Sector: French e-commerce · Premium houseplants
“For weeks, I had a hunch. With Korvus, I had a number.”
Key figures
1 in 10 visitors
of the configurator affected
0 alerts
from the standard stack
€4,000 to €8,000
in lost revenue over the period
The hunch
By Charlot is a French online retailer of premium houseplants. Customers can write a few words and have them printed on the pot, and that option converts best.
“The message on the label is why people choose us. When it stops working, our best-selling feature stops selling.”
In late April 2026, Charles senses that something is holding back sales on these pages. No report or alert flags an anomaly, and he has nothing to prove it.
What visitors went through
When personalizing their pot, some visitors typed their message, then waited for the label preview. It never appeared. There was no way to check the result before ordering.
The configurator made three successive calls to a third-party label generation service:
- File upload: 200 OK
- Render key generation: 200 OK
- Preview display: HTTP 500 (failure)
The rendering engine crashed at the third step: a clean server crash, undetectable by the tools in place for weeks.
The bug stayed invisible anyway. Charles had suspected a problem since late April without being able to pin it down, because three technical blind spots reinforce each other.
Why no one saw it
The problem happened at an external vendor, at the very last moment. From the site's point of view, nothing was wrong. The errors got lost among tens of thousands of monthly visits.
When the tech team tried to reproduce the bug, the preview link had already expired. They tested, everything worked, case closed.
No one had been careless. The problem was simply outside the field of view of standard tools.
Blind spot 1: the image that says nothing
When a browser loads an <img> tag without a crossorigin attribute and the server responds with a 500, here is what the page's JavaScript receives: a binary error event. It carries no HTTP status and no response body, nothing that helps.
On the front end, it is as if the image never existed. Standard analytics tools capture nothing from this category of error. The visitor sees a broken page, and by default the site has no way of knowing it is broken.
Blind spot 2: volume under the radar
The HTTP 500 errors did show up in the server logs. But buried in tens of thousands of monthly sessions, they never reached the alert threshold of a standard monitoring probe, which made them statistically invisible by design.
Blind spot 3: the key that expires
When the tech team is asked to reproduce the bug after the fact, the mechanism works against them: the preview key expires after fifteen minutes. An hour after the visitor hit the error, the exact URL of the incident is no longer valid. The bug can no longer be reproduced.
The team tests on a fresh journey, everything works, and the case is closed. The evidence vanished before anyone could look at it.
What Korvus saw from day one
Korvus, installed on May 21, 2026, adds an observability layer directly in the visitor's browser to continuously capture the events that standard telemetry misses.
In 16 days, Korvus measured three things. First, 1 in 10 configurator visitors was affected, or 5.3% of the site's total traffic. Second, errors spiked on May 27, concentrated on a single day. Third, 34% of affected visits showed signs of frustration, versus 21% for the others.
Every error is recorded the moment it happens, with the exact link to the page involved. That is precisely what was missing to reproduce it.
Within minutes, Charles's hunch becomes a quantified incident: time period, scale, profile, post-error behavior.
Estimated lost revenue
- €4,000 to €8,000 over the documented period (May 21 to June 6)
- €30,000 to €60,000/year projected over 12 months under normal conditions
The striking figure is the signal-to-noise ratio: in the middle of the sales season, 1 in 10 configurator visitors saw a broken image, and not a single tool in the tech stack caught it. Korvus had documented it on day 0.
The proof that unblocks everything
On June 5, Charles and his technical vendor receive a complete case file. It contains links that reproduce the error live, figures over 16 days and the behavior of affected visitors.
The vendor opens the links, sees the error and identifies the cause. The fix goes live overnight June 6-7.
Incident timeline
- May 21: Korvus installed, first alert the same day
- May 27: error spike on a single day
- June 5: proof sent to Charles and his vendor
- Night of June 6-7: fix goes live
- June 7: no visitors affected anymore
48 hours passed between delivering the proof and resolving the bug.
The takeaway
By Charlot is not an isolated case. Most e-commerce sites rely on several vendors and external services. When one of them fails, the visitor sees it and standard tools see nothing.
For a CMO, that means an acquisition budget sending visitors to a broken page, with nothing surfacing.
With a guardian placed on the visitor's side, you can prove what you sense, put a figure on it and decide what to fix first.