DEV Community

Dan Liang
Dan Liang

Posted on

How Much Should You Offer in a Free Trial?

Users get free AI generations. The founder pays for every one. The first sale comes in, but the numbers still show a loss. Tighten the free allowance, though, and there's another worry: customers might run out before they've had a chance to properly try the product.

What if losing those few attempts means they can't tell whether the product will work for their business?

Stone_Three described just this situation: his product had made its first sale but was still losing money. He was covering the cost of users' free generations and had already started limiting trial usage.

Whether customers still have enough room to reach a decision after those limits tighten depends on what they were using those attempts to do.

Take voiceovers, for example. A customer pastes in a script, chooses a voice, adjusts the settings, generates the audio and listens. The audio is ready, but hearing it once may not be enough to tell whether it'll work. They may need to generate a few more versions and compare the results. If their allowance runs out during that comparison, they may still not know whether the voiceover is usable in their own content.

ElevenLabs' official documentation says the voice and model significantly affect the output, and the same settings don't guarantee the same result every time. If users want a livelier, more dramatic performance, the documentation recommends lowering stability and generating several versions to choose from. In that situation, another generation may still help the customer judge the result.

But not every retry means the customer hasn't had enough of a chance to decide. The first result may already have been enough to make a judgment, with later generations simply producing more content. Or they may already have heard enough to recognize a poor fit, and a few more attempts won't necessarily make that any clearer.

Running out of allowance while still comparing results is different from continuing to use the product for free after making a decision. Whether a few more attempts help depends on whether they let the customer finish a comparison that's still incomplete.

Counting generations alone isn't enough to work out the cost. One generation doesn't necessarily consume the same resources as another. The model and the workload both affect resource use. Even after new generation stops, files already created may still incur costs: keeping them takes storage, and accessing or downloading them involves data transfer.

If you only look at how much the call count has dropped, you can miss both the differences in resource use per call and the costs that continue after generation.

In my work reviewing stalled businesses, I look at the money spent on a free trial alongside the decision it's meant to help customers make. Does that spending give them enough of a chance to find out whether the product works for what they need? Can the operator afford to provide that experience? If the part of the experience that helps customers decide is already more than the operator can afford, the scope of what's offered for free needs to be reconsidered.

A first sale confirms that someone is willing to pay, but that sale alone can't tell you how far the free trial should go. A lower bill confirms that spending has fallen. Getting through the workflow confirms that the steps have been completed. Once the budget and usage limits are set, you still need to come back to this: does the revised free trial give customers enough of a chance to decide whether the product fits their needs?

Top comments (0)