Carlytics 2026 Mid-Year Transparency Report

TL;DR: We've been running Carlytics for the first half of 2026 with the explicit policy that "we'd rather show a gap than a guess." This post is the qualitative debrief: the categories where our decoder works well, the categories where it doesn't, the refund pattern we've seen, and what we're working on for the second half of 2026. No record counts, no marketing graphs — just the honest review.

Every B2C product owner writes a victory-lap post at the six-month mark. This isn't that. This is the post I'd want to read about a service I'm considering paying nine euros to.

The reason it exists is that, in the absence of an honest review from the people running the product, the only signals you have are the ad copy and the Trustpilot reviews. The ad copy is optimized for conversion. The Trustpilot reviews are dominated by either the very happy or the very angry. Neither captures the day-to-day quality of what we actually do.

This is the founder filling in the middle.

What we set out to be

The thesis when we started: European used-car buyers were paying EUR 25–35 for VIN reports that were over-rich for the typical question ("is this car what the listing says it is?"). A leaner report at a third of the price would convert better and serve the buyer better on the question they actually had.

That thesis has roughly held. The conversion rate from free VIN check to paid report has tracked above the industry comparables we have a read on. The customer-question split — what people email us asking — has matched the original intent: triage, not exhaustive history.

What works well

A short list of cases where I am confident the report earns its EUR 8.90.

Mileage rollback detection on German cars. The German TÜV / HU records are dense and consistent enough that, for a German-registered car with at least two TÜV cycles, our odometer comparison catches rollbacks at a high rate. This is the single highest-conversion use case and it has been since launch.

Cross-border specification verification. The free decode catches the case where a Polish listing says "VW Passat 2018 2.0 TDI 190 hp" and the VIN decodes to a 2017 with a 150 hp engine. We see this case daily. The free check defuses it without anyone paying us anything, which is fine — the user remembers and comes back when the next car works.

Recall flag visibility. The EU Safety Gate recall data plus the NHTSA recall data, intersected against the buyer's VIN, surfaces open recalls clearly. For a Hyundai Kona EV in the 2020-2021 battery-fire VIN range, we render the recall, link to the dealer-completion process, and that information is enough to drive a meaningful negotiation. We get email confirmations of this outcome roughly weekly.

EU country breadth. We cover all 27 EU countries plus UK/NO/CH/RS to roughly equivalent depth. Not exhaustive on any one country the way a national specialist is, but no big gaps. A buyer in Greece looking at a Romanian car gets close to the same coverage as a buyer in Poland looking at a German car.

What works less well

This is the part most companies in this space don't publish.

Damage and accident history outside the German lane. The German TÜV records are dense enough that we can construct a reasonable damage signal. Outside Germany, the underlying data is less queryable and our reports are sparser. For an Italian-registered car we know less about prior insurance claims than we do for a German one. We disclose this in the report — fields that we couldn't populate are marked "data not available" — but the disclosure is no substitute for actually having the data. We're closing the gap, but slowly.

Pre-2010 VIN coverage. Our VDS decoder is trained primarily on Finnish, Norwegian, Dutch, and Czech registry data, and the deeper we go before 2010 the more we encounter VDS patterns we don't have authoritative coverage for. A 2008 Renault Mégane gets a thinner decode than a 2018 Mégane. The decoder falls back to make/model from the WMI and leaves the trim and engine variant null. We tell the user this; some refund anyway, and we honor those refunds.

Engine-variant disambiguation for VW Group cars. Volkswagen's 2.0 TDI engines (the EA189, EA288, then various EA288 evo iterations) share large parts of the VDS pattern, and we sometimes return the wrong variant inside the family. The macro-decode (year, model, body, fuel type) is right; the micro-decode (specific engine code) sometimes isn't. Real fix: more Finnish data with engine codes attached. In flight.

Motorcycle coverage. We launched motorcycle VIN check in Q1 2026. It works but the depth is shallow — a motorcycle VIN tells you the make, model family, year, and engine displacement reliably, and not much else, because the motorcycle registry data we have access to is far less rich than the car data. We've been explicit about this on the motorcycle pages and the refund rate on motorcycle reports is roughly double the car rate. We may pull the motorcycle product if it doesn't reach the quality bar by year-end.

