A conversion finding without a price is an opinion. Here’s the arithmetic that turns one into a decision, using a worked example with hypothetical numbers.
The worked example
Say a store’s mobile product pages get 12,000 sessions a month. Desktop visitors add to cart at 6%, mobile at 3.4%. Session recordings show why: the variant selector misfires on a meaningful share of mobile taps.
The sizing question is: if the selector worked, how much of that gap closes?
Not all of it: mobile always converts below desktop. Be conservative and attribute a third of the gap to the broken selector: 0.9 percentage points. That’s roughly 108 extra add-to-carts a month (12,000 × 0.9%). Apply the store’s cart-to-order rate (say 40%) and average order value (say $70):
108 × 0.40 × $70 ≈ $3,000/month at stake. A fix that recovers even a third of that is worth ~$1,000/month.
Now “fix the variant selector” can be ranked against everything else on the roadmap, and it probably wins.
The honesty rules
Sizing goes wrong in predictable ways. Four rules keep it trustworthy:
- Show the arithmetic. A number without its derivation is marketing. Anyone should be able to challenge each step.
- Attribute conservatively. Never assign the whole gap between two segments to one cause. A third is often defensible; the whole gap almost never is.
- Respect small samples. If a page gets 300 sessions a month, a week of data proves nothing. Say “we don’t have enough data to size this yet” and rank it below what you can size.
- Split before you size. Store-wide averages hide the finding. Split by device first, then by traffic source.
Close enough to rank the roadmap
The point of $/month sizing is to make conversion work comparable to everything else competing for developer time. An estimate that’s honest about its assumptions and off by 30% still ranks the roadmap correctly. A recommendation with no number at all doesn’t get ranked, so it gets postponed. This is how every insight in HI Engine arrives: an estimated $/month impact, the arithmetic behind it and a confidence level. Steal the method even if you never use the product.