The lowest-performing version earned $31.42 RPM. The highest? $469.65, normalized to a $100 average order value. That’s not a marginal improvement. That’s the difference between breaking even and compounding return on every email send. And the variable wasn’t the product. It was the design of the cell: image treatment, pricing visibility, star ratings, urgency cues, and how those elements were layered into a single, on-brand image at open time.

This post breaks down how to design every pixel of the product grid, not just which products appear, but how each one is presented. We’ll cover image types, data hierarchy, customer-specific composition, and how to measure what actually drives revenue. Because when 7% of shoppers generate 26% of revenue from clicking a recommendation, you can’t afford to leave the cell to chance.

The Product Grid Is Your Highest-Earning Block, And Its Cells Are the Most Ignored

It’s a paradox: the product grid earns more revenue than any other block in broadcast and triggered emails, yet it receives the least creative attention. Why? Because for years, we’ve treated it as a mechanical output, pull products, render cards, send. But that mindset ignores a fundamental truth about attention: predictable = invisible.

The Nielsen Norman Group has shown across two decades of eye-tracking research that users almost never look at content that appears in predictable positions or resembles an ad. That’s banner blindness in action. When your product grid looks the same every time, with identical HTML styling and fixed data fields, you’re training subscribers to skip it. And when the same shopper sees the same flat-lay image with price and stars in every post-purchase email, the cell fades into the background noise.

Yet the potential is massive. According to Salesforce Shopping Index data, 7% of ecommerce shoppers click on a product recommendation, but they generate 24% of orders and 26% of revenue. That concentration means even small improvements in cell design compound into significant P&L impact. The question isn’t whether to optimize the grid. It’s whether you’re optimizing the right thing.

Why the Abandoned Cart Lesson Applies to the Product Grid

In our post Email Variant Testing: 22 Abandoned Cart Combinations, a 15x Revenue Spread, and the Data Behind Both, we showed how minor changes to a single Smart Block, image type, coupon visibility, urgency language, could swing performance by orders of magnitude. The highest RPM version used an on-model lifestyle shot, a time-bound discount, and a star rating. The lowest? A flat lay with no pricing and no urgency.

That same variability applies to product recommendation emails. The grid isn’t one block. It’s a portfolio of 20 to 50 individual cells, each one a micro-experience. And each cell has the same variant surface: image, price, stars, badges, urgency, copy. The only difference is scale.

The problem with traditional HTML product cards is they’re static. You can’t A/B test flat lay vs on-model across 30 products without creating thousands of combinations. But with open-time composition, you can. Zembula’s Product Grid feature renders each cell as a Smart Block, a single, brand-controlled image composed at open time, with full control over typography, layout, and data hierarchy. That means you can test which image treatment wins, which data elements convert, and which combinations drive the highest RPM, without blowing up your ESP workflow.

Image Treatment Matters, By Audience and Context

Does a flat lay outperform an on-model shot? The answer: it depends. In browse abandonment emails, lifestyle and on-model images consistently outperform flat lays because they help the shopper visualize the product in use. But in post-purchase or replenishment emails, flat lays often win, clean, minimal, and focused on the product itself.

There’s no universal winner. But there is a universal rule: image treatment must align with intent. A 2025 analysis by Blend found that professional product photography improves ecommerce conversion by 10-33% across categories. But no study tests how those treatments perform in email by context. That’s where variant-level testing fills the gap.

For example, in a recent test for a premium apparel brand, on-model shots in a browse abandonment product recommendation email drove a 38% higher CTC (click-to-conversion) than flat lays. But in a post-purchase ‘you might also like’ email, flat lays performed 12% better. The audience was the same. The products were similar. The difference was intent, and the image had to reflect it.

Data Elements Earn Their Pixels, Not All Information Converts

More data in a cell doesn’t mean better performance. In fact, our Q2 2026 retail benchmark shows that urgency-heavy combinations pull more clicks but lower CTC, while BNPL callouts drop CTC to 6.1%. Why? Because not every click is valuable. Urgency badges attract bargain hunters. BNPL cues attract price-sensitive shoppers. Both increase volume but dilute quality.