The refund pattern

The refund cases break into roughly four categories:

Category A — decoder error. We returned a wrong year, wrong model, or wrong engine variant. The customer's listing was right; we got it wrong. We refund and re-run the VIN by hand to fix the underlying mapping. This is the case I lose sleep over. We track every one. The pipeline that catches them is the same pipeline that should prevent them — when it doesn't, the refund is automatic.

Category B — gap that was expected. We told the customer in the report that a field was "data not available" and they still feel the report wasn't worth EUR 8.90. We refund these too, with educational framing — explaining that the report did the cross-decode and the stolen-vehicle check and the recall match, and that the gap was disclosed before payment. About a third of these customers stay with us; the rest leave.

Category C — misread expectations. The customer expected the report to include a damage history we don't have for that country, or an insurance-claim record that no European service has for that country. This is a UX failure on our part — we should have communicated the scope better upfront. We've been rewriting the sample-report page to make scope more explicit.

Category D — the seller scam wasn't the kind the report would have caught. Customer paid, bought the car anyway, found later that the car had a problem (mechanical, undeclared flood damage, forged document) that the VIN report wasn't ever going to catch. We refund as a goodwill gesture and reference our post on when a VIN check doesn't help so the next customer in the same situation knows up front.

Category A is the one that drives the engineering roadmap. Category B is the one that drives the disclosure language. Category C is the one that drives the marketing. Category D is the one that drives the editorial — posts like this one.

What we're not doing

A short list to be unambiguous:

• We are not training a generative model to write blog posts. Posts like this one are written by humans. We use AI for translation and for drafting brief outlines and that's the limit. The product is supposed to be the trustworthy answer to a small question; we don't want to undercut that with fake content on the marketing side.
• We are not publishing a "Best [X] of 2026" series. We don't have the editorial authority for that lane and faking it would erode the trust we are spending years building.
• We are not buying backlinks. We are building organic reach and a small set of legitimate partner integrations. The pace is slow. The compounding is real.
• We are not gating purchases pre-emptively. If you want to buy a VIN report for a Belarusian car or a 1973 Mercedes or a motorcycle, the system will sell it to you. The disclosure of what we can and can't do is in the report. The refund policy is what catches the edge cases. We'd rather refund 5% than gate the 95% out of a sale they wanted.

What we're working on for H2 2026

Stating the focus areas without committing to dates, because dates miss:

• Engine-variant disambiguation for VW Group 2.0 TDI and 2.0 TFSI families
• Damage-history breadth outside the DE lane, starting with NL/BE
• Pre-2010 VIN coverage for the makes we know we're weakest on (Renault, Peugeot, Citroën, Fiat)
• A redesigned sample-report page that's more honest about per-country depth, so users price-in the scope before they pay
• A revised refund policy that captures Category B and D faster, ideally pre-purchase via clearer disclosure rather than post-purchase via refund

What we are not promising: feature parity with the larger competitors on damage history. That gap is real and closing it costs more than we can spend at our current pricing. We will close it; the question is whether by mid-2027 or mid-2028. Saying anything more precise would be marketing.

What I'd ask of you

If you've used Carlytics: tell us when we got something wrong. The most valuable email I get in any given week is the one that says "VIN ABC123 — you returned a 2017 X1, the car is a 2019 X3." That email goes into the queue and the next person to enter that VIN gets the right answer. Without that signal we don't know which VDS patterns to prioritize, and the decoder improves slower.

If you haven't used Carlytics: try the free check on the VIN of any used car you're considering. The free check is the same code path as the paid report — if the free decode is solid, the paid report's underlying data is solid. If the free decode is patchy, the paid report's data on that VIN will be patchy too. We'd rather have you find out before paying.

The honest closing

We are six months into building a thing that, if we get it right, will be useful for a generation of European used-car buyers. The product is improving. The honest version of "improving" is that some weeks we ship a meaningful upgrade and some weeks we ship a typo fix and a bug report. Trust is built by the long arc, not the launch posts.

Thanks for reading. The next transparency report goes out in November 2026.

Related reading: What is a VIN really? And why your decoder can be wrong · When a VIN check does not help · Carlytics vs carVertical vs autoDNA: founder comparison