Users don't hate credits. They hate paying for your retries.
Three launches on the board today price in credits, and the complaint underneath each one is the same.
Solid's reviews say credits burn while fixing errors the tool itself caused. Speechka charges 1 credit per minute of translation mode, so you pay full freight for the call where the other person does most of the talking. GBrain bundles $100 of usage into a $199 seat, which is fat margin or underwater depending on which model the user happened to pick that week.
Those aren't three pricing mistakes. They're one. The meter measures something the vendor can see and the user can't feel.
After building two products on usage pricing, the rules I'd defend: never charge for a retry you caused, show the cost of an action before it runs, and pick a unit the user could count themselves if they had to. You give up revenue on the first rule and get it back because people stop rationing.
Still unsolved for me is routing. The same job can cost 3x depending on which model it lands on, and I haven't found a credit model that survives that without either lying to the user or eating the variance myself.
Replies
Softorino 💻📲
The 'catch' question is one I answer daily. I write copy for a free browser tool with zero credits, and even at $0 people still search for the fee they think is hiding. Have you tried a flat cap instead of metering every action?
@yelyzaveta_kibets A flat cap is what I'd run if our costs were flat. A video render here costs roughly 40x an image, so one cap either prices out the person doing images or funds someone rendering all day, and I haven't found a number that does both. What we've ended up with is a ceiling generous enough that normal use never touches it, and metering only in the one studio where the unit cost is actually big. On the free tool, people hunt for the catch because pricing is the only page where they can check if you're honest, so naming what you do with their files kills the question faster than the word free does.
Softorino 💻📲
@asadmalik901Â Ceiling generous enough nobody hits it, metering only where the unit cost is actually big, beats one number for everything. Naming what happens to the files is the line I'd push onto our own FAQ, since silence is what invites the what's the catch searches. The video-versus-image cost gap alone rules out a flat cap.
I charge one credit per generation in a flashcard generator tool I build. A timeout or a 5xx always refunds, no limit. But an empty answer, or 12 cards when someone asked for 30, might be my fault or might be the input. Ask for 30 cards on a topic that only has 15 real words in it and you'll get an undercount every time. With a plain refund rule that's a free generation loop.
So those refunds are capped at 3 a day per user, and an undercount only refunds below 60% of what was asked. Above that you keep the cards and get an honest warning instead.
@siarheihamanovich Capping the refunds is the part everyone skips, and without it the rule is just a free tier with extra steps. The 60% line is the bit I'd steal, because it draws the boundary at unusable rather than imperfect, and that's the only definition that survives contact with real users. Where it breaks for me is that a generous refund on a video render is a real bill and on an image it's rounding error, so the same rule costs wildly different amounts depending on where it fires. I've kept one rule anyway, because two rules is something nobody can hold in their head.