There's no magic percentage. The right budget comes from what a realistic gain would earn you in margin and how long you're willing to wait to recoup it. With this budget, you don't buy a gain. You buy a number of attempts.
The benchmark
The math, with made-up figures. At €10M in sales, a 5% increase in revenue per visit is worth €500,000 in revenue. With a 30% contribution margin, that's €150,000 in margin per year. The same logic gives €75,000 at €5M and €300,000 at €20M.
These amounts set a ceiling. An annual budget that pays for itself in under twelve months on this basis is defensible, as long as you accept that the first year won't capture the full gain.
To see this, we modeled a year of optimization on a €1.5M store whose traffic stays flat. The first quarter goes to making measurement reliable and to learning, and it produces no winning test.
After that, out of about thirty tests, three or four win, and the store ends the year with roughly 10% more revenue. That win rate isn't pessimistic: Optimizely, which drew lessons from 127,000 experiments run on its tool, found that only 12% of them significantly improve the primary metric.
The rule
The budget follows the expected margin, adjusted for your odds of getting there, and those odds depend first on how many attempts you can afford. Divide the budget by the full cost of one test, multiply by a win rate of about one in eight, and you'll know how many winners you're buying.
As a rule of thumb, when sizing an engagement, the authors count about forty hours a month to run two to three tests, from mockup to reading the results.
Then review the budget every quarter, in light of what the first projects delivered.
The steps
- The gain. Estimate a conservative increase in revenue per visit over one year, leaving the first quarter at zero.
- The margin. Convert that gain into contribution margin.
- The timeline. Set the payback period you're willing to accept.
- The attempts. Translate the budget into tests, then into likely winners.
- The split. Divide the budget among internal time, service providers and tools, funding time first.
- The line for leadership:
"With this budget, we can run [N] tests this year. If one in eight wins, the margin we recover covers the spend in [M] months, first quarter included."
The trap
Buying the tool before you have the time. A testing solution with nobody to write hypotheses, build variants and read results is just a subscription. And hours disappear faster than you expect.
With a brand the authors work with, the planned budget was around fifty hours a month: one meeting that runs to an hour and a half, one half-hour readout, and 1/10 of the month was gone.
At €5M, most of the budget should therefore go to research and fixes. A continuous testing program pays off when your traffic lets you conclude a test in a few weeks, so run that calculation before you sign up for a tool.
