Most people arrive at a legal-rights product unsure whether they have a case at all. They are not shopping, they are checking. If the first thing you show them is a price, you are asking them to pay to find out whether they should have come.
So on BiroClick the split runs along the line where certainty appears. The analysis is free: you enter the details, and the system tells you whether there is a claim and how strong it looks. Only after that does the send-ready document cost money, between 9.90 and 29.90 shekels depending on the claim type.
Putting the line in the right place
The usual instinct is to draw the free line where your costs are, giving away the cheap part and charging for the expensive one. That is the wrong axis. Draw it where the customer's doubt is, and charge on the far side of it.
Here the customer's doubt is do I even have a claim, so that question has to be answered before any price appears, even though answering it is the expensive half to run. On Hero of You the doubt is will this actually look like my kid, so the preview is free and the parent approves the likeness before paying. Different product, same rule: free ends where uncertainty ends.
What the filter actually does
- It removes the worst outcome, which is a person paying and then learning they had no claim. That customer does not come back, does not recommend you, and tells other people why.
- It moves the sales argument from a promise to a result. By the time the price appears, the person has already seen their own case assessed. You are not asking them to believe you.
- It makes the paid thing concrete. They are not buying access to a tool, they are buying a specific document about their specific problem.
- It sets the price expectation honestly. Someone who has seen a real assessment has a much better sense of what the document is worth than someone reading a landing page.
The free tier is where you disqualify people. If it never disqualifies anyone, it is not a filter, it is a demo.
Where it costs you
This is not free to run. The analysis is the expensive half, because it is the part that reads the input and reasons about it, and you pay for every visitor including the ones the filter correctly turns away. That cost is only worth carrying if the paid step converts well enough on the people who pass.
Which means the filter has to be honest in both directions. A filter that waves everyone through to protect the funnel is worse than no filter at all: you have now paid for the compute and still sold somebody a document they cannot use, so you have bought yourself a refund and a bad review at full price.
It also has to be honest the other way. Turning away a marginal case that would have been worth filing is a real loss, and it is invisible in your numbers, because the person who was wrongly told they had nothing simply leaves. Free tiers fail quietly in that direction and nobody notices, because the metric that moves is one you are not looking at.
The economics, plainly
The arithmetic is not complicated but it is unforgiving. Take the cost of one analysis, divide by the share of people who pass and then pay, and that is what you are actually paying to acquire a customer, before any marketing at all. If that number is close to the price of the document, the product does not work, and no amount of funnel optimisation will fix a structural gap.
Two levers move it. Make the analysis cheaper to run without making it worse, which is mostly a prompting and context problem. Or qualify earlier and more cheaply, so obviously hopeless inputs never reach the expensive step. The second is usually the bigger win and it is the one people skip, because it feels like adding friction, and it is: friction aimed at exactly the people you do not want.
When a free tier is the wrong shape
If your product's value is obvious before use, a free tier gives away margin for nothing. Nobody needs a trial to understand what a domain name does. Filters earn their cost specifically where the customer cannot tell in advance whether the thing applies to them.
And if the free tier is a smaller version of the paid one rather than a different question, it is not a filter. It is a sample, and samples train people to wait for more sample. The distinction is whether the free step ends with a verdict or with an appetite.