How to
How to check your vehicle mapping accuracy with a 300-vehicle audit
A practical way to measure how often your RC-to-insurer vehicle mapping picks the right code: what to sample, how to judge it, and what the result costs you.
Ask a motor insurance team how accurate their vehicle mapping is and you usually get one of two answers. "Pretty good", or a story about the last time it went badly wrong. Hardly anyone has a number. That's not laziness. Getting a number takes a bit of method, and nobody writes the method down.
Here's ours. It takes a few days, needs nothing more than a spreadsheet, and works whether you map by hand, with a vendor or with something you built.
1. Pick 300 vehicles from real traffic
Take them from the last month or two of actual quote requests, not from a list someone picked. Recent traffic has the new models in it, and new models are where mappings break first.
Keep the mix true to your business. If most of your quotes are two-wheelers, most of the sample should be too. Don't remove the odd ones. The odd ones are the point.
Send only the vehicle fields: maker and model as written on the RC, fuel, cc, seats, body type, emission norms and vehicle class. You don't need registration numbers, chassis or engine numbers, or anything about the owner, and leaving them out makes the audit a much easier conversation with your security team.
Why 300? It's enough to show the big problems, and small enough that two people can judge it in a day or two. On 300 vehicles a measured rate is good to about 6 points either way. That's plenty to tell a decent mapping from a bad one. To compare two systems that are close, use the side-by-side method in step 3, which picks up smaller differences.
2. Agree what "right" means before anyone looks
This step gets skipped, and then everyone argues later. Write four definitions down first and judge every vehicle and insurer pair against them:
- Exact means the listing a careful person would pick: same model, same variant, same engine size.
- Acceptable means a different code for exactly the same vehicle. Insurer masters list some vehicles more than once, so this comes up more often than you'd think.
- Wrong means a different variant, engine size or model.
- No listing means the insurer doesn't cover this vehicle in its master. When that's true, saying so is the right answer.
Judge per insurer. A vehicle that's right for four insurers and wrong for the fifth still produces one wrong quote.
3. If you're comparing two systems, judge blind
Put both answers side by side, shuffle them, and label them A and B so the judge can't tell which is the current mapping. People favour the answer they're used to, even when they're trying not to.
Have a second person judge the same 50 rows. Where the two of you disagree, the definition is usually the problem. Fix the definition, then re-judge, rather than averaging the scores.
4. Turn the result into rupees and hours
You'll end up with an exact rate, a wrong-pick rate and a no-listing rate. The wrong-pick rate matters most, because a confident wrong code goes straight into a quote and nobody checks it. A missing code at least gets noticed.
Then do the arithmetic for your own volumes. Here it is with made-up rates, purely to show the sum:
| Today | After the change | |
|---|---|---|
| Vehicles quoted a month | 20,000 | 20,000 |
| Insurers quoted per vehicle | 5 | 5 |
| Wrong picks per 100 pairs (from your audit) | 6 | 3 |
| Wrong codes a month | 6,000 | 3,000 |
| Hours to fix them, at 5 minutes each | 500 | 250 |
Swap in your own rates and your own minutes. Then add what doesn't fit in a spreadsheet: the customers who walked away from a bad quote, and the endorsements and refunds that came later.
5. Check how much of the damage gets caught
One more number is worth having. Of the wrong picks, how many did your process flag before a customer saw them? If your mapping gives a confidence level, look at the wrong picks and see where they landed. A system whose errors mostly show up as LOW or "check this" costs you a review. One whose errors look as confident as everything else costs you a customer.
That's how Dorvy Tech is built. Every pick comes with HIGH, MEDIUM or LOW confidence and the closest alternatives, and when nothing fits it says "no listing" instead of guessing. On our own held-out test of 150 real RC records across five insurer masters, it picked the exact listing 77.0% of the time against 66.2% for the best standard matcher, and it was right 84.9% of the time when it gave an answer. You can read how we measured it.
Or let us run it for you
If you'd rather not do the work, send us the 300 vehicles with the code you use today for each insurer. Within 3 working days we send back our pick, its confidence and the alternatives next to your code, and your team judges which is right. The comparison is free, and you keep the spreadsheet either way.
