Hertz: The AI Found the Damage and Lost the Customer

What happened

Hertz began installing AI vehicle scanners in autumn 2024, and announced the UVeye partnership in April 2025 as a maintenance improvement. By that summer renters were reporting automated damage bills arriving within minutes of return, including $195 for a dent made up of $80 of repair and $115 in fees to detect and process it. In August 2025 a House Oversight subcommittee wrote to Hertz noting it appeared to be the only US rental company issuing such charges this way, and Senator Blumenthal followed with ten questions, including how soon a customer could reach a human. The system was more accurate than the staff it replaced, and those staff had been quietly doing a second job nobody had written down.

The case, start to finish

The scanner did not invent the charges. It withdrew a discount nobody had ever named.

A leak that had always been there

Three numbers decide a fleet-rental year: utilization rate, the residual value you get when the car is sold, and damage recovery, meaning the money you get back for damage done while the car was in someone else’s hands. The third is real money, and it had always leaked, because catching it depended on a tired agent walking around a car at the end of a shift.

From fall 2024 Hertz installed AI vehicle scanners at airports, at least six of them by Senator Blumenthal's account, and in April 2025 announced a partnership with UVeye. The framing in that announcement was maintenance: an executive vice president described bringing efficiency and greater accuracy to the maintenance process. The system scans body, glass, tires and undercarriage in seconds and compares before-and-after images.

What follows is a judgment about how that rollout was designed, not about Hertz's business overall, and as of the two congressional letters in August 2025 the matter was still open.

The technology worked, which is the problem

Nobody in this story disputes that the scanner sees what it claims to see. The House Oversight letter opens by granting that the technology has the potential to improve efficiency and strengthen customer trust. Hertz told CBS News that of more than 675,000 rentals scanned, more than 97% showed no billable damage, which is a genuinely low hit rate and the number the company kept repeating.

The mechanism nobody planned for sits in the vendor's own pitch, advertised, per reporting quoted to Congress, as detecting five times more damage than manual checks and producing six times higher total value of damage captured. The first of those is an accuracy claim. The second is not.

Consider what the imprecise walkaround had quietly been doing. A scuff overlooked at the end of a long shift was money the company chose not to chase, over and over, without ever calling that a decision. It was an unpriced goodwill allowance running at scale. Automating detection did not manufacture charges out of nothing. It withdrew a subsidy that had never been named, and from the passenger seat, withdrawing an invisible subsidy is indistinguishable from inventing a new fee.

Three failures, none of them the model

The first was that no person stood between the model's output and the customer's card. The House letter singles Hertz out as apparently the only U.S. car rental company issuing damage assessments without human review, while a competitor running comparable scanners has employees review every flagged item before anybody is charged.

The second was the fee stack. One renter was billed $195 against $80 of actual repair, the remaining $115 being fees to detect and estimate the damage and to process the claim. Once the fee exceeds the repair, the customer is no longer arguing about a dent. Another reported bill was $440 for a minor scuff.

The third was the clock. Appeals ran through a chatbot that could only flag a case for later human review, and one renter was told to expect an email reply in ten business days against a discount that expired in seven. Blumenthal's letter describes reductions offered for payment within one or two days, presumably to encourage a quick settlement. Underneath all three sits a framing gap: the program was announced as maintenance, and customers met a billing machine. On the company's own two figures, 3% of 675,000 scans is on the order of twenty thousand people billed, which is arithmetic on those numbers rather than a figure Hertz published.

Find the discretion before you delete it

The transferable rule is not do not automate. It is that the humans being replaced are usually doing a second job nobody wrote down, and that job is often absorbing edge cases, granting small mercies, and keeping the company's relationship with its customers slightly better than its policies technically require.

So before automating any judgment that moves money out of a customer's account, find the discretion the old process was exercising and decide on purpose whether to keep it. Withdrawing it can be a legitimate choice, but it is a pricing change and it deserves to be announced like one rather than discovered in a charge.

And whatever the model outputs, put a person and a working appeal between it and the card, on a timeline no slower than the discount you are offering. The vendor's metric is damage captured. The metric that predicts whether this works is damage captured and not disputed, and only the human step produces it. And remember who does the describing afterwards. The customers you never charge have no story to tell about any of this. The ones you bill have nothing else.

Timeline

• Fall 2024 First AI vehicle scanners installed. Senator Blumenthal's letter records that “since fall 2024, Hertz has installed AI vehicle scanners at at least six U.S. airports.”
• April 16, 2025 Hertz announces the UVeye partnership, and announces it as MAINTENANCE. EVP Mike Moore: “We are excited to partner with UVeye to bring efficiency and greater accuracy to our maintenance process.” Deployment starts at Hartsfield-Jackson Atlanta.
• June-July 2025 Renters report automated damage bills arriving within minutes of return. One is charged $195 for a dent: $80 of repair plus $115 in fees to “detect and estimate the damage” and to “process” the claim, cut to $130 if paid within one day. Another is charged $440 for a minor scuff.
• August 20, 2025 House Oversight's cybersecurity and IT subcommittee writes to Hertz, noting that Hertz “is apparently the only car rental company in the U.S. that issues damage assessments to customers without human review.”
• August 26, 2025 Senator Blumenthal writes with ten questions, including when in the appeals process a consumer can reach a human, and whether Hertz plans to phase out employee visual inspections entirely.

You're in the owner's chair

Your rental fleet genuinely loses money to damage that tired agents miss at return. A vendor sells you a drive-through scanner that catches five times more of it in seconds, and it works exactly as advertised. How do you roll it out?

