The Intelligence Behind Every Sponsored Placement

In retail media, autobidding systems estimate the value of every advertising opportunity before an auction runs. They combine predictions such as click-through rate (CTR), conversion rate (CVR), and expected order value (EOV) to understand how each placement is likely to perform. That intelligence helps determine how a retail media platform competes for each placement — not the bid alone.

Editor's note: This is the second article in our Inside Modern Commerce series. Part 1 explored why modern autobidding systems optimize under budget and ROAS constraints. This article explores how they estimate the value of every advertising opportunity before an auction begins.

When people think about retail media auctions, they usually picture brands competing with bids. The advertiser willing to pay the most wins the placement, and the rest is simply auction mechanics.

That mental model isn't wrong, but it's incomplete.

Modern commerce platforms don't simply compare bids. Before an auction can determine a winner, the platform first has to answer a more fundamental question:

How valuable is this advertising opportunity likely to be?

That answer can't be observed directly. It has to be predicted. Every sponsored placement shown to a shopper is ultimately the result of a prediction.

Why bids alone don't decide retail media auctions

Imagine two products competing for the same sponsored placement.

One brand bids $2.00 per click. Another bids only $1.50.

At first glance, the decision seems obvious. But what if the first product is unlikely to be purchased, while the second consistently converts into high-value orders? Comparing bids alone is no longer enough.

Before deciding how aggressively to compete for an opportunity, a retail media platform first needs to estimate the value that opportunity is expected to generate. The bid is only one input into that decision.

In modern commerce, auctions aren't driven by bids alone. They're driven by the platform's ability to estimate value.

What does an autobidding system actually predict?

At the heart of every autobidding engine is a surprisingly simple question:

How much value is this advertising opportunity expected to create?

Answering that question requires more than looking at historical performance. Before making a bidding decision, the system needs to estimate how this sponsored placement is likely to perform for this shopper, in this context.

That estimate is built from multiple predictions: 

Click-through rate (CTR) estimates how likely a shopper is to engage with the sponsored placement.

Conversion rate (CVR) estimates the likelihood that engagement ultimately leads to a purchase.

Expected order value (EOV) estimates how much value that purchase is likely to generate if it happens.

No single prediction tells the whole story.  Looking at any one metric in isolation provides only part of the picture. Modern commerce systems combine these predictions to build a richer estimate of an opportunity's expected value. That estimate becomes one of the key inputs that informs how the bidding engine competes in an auction.

Ad value is contextual, not fixed

One of the biggest misconceptions in commerce is that a product has a fixed value. It doesn't.

The value of an advertising opportunity depends on context.

The same sponsored placement can produce very different outcomes depending on where it appears, who sees it, and when it appears. Placement, shopper location, audience, product category, seasonality, and countless other signals all influence both the likelihood of a purchase and the value that purchase is expected to generate. A product that performs exceptionally well in one context may perform very differently in another.

That's why retail media platforms don't rely on static assumptions or historical averages alone. They continuously learn from changing marketplace signals, updating their estimates as shopper behavior and marketplace conditions evolve.

In modern commerce, value isn't an intrinsic property of a product. It's something the platform must estimate, moment by moment, for every opportunity.

Better predictions lead to better markets

Every auction is only as good as the information behind it.

If a platform cannot accurately estimate the value of an opportunity, even a well-designed second-price auction will struggle to allocate budget effectively. The opposite is also true.

As prediction quality improves, bidding decisions become more informed. Advertisers spend budgets more efficiently. Retailers monetize inventory more effectively. Shoppers are more likely to see sponsored products that are genuinely relevant to what they're trying to accomplish.

Commerce is becoming an intelligence problem

Prediction has always mattered in commerce. What's changing is the number of decisions that depend on it.

Search is no longer the only place where products are discovered. Recommendations, merchandising, sponsored placements, and emerging AI-powered shopping experiences all require platforms to evaluate opportunities in real time, often with far less explicit intent than a search query provides.

As the number of decisions grows, so does the importance of estimating value accurately.

Commerce media platforms are no longer simply matching advertisers with inventory. They're continuously interpreting context, estimating outcomes, and deciding where attention and budget are most likely to create value.

The question is no longer simply, "Who bid the highest?" It's:

"Given everything we know about this shopper, this product, and this moment, what is the best decision we can make?"

As commerce becomes more dynamic, the ability to make intelligent decisions at scale is becoming one of the defining capabilities of modern commerce systems.

Want to see value-based autobidding in action?

Topsort's bidding engine handles CVR and EOV prediction out of the box, so your advertisers get performance without managing bids manually. Book a demo or explore our auctions API documentation to see how it works under the hood.

Frequently asked questions

What is autobidding in retail media? Autobidding is an automated system that sets bids for sponsored placements on behalf of advertisers. Instead of advertisers manually choosing a bid for every keyword or placement, the autobidding engine estimates the expected value of each opportunity and bids accordingly, while respecting the advertiser's budget and ROAS constraints.

How do retail media platforms estimate the value of an ad? Platforms combine two machine-learned predictions: conversion rate (CVR), the probability that a sponsored listing leads to a purchase, and expected order value (EOV), the revenue that purchase is likely to generate. Expected value = CVR × EOV, adjusted continuously for context like placement, audience, and seasonality.

What is the difference between CVR and EOV? CVR measures how likely a purchase is; EOV measures how much that purchase is worth. A product can have a high CVR with small orders, or a low CVR with large orders — a bidding engine needs both to compare opportunities fairly.

Why don't the highest bids always win retail media auctions? Because modern auctions rank ads by expected value, not raw bids. A lower bid attached to a product with a much higher predicted conversion rate can outrank a higher bid, which keeps sponsored results relevant for shoppers and efficient for advertisers.