A while ago, I wrote a post in which I described an algorithm for turning your amazon sales rank into an absolute number of sales. I have since gotten a bit more data. I won’t reprise my entire methodology here; you can check the original post if you are interested. However, the upshot is this:

- You get 1 “point” for every sale.
- Your “score” decays with an exponential timescale of about a day.
- Thus, if you sell at a constant rate, your score is essentially your average number of sales/day.

My fit:

x=A*(rank/C)

^{-(n0+b*rank)}

where x is the score, A=2.4, C=150,000, n0=0.43, b=2e-6.

In graphical format:

To a good approximation a doubling of sales improves your rank by a factor of 4.

This means, for example:

Rank | score |

100 | 58 |

500 | 28 |

2000 | 16 |

10000 | 8 |

50000 | 4 |

There are some caveats:

- These are, of course, only your amazon US sales.
- This relation is only true for amazon US. Other stores (Canada, UK) presumably use the same algorithm, but since they produce fewer sales, the actual numbers in the fit are going to be different.
- Amazon makes fewer sales at night, and therefore there is a 24 periodicity induced. At the same score, your rank will be slightly better at night, slightly worse during the day.
- This does
**not**not include long term weighting. Amazon uses a secondary (and perhaps tertiary) weighting to reflect long term sales. A book that did very well last month but hasn’t sold since won’t actually have a score of zero. However, unless you sold a million books a year ago, but nothing now, this probably won’t affect you.

If anybody has any thoughts (or contradictory ideas), please let me know.

**-Dave**

Oh, and there are a couple of other caveats I forgot to mention:

1) This is only true now. As the number of books on amazon increases and overall sales rates change, so to will the parameterization.

2) I wouldn’t expect this relation to hold into, say, the top 20.

3) This doesn’t include kindle.