Insights

Detecting damage before it becomes a mystery

Carsharing removes the handover – and with it the moment when a vehicle's condition is recorded. How AI-based damage detection with carvaloo closes that gap.

Roland Kainbacher – Porträt Roland Kainbacher 7 min
Aufprall am Heck, von der Erkennung hervorgehoben (carvaloo)

A scratch on the bumper, a dent in the door: in carsharing, damage turns up that nobody reported. Who caused it can rarely be established afterwards – several rentals often lie between the incident and its discovery. The cost stays with the operator.

Why damage is so hard to attribute

In classic rental there is a handover: someone walks around the vehicle with the customer and ticks the boxes. Carsharing removes exactly that step – which is the whole point. But it also removes the moment when the condition is recorded.

The usual workarounds all share the same flaw: photos before and after the trip get skipped by many users or are useless in the dark. Visual checks by your own staff cost time and only happen at stations. And self-reporting requires someone to notice damage – and to admit it.

How detection with carvaloo works

carvaloo is a German company that solves damage detection through motion data. A small sensor in the vehicle measures acceleration. An impact, a kerb strike or a bump produces a characteristic pattern – different from a slamming door or a speed bump.

A trained model tells these patterns apart and reports a damage event with time, location and estimated severity. Because the time is known, the event can be linked to the ongoing booking – and thus to the person driving at that moment.

carvaloo states that this detects around 95 percent of damages and assigns them to the causer, while reducing manual visual inspections by more than 90 percent.

What this means in operation

  • Attribution instead of guesswork. You know during which booking the damage occurred, instead of discovering it weeks later.
  • Fewer inspection trips. Your team drives to vehicles where something was actually reported.
  • Solid ground for follow-up. Time, location and event data are documented – helpful both with customers and with insurers.
  • Fairness towards honest users. Those who caused no damage do not pay for it through higher rates.

How it works together with Zemtu

Zemtu has integrated carvaloo's damage detection. A small sensor box in the vehicle measures impacts and vibrations. When the model recognises a damage event, it reports the time, the place and an estimate of its severity – without anyone having to file a report.

Because the time is known, the report lands with the booking it falls into. Added to that are the photos from the guided walkaround before and after the trip, which carvaloo also evaluates: the image analysis compares the condition with the one after the previous trip and marks what has changed.

Report, evidence and billing therefore stay within the same system. Your team does not look at every trip, only at the cases where something actually stood out.

How this comes together in daily operation – from registration through handover to the report – is described on the page Artificial intelligence in operations.

What the technology does not do

An honest assessment belongs here: a detection rate of 95 percent also means part of the damage still goes unnoticed – especially at very low forces, such as a scratch from someone walking past slowly. The system also does not replace an appraisal: it reports that something happened, not what the repair will cost.

And it creates additional data about driving behaviour. Anyone evaluating motion data should state clearly in the terms of use what is recorded, how long it is kept and what it is used for. For works councils in corporate fleets this is often the decisive point.

Who benefits most

The larger the fleet and the more anonymous the usage, the greater the benefit. With a handful of vehicles in a fixed group you know the people involved and settle damage in conversation. From around twenty vehicles in public operation, unattributed damage adds up to a line item that noticeably affects your calculation.