Conversely, Cart + Coupon + Ratings & Reviews achieved a 24.6% CTC and $273.75 normalized RPM, nearly triple the $85.18 RPM of a basic cart reminder with no social proof. That aligns with the Spiegel Research Center finding that displaying reviews can increase conversion by 270%, especially for high-consideration products. But there’s a twist: eevy.ai’s 2026 analysis found that products rated 4.0-4.5 stars convert better than those rated 4.5-5.0, perfect ratings trigger skepticism.

So the rule isn’t “add star ratings.” It’s: add the right star rating to the right product for the right audience. And you can’t know that without testing. In Zembula, we layer star ratings only when review volume is between 11 and 30, based on PowerReviews data showing that threshold drives the highest lift. For products with 1-10 reviews, we suppress stars. For 5-star products, we may show ‘Top Rated’ instead of a full 5.0 badge.

The point? Every data element, price, stars, trending badge, urgency, must earn its place in the cell. And only variant-level attribution can tell you which ones do.

Personalization Goes Beyond the Product, It’s in the Cell

Most brands personalize the product. Few personalize the presentation. But the same product should not look the same to every subscriber. A first-time buyer sees a different cell than a loyalty member. A high-AOV customer sees pricing; a discount-driven shopper sees a coupon. This isn’t just segmentation. It’s real-time, open-time composition.

Zembula’s Product Recommendations engine uses signals like last order, browsing behavior, and lifetime value to pick the right product. Then, our open-time composition engine decides how to render it, as a flat lay or on-model, with or without price, with urgency or with social proof, based on what that individual is most likely to respond to.

And because we use Smart Banners and Smart Blocks, the entire cell is a single image. That means we preserve brand fonts, art-directed layouts, and pixel-perfect spacing, things HTML email can’t reliably deliver, especially across Gmail’s limited web font support.

How to Measure the Cell: RPM and CTC Together

You can’t optimize what you can’t measure. And most email analytics only go to the campaign or CTA level. But to improve the product grid, you need block-level and variant-level attribution.

In our guide to Email Block Analytics, we break down how to measure revenue from every module. But for the product grid, you need more: variant-level RPM and CTC, with 7-day click attribution.

Here’s the two-step test we recommend:

1. Replicate your current HTML product card as a composed image, same image, same data, same layout. This isolates the impact of rendering format. In most cases, just switching to a Smart Block lifts CTC by 5-7% due to better branding and consistency.
2. Test one variable at a time: flat lay vs on-model, price vs no price, stars vs no stars. Use equal-size content control via open-time composition so every subscriber sees one variant, no ESP splits needed.

The goal isn’t just more clicks. It’s higher-quality clicks. That’s why RPM and CTC must be read together. A variant might drive 2x more clicks but only 1.2x more revenue, meaning lower conversion quality. That’s what we saw with urgency-heavy cells: high volume, low CTC.

Key takeaways

• The product grid earns more revenue than any other email block, but its cells are often the least-designed elements.
• Image treatment (flat lay, lifestyle, on-model) must align with shopper intent and email context, test to know what wins.
• Not all data belongs in every cell. Urgency and BNPL cues can increase clicks but reduce conversion quality.
• Star ratings lift conversion, especially for high-consideration products, but perfect ratings (5.0) can trigger skepticism.
• Personalization must extend to the presentation: the same product, different cell per subscriber, composed at open time.
• Measure with variant-level RPM and CTC, never rely on campaign-level metrics alone.
• The algorithm is table stakes. The cell is the frontier. Test, measure, and optimize every pixel.

For more on how to build a performance-driven email strategy, see The Ultimate Guide to Smart Banners and The On-Brand Image Premium. And to understand how your grid performance compares to industry peers, download our 2025 email performance benchmark report.