Finding a revenue leak across marketing and inventory

The problem

Between 7% and 9% of the retailer’s product-page visits were landing on items that were fully unavailable: 282,000 sessions a week at the peak. The rate had been stable for months, no alarm fired on it, and it was owned by merchandising as a chronic inventory problem.

Meanwhile the metric leadership actually watched, mobile-web add-to-cart rate, was falling, and was being investigated as a page-design problem. The two were never the same conversation. The unavailable-page rate was an inventory metric, the add-to-cart decline was a product metric, and the automated bidding that connected them was a marketing system. Three functions, three dashboards, no shared view, and no anomaly in any single one of them.

How Augmeta helped

The Operator reframed the question, from "why is the unavailable-page rate high" to "why is mobile-web add-to-cart falling," and followed it across traffic sources, product availability and device behaviour.

The chain:

1. Automated bidding bought clicks on items that had just gone out of stock.
2. Those clicks landed on product pages with nothing to buy.
3. Mobile-web visitors recovered from that dead end far less often than desktop visitors did, so the leak surfaced first in the metric leadership was watching.

Most of the affected traffic came from two places: paid search, and the retailer's own category pages, whose availability information lagged the product pages. That second share was the part the retailer controlled directly.

Two of the retailer's own product managers independently reproduced the analysis from their own tools.

The finding we didn't ship

A parallel analysis suggested the same paid channel explained an entire year-over-year decline in desktop cart-to-checkout. Before it went anywhere, a check of the channel’s history found that two paid channels had swapped a large volume of visits between them in a single month: a reporting reclassification, not a real shift in media. The analysis was parked pending validation with the media team rather than delivered. A conclusion that would have been wrong for a taxonomy reason never reached the customer.

What changed

The work split into two tracks.

Track 1: the retailer's own pages. The team shipped fresher inventory information on category pages, so shoppers stop being routed from the retailer's own pages to items they cannot buy. It addressed about 30% of affected sessions, and it was live two days after the investigation started.

Track 2: the paid media feed. The feed driving automated bidding was updated with current inventory, so the bidding system stops buying clicks on items that have just gone out of stock. [status]

A persistent issue that three functions had each half-explained now had one shared explanation and two targeted changes. The conversion and revenue effect was still awaiting measurement when this was written, and it is not presented here as a closed loop. What is demonstrated is the diagnosis, the speed of the first fix, and a stated model of what is recoverable: those sessions converting at the site average instead of at zero.