FAQ

Where do you get used car data from?

Our data comes from publicly listed listings on Facebook Marketplace and multiple Instagram pages.

How can we see the confidence of the car evaluation model?

The confidence of the model post evaluation is displayed within the evaluation report. A confidence percentage denoted by light green stands for high confidence, amber for medium confidence and red for low confidence.

How can we compare two number plates value?

You can compare between two different numbers by using the compare number button within the same page.

How can I compare between two identical cars value?

Identical or non identical cars can be compared against each other by clicking the compare car prices button in the same screen.

Why are number plates expensive in Qatar?

Number plates are transferable in Qatar. A number is linked to a person and not a vehicle, so limited numbers are often seen as an investment.

What number plates are generally expensive?

3, 4, and 5 digit numbers are generally expensive. Lower digit counts are more expensive, and 6 digit numbers can also be expensive when they follow a pattern.

What makes a number plate expensive?

Patterns can make a number more expensive than other numbers with the same digit count. Palindromes, birth year numbers, sequential numbers, repetitive digits, and other locally desired values can all increase value.

What is this tool?

This tool is a number plate estimator, evaluator, and predictor for Qatar number plates. It helps you judge a number before buying it and helps you understand the right value before selling.

How does this tool work?

The tool uses machine learning models trained on historical data to estimate where your number stands and predict its worth.

How is this tool accurate?

It is trained on a large amount of real data gathered from public sources, which helps it make a strong estimate based on past market behavior.

Are number plates a good investment in Qatar?

Number plates generally appreciate over time, and many people consider them a good investment in Qatar.

Are these prices accurate?

You can test the tool yourself by comparing the predicted value against the asking prices of plates listed for sale on one of the many online marketplaces where numbers are offered for sale. This tool was not trained specifically on one marketplace listing, but the accuracy is still strong as a general market estimate.

Some numbers look off. Why does that happen?

A small number of plates will vary by a high margin. In most cases that is because the seller is asking above the market price, or because the plate is so rare that the dataset did not include enough similar patterns.

Can the tool account for patterns like 974?

Numbers like 974 fall into the rare-number warning mentioned in the disclaimer. These plates are usually auction-grade, very expensive, and often do not show up in normal listings. Even when they do appear, the model still needs enough similar examples in the dataset to learn how much weight to give that pattern. In machine learning terms, these are outliers.

How frequently do you retrain?

We retrain frequently to keep the model as accurate as possible as new data comes in.

Does it get more accurate with time? I found 1234 cheaper than 7020.

Yes, the model generally improves as more data becomes available. Cases like 1234 versus 7020 can happen for two reasons: either the training data genuinely suggests that 7020 is more expensive, or the dataset may not yet contain enough sequential-number examples like 1234 or 4567 to fully capture that pattern. We retrain often to reduce these gaps.

Why didn't you add 3 digits?

The dataset for 3-digit plates was not large enough to support a model with confidence. There are only 900 possible 3-digit numbers in total, and only the ones that appear in public data sources are useful for training.

How is the value calculated? Is it rule-based or data-driven?

It is primarily data-driven. Rule-based pricing does not work well for number plates because market behavior is too nuanced. We train models on publicly available data and run each new number through the model so it can pick up patterns that are hard to encode manually.

Are accidents and warranties taken into account by the model?

That's a good question that deserves a nuanced answer. The answer is mostly yes — but also not exactly yes. ML models don't really understand what "warranty" or "accident" means in a literal sense. All they do is build lots and lots of connections and relationships using the parameters we provide during the training phase — make, model, age, mileage, cylinder, trim — so the model can finally predict the price of the next car. Accidents are taken care of indirectly, even though the dataset does not really contain information about the number of accidents on a vehicle. You could say two things about accidents: 1. They can only increase with time, never go backwards (definite). 2. They tend to increase more with time (probable). The patterns in the number of accidents reflect on the prices in the dataset because the average number of accidents on any car driven by an average driver is roughly fixed for a given age and mileage. The only challenge for the model would be outliers — too many major accidents in a short span, or too few minor ones over 10 years — but that's not how the real world usually works. Regarding warranty, the dataset does not contain whether a car is still under warranty, but a steep drop in prices that cars like BMW take once the warranty window ends is captured implicitly through the age/mileage relationship. As long as the model learns the price-vs-age curve and the drop after roughly 5 years, the warranty effect comes through automatically.

What parameters or functions does the model use to estimate a car's value?

There are definite things — like mileage, trim, year, make, model, and cylinder — that we take directly from you in the input fields. But there are also other relationships the model builds out of these inputs that aren't directly visible. For example: the resale value of the car itself (a Toyota Corolla holds value differently than a Range Rover), the popularity and demand for that car in the region (some models trade easier than others), brand-specific depreciation curves, how mileage affects price differently across body styles, and a lot of other connections — some we can interpret, others we can't. The model discovers these patterns implicitly during training, even though we never explicitly tell it "this model is popular" or "that brand depreciates quickly."