Parked 96% of the Time: What AI Changed About Peer-to-Peer Rental

The Arithmetic of Standing Still

Research by John Bates and David Leibling for the RAC Foundation found that the average car in England is parked at home about 80% of the time, parked somewhere else about 16% of the time, and actually moving for roughly 4%. The report dates from 2012 and the proportions have not meaningfully improved since. Whatever else a car is, it is an asset that spends nineteen days out of twenty doing nothing while depreciating, insured, taxed, and taking up space.

Every business that owns equipment recognises this shape. The van used three mornings a week. The trailer that comes out in summer. The boat, the camera rig, the second delivery vehicle bought for a contract that ended. Accountants call it capital tied up in underutilised assets; owners usually call it “the thing on the drive.”

The obvious answer — rent it to someone else when you are not using it — has been obvious for a very long time. It stayed impractical for reasons that had nothing to do with demand and everything to do with cost per transaction. That is the part that changed, and AI is most of the reason.

Why Renting Out Idle Assets Used to Be a Bad Business

Three costs sat between an idle asset and revenue, and all three were fixed human costs charged against a thin per-booking margin.

Screening cost. Handing keys to a stranger requires knowing something about them. Traditionally that meant a person checking a licence, comparing a face to a document, making a judgement call, and being wrong often enough to matter. Ten minutes of skilled attention against a €45 rental is not a business.
Pricing cost. An owner with one vehicle has no idea what it is worth on a wet Tuesday in November versus a Friday in August, or how a three-day lead time should be priced against three weeks. Rental companies employ revenue managers for this. Individuals guessed, and mostly guessed low, which made the whole exercise feel not worth doing.
Dispute cost. The scratch argument. Was it there before? Neither party has evidence, both are certain, and resolving it costs more in time and goodwill than the booking earned. A single bad handover could wipe out a month of income.

Add these up and you get the reason rental remained the business of companies with fleets, airport counters, and legal departments. They could amortise the fixed costs across thousands of transactions. One owner with one car could not.

What AI Actually Changed

None of the shifts below are exotic. They are ordinary applications of well-understood technology, and their combined effect is that the three costs above fell far enough for the arithmetic to invert.

Screening became a computation. Document capture, optical character recognition, face matching against the document photo, liveness checks, and a risk score assembled from history and behaviour now run in the seconds between someone tapping “book” and seeing a confirmation. The accuracy of the underlying face-matching step is independently measured — the US National Institute of Standards and Technology runs an ongoing evaluation of face recognition algorithms that has tracked substantial error-rate reductions across successive generations of submitted systems. The relevant consequence for a marketplace is not that machines are infallible. It is that a check which used to cost ten minutes of a person’s attention now costs a fraction of a cent, so it runs on every booking rather than on the ones that happen to look suspicious.
Pricing stopped being a guess. Demand forecasting on rental inventory is a well-trodden problem: seasonality, local events, day of week, lead time, what comparable assets nearby are charging, how quickly similar listings filled. Platforms now push a suggested price to the owner and adjust it as the date approaches. Owners can override it, and the ones who do usually discover the model was right.
Condition became evidence. Timestamped photo sets at pickup and return, compared automatically, with differences flagged. This sounds minor and is not: it converts the most common dispute from an argument between two confident people into a comparison of two images. Deposits can then be held and released against something, which is what makes one stranger willing to hand over keys and the other willing to leave money on the table.
The administrative tail collapsed. Listing descriptions, translations for foreign visitors, answers to the twenty questions every renter asks before booking, reminders, handover instructions. Individually trivial, collectively the reason most people who tried this in 2015 quietly stopped. Much of it now runs without the owner opening the app — the same pattern we have described for support automation in small businesses, applied to a different queue.

What AI Did Not Solve

An honest accounting matters more here than in most software categories, because the downside is a physical asset in someone else’s hands.

Insurance and liability are legal products, not features. Whether your own policy permits commercial use, what the platform’s cover actually includes, and what happens in a total loss are questions with answers specific to your country, your insurer, and your asset. Read them before the first booking rather than after the first incident.

