Pay
Annual totals hide everything interesting
Two people with the same yearly figure can have had completely different years.
Guides on Pay: Pricing, from first principles to the annual review, The ceiling is hours, and it arrives sooner than people expect, What rating work actually pays
The annual total is the least informative number in your records. Two earners who both finish on the same figure can have had a steady year and a chaotic one respectively, and only one of them can plan the next twelve months.
What makes a year good is not the total. It is how predictable the total was, how much of it came from work you want to keep doing, and how much of it you actually kept.
The four readings that matter
Take your monthly figures and produce four things before you look at the sum.
Spread. Highest month against lowest month. A year where the best month is twice the worst is normal. A year where it is six times the worst is a year you survived rather than ran, and the fix is usually pipeline rather than price. Some variation is simply the shape of self-employment: in the Federal Reserve's survey of US households in 2024, 59 percent of self-employed adults said their income varied from month to month, against 28 percent of traditional employees.
Median month, not mean month. One exceptional month drags a mean upward and tells you nothing about a typical one. The median is what you can budget against, and budgeting against the mean is how people end up short in February.
Format mix. Revenue by format, and hours by format, side by side. The two orderings are frequently different, and the format that produced the most revenue is often not the one that produced the most income per hour.
Take-home fraction. Gross booked against what reached your account after commission, processing, currency and refunds. If you have not been tracking take-home per format, this is the year to start, because it is the single number most likely to be worse than you assumed.
A worked illustration
The figures below are an illustration, not data. They exist to show the shape of the reading, and your own numbers will look nothing like them.
| Format | Share of revenue | Share of hours | Revenue per hour (indexed) |
|---|---|---|---|
| Written assessment | 45% | 35% | 129 |
| Recorded video | 35% | 45% | 78 |
| Comparison jobs | 12% | 15% | 80 |
| Rush work | 8% | 5% | 160 |
Read that and the year has already told you three things. Video is the largest time sink and pays worst per hour, which is a pricing problem rather than a reason to drop it. Rush work is the best-paid hour you sell and is 5 per cent of the year, which is a supply problem: you are probably not advertising availability for it. And comparison work is doing neither job well, which matches why comparison jobs price badly almost every time it gets measured.
Seasonality, separated from noise
You need two years before a seasonal pattern is real, and most people conclude one from four data points.
With one year, mark the quiet stretches and treat them as a hypothesis. With two, overlay them and look for months that were quiet both times. Anything quiet in only one of the two years was an event, not a season.
Where the pattern holds, the response is a cash-flow response rather than a marketing one. Budgeting for the quiet season is the mechanics of that, and it is far cheaper than trying to force demand into a month that does not have any.
There is a second-order effect worth naming. Your own availability creates apparent seasonality: if you took two weeks off in August, August looks like a quiet month forever afterwards unless you note why. Records that capture what you did, not only what you earned, are the difference between reading a market and reading your own diary - which is the practical argument in the records post. The general discipline of keeping figures in a form that can be compared across periods rather than admired once is set out in the case for treating your own numbers as a dataset, and a year of monthly totals is exactly the size where that starts to pay.
What to change for next year
Pick two things. An annual review that produces nine changes produces none, because you cannot attribute anything afterwards.
The two with the best track record are a price move on your worst-paying format, and a deliberate expansion of whatever your table shows as underpriced and undersupplied. Both are measurable within a quarter, and both leave the rest of the year stable enough that you can tell which one worked.
Resist the third change, which is always "get more visible". Visibility without a pricing change converts a quiet year into a busy year at the same hourly rate, and the year after that you are working weekends.
Reading it against the outside
Your year happened inside a market, and some of what looks like your performance is the market's. Rates across generic assessment have been under pressure since automated scoring became free and instant - the technical account of what those systems now do explains which parts of your mix are exposed to that and which are not.
Buyer behaviour moves too. What buyers say about the reviews they receive is the closest thing to a demand-side read available, and a year where your specific format lost ground is worth checking against it before you conclude anything about yourself.
Platform-side annual reports, where they exist, are the source for the gross figures you are reconciling against, and their format and timing differ by marketplace - for the one this site sits next to, Rate Cock's judges page is the current reference. Reconcile against the report rather than your memory, then keep both, because the year after next you will want three points on the line and not two.