• Deploy it, bill automatically, but soften the blow with a prompt-payment discount
• Deploy it, but route flagged charges through a human appeal first
• Deploy it and bill automatically — removing the human cost is the entire business case

Where a $195 rental damage bill went

• Actual cost to repair the dent: $80
• Fees to detect, estimate and process the claim: $115

The single claim described in Senator Blumenthal's August 26, 2025 letter to Hertz, citing New York Times reporting. Hertz offered to cut the total to $130 if the renter paid within one day. Once the fees exceed the repair, the customer stops arguing about the dent and starts arguing about you.

Business model

Rent a vehicle by the day from a fleet you own or lease, concentrated at airports where demand concentrates. Three numbers decide a fleet-rental year: utilization, the residual value you get at disposal, and damage recovery, which is what you recoup for the dents that arrive during somebody else's rental. The third one is real money, and it had always leaked.

Revenue model

Time and mileage, plus a stack of ancillaries: insurance products, fuel, upgrades, and damage recovery. Damage recovery is unlike every other line on that list. It is billed to a customer who has already left the lot and never agreed to the amount. That single structural fact is what makes automating it dangerous in a way that automating, say, fuel pricing is not.

Cost structure

Vehicle depreciation dominates, then fleet operating costs, airport concession fees, and labor. A human walkaround at return costs staff minutes on every one of millions of returns, which is exactly the cost the scanner was bought to remove, and exactly the judgement it removed along with it.

Strategic challenge

The technology worked. That is the problem. UVeye's system, per Hertz's own description, scans body, glass, tires and undercarriage in seconds and compares before-and-after images. And per reporting quoted in Blumenthal's letter, it was advertised as detecting “5X more damage than manual checks” and generating “6X higher total value of damage captured.” The second half of that pitch is not an accuracy claim. Here is the mechanism nobody planned for: the sloppy human walkaround was not merely inefficient, it was an unpriced goodwill allowance. Every scuff a tired agent waved through on a Friday night was a discount the customer never knew they received and the company never booked. Automating detection did not invent charges out of nothing; it withdrew a subsidy that had never been named. Withdrawing an invisible subsidy feels, from the passenger seat, precisely like inventing a new fee.

Key decision

Two decisions, and the second is the fatal one. First: deploy the scanner. Second: bill the result automatically, with no person standing between the model's output and the customer's card. Both congressional letters land on the second.

What worked

The detection, and the base rate. Neither letter disputes that the system sees what it claims to see; the House letter opens by granting that the technology “has the potential to enhance operational efficiency and strengthen customer trust.” Hertz told CBS News that of more than 675,000 rentals scanned, more than 97% showed no billable damage. That is a genuinely low hit rate, and it was the number Hertz kept repeating.

What failed

Three things, none of them the model. FIRST, no human in the loop: a competitor running comparable scanners has employees review every AI-flagged damage before a renter is charged anything, and the House letter singles Hertz out as the exception. SECOND, the fee stack: a renter billed $195 against $80 of actual repair is not being charged for a dent, they are being charged for the detection of the dent, and once the fee exceeds the repair the argument stops being about the car. THIRD, the clock. Appeals ran through a chatbot that could only flag a case for later human review; one renter was told to expect an email response in ten business days, against a discount that expired in seven. Blumenthal's letter names the pattern directly, with reductions offered if renters pay within one or two days, “presumably to encourage a quick settlement.” Underneath all three sits a framing gap: Hertz announced a maintenance and vehicle-availability program. Customers met a billing machine. Nobody was told the rules had changed until the charge arrived.

Risk factors

Automating a decision that moves money out of a customer's account without their agreement; buying on a vendor metric (value captured) that is not the metric you actually need (disputes avoided); deleting human judgement that was silently absorbing edge cases; a fee structure where the process costs more than the repair; and legislative attention arriving faster than the deployment schedule: six airports to two congressional letters in under a year.

Lesson summary

An automation can be more accurate than the humans it replaces and still destroy value, because those humans were quietly doing a second job nobody wrote down. Before automating any judgement that costs a customer money: find the discretion the old process was exercising, decide ON PURPOSE whether to keep it, and never let a model's output reach a customer's card without a person and a working appeal in between. The 3% you bill are the only 3% who will ever describe your program in public.

Key data

• 675,000+ Rentals scanned (Hertz-reported, to mid-2025)
• more than 97% Share with no billable damage
• ~20,000 — our arithmetic on those two figures, not a Hertz number Implied customers billed
• $195 = $80 repair + $115 fees One reported bill
• $440 for a minor scuff Another reported bill
• “6X higher total value of damage captured” Vendor claim, per reporting cited to Congress
• 7-day discount vs. 10-business-day email reply Appeal timing

Sources & basis

The company here is real and named, and nothing about it was invented to make the story land. The list below is where each fact came from — public filings, court records, published reporting — so you can open a source and check it against the sentence that used it.

1. Hertz newsroom, April 16, 2025: Hertz and UVeye Partner to Modernize Vehicle Maintenance with AI Technology — the company's own framing of the program and the Mike Moore quote View source ↗
2. Letter from Chairwoman Nancy Mace, House Oversight Subcommittee on Cybersecurity, Information Technology, and Government Innovation, to Hertz CEO Gil West, August 20, 2025 — source for the six-location/100-location figures and the “only car rental company... without human review” finding View source ↗
3. Letter from Senator Richard Blumenthal to Hertz CEO Gil West, August 26, 2025 — source for the $195 and $440 claims, the fee breakdown, the prompt-payment discounts, the chatbot appeal path and the quoted UVeye marketing claims View source ↗
4. CBS News, Behind the controversial AI tech used to inspect rental vehicles for damages — source for Hertz's 675,000+ scans, the 97% no-billable-damage figure and the ten-airport deployment View source ↗