Regulation is local and moves. Short-term vehicle and vessel rental sits under different rules in different jurisdictions, and some cities have tightened them specifically in response to platform growth. That is a question for your accountant and, in some places, your municipality.

The physical work stays physical. Cleaning, fuel, keys, being reachable, meeting someone at an awkward hour. Software reduces the coordination, not the labour.

Wear and depreciation are routinely underpriced. An asset used by strangers ages faster than one used by you. If your mental model of the income is “free money from something already sitting there,” you are not subtracting the maintenance those extra kilometres will cause.

Does the Arithmetic Work for You?

Before evaluating any platform, run four numbers.

Genuine idle time. Not “I rarely use it” — actual days per month you would be comfortable not having it, with a buffer for the days you thought you would not need it and did.
Realistic day rate, after fees. Take the platform’s commission off, then the cleaning, the fuel, and the drive to the handover point.
Incremental costs. Extra maintenance, higher insurance if applicable, the occasional deep clean, the occasional lost day for a repair.
Your own time, priced honestly. If a booking costs you 45 minutes across messaging, handover, and return, and you value your hour at all, that comes off the top.

Assets with high idle time, high day rates, and low handover friction work. Assets with the opposite profile do not, and no amount of platform quality will fix that. This is a spreadsheet question before it is a technology question.

Where the Demand Actually Is

Rental demand concentrates where visitors do. UN Tourism’s statistics put Spain consistently among the world’s most-visited destinations, and its coastal regions in particular receive visitors who arrive by air, stay a week or two, and need transport a hotel does not provide. That profile — dense short-stay demand, thin public transport to the places people actually want to reach, and a long season — is what makes some markets work for peer-to-peer rental while others never get off the ground. Before committing, check whether your location has that shape.

Choosing a Platform

The category has settled into a few models worth understanding before you pick.

Turo is the largest peer-to-peer car rental marketplace, strongest in North America with a growing European presence, and the most useful reference point for how the model works at scale. Getaround built its product around keyless access — the renter unlocks the car from the app, which removes the handover meeting entirely, at the cost of hardware installed in the vehicle. Both are car-only.

Juuh takes a different cut at the same problem: it focuses on Spain and lists motorbikes and boats alongside cars, which matters in coastal markets where a scooter or a small boat sits idle for most of the year and carries a higher seasonal day rate than a car. Hosts are verified before listing, each booking carries a security deposit, and cancellation terms appear on the listing rather than getting negotiated afterwards. For an owner whose idle asset is not a car, the choice of platform is often decided by that single question of what the marketplace will actually list.

Whichever you evaluate, the test is the same as for any business software: check what the platform does when something goes wrong. Read the damage claim process, the cancellation policy, and the payout schedule before you read the marketing page. Those three documents tell you more about a marketplace than any feature list.

Testing It Without Betting the Asset

A sensible first pass takes about six weeks.

List during a period you genuinely do not need the asset. A fortnight you were going to be away anyway. The goal of the first bookings is information, not income.
Accept the suggested price at first. You will be tempted to price above it. Resist for the first three bookings, so you learn what the market pays before you start theorising about it.
Photograph obsessively at handover. Every panel, the odometer, the fuel gauge, the interior, timestamped through the app so the record lives with the booking rather than on your phone.
Track the real numbers, not the gross. Revenue minus commission, cleaning, fuel, extra maintenance, and your own hours. The same discipline we recommend for measuring returns on any AI-enabled process: if the only figure you track is the headline one, you will conclude it works long after it stopped.
Decide on evidence. After six weeks you will know your real utilisation, your real hourly return, and whether the handover friction is tolerable. One of those three usually settles it.

Six weeks is deliberately short. The cost of calling this wrong is one season of imperfect pricing on one asset, which puts it firmly in the category of decisions worth making quickly rather than perfectly.

The broader pattern is worth noticing beyond vehicles. A category of business that was uneconomic for decades became viable not because demand appeared, but because the per-transaction cost of trust, pricing, and dispute resolution fell far enough for demand that already existed to be served. That is a fair description of what AI has done to several markets, and a reasonable place to look when deciding where it might do it next in yours.