Free trial or paid: what your ads should optimize toward.

Every subscription ad account is an instruction. Find me more people who do this. The platform follows the instruction extremely well, which is exactly the problem when the instruction is “start a free trial” and what you meant was “become a customer who pays.”

Trial design decides how far apart those two things are. The closer the trial start is to a real commitment, the safer it is to optimize toward it. The further apart they are, the more the account needs help telling a good trial from a bad one.

The division that decides it

Here is the same $20,000 of spend under two trial designs, as arithmetic rather than any client’s result.

The ad account, left to itself, prefers the first column. Its trials cost a third as much. The business prefers the second column, because it bought a hundred more paying subscribers for the same money and each one cost about $15 less. The two designs are not a cheap option and an expensive one. They are an expensive option that looks cheap, and a cheaper one that looks expensive.

This is not an argument that card-required trials always win. Some products need a low-friction trial to be understood at all, and the numbers can come out the other way. It is an argument that the comparison has to be made on paying subscribers, and that the