Pennyfill and the problem with official MPG figures

Pennyfill has a “worth the trip” feature. It tells you whether driving to a cheaper station saves you money once you count the fuel used getting there. It only works if it knows how many miles per gallon your car does, and this week I spent my time on where that number comes from.

The problem with asking people for their MPG

Most people have no idea what their MPG is. If the box is empty, the feature has nothing to work with, so a station five miles away looks like a bargain whatever the car. The fix I added is a registration lookup in the garage. Type in your number plate and the app now gets the engine size and year, then matches the car to its official combined MPG from the VCA car fuel data.

One rule I set early: if you have typed in your own MPG, the lookup never overwrites it. Your figure beats a table every time.

Then I noticed the numbers were too kind

The first version used the official figure as it stood, and it made far-away stations look better value than they really are. Official lab figures are optimistic. That is no secret, but I had not allowed for it.

The rough corrections I settled on, from what I read while building this:

• Cars tested under the old NEDC test (the VCA guides run to 2017): official figures flatter them by about a quarter.
• Cars tested under WLTP (2019 on): about a tenth.

So there is a small function, realisticMpg(), that scales the matched figure by the test era. The multipliers are 0.78, 0.84 and 0.9. They are blunt, and I would not claim they are right for any single car. They are meant to be closer than the lab number, and they err on the side of making a long detour look less tempting.

Showing both

I did not want to hide the official figure, so the garage shows both: the official one and the realistic one the app actually uses. If the two look far apart for your car, you can see why and overwrite it with what you really get.

What I took from it

• A default is a decision. Using the lab figure looked neutral, but it quietly pushed people towards longer drives.
• Prefer a blunt correction you can explain to a precise number you cannot.
• Keep the user’s own number in charge.

The other change worth a mention this week was small. For App Store review, the button before the location permission prompt in the app now says “Continue” and explains why the app wants your location. It took ten minutes and it is the kind of thing that decides whether a build goes through.

If you drive a lot and want to try the worth-the-trip numbers on your own car, it is on pennyfill.co.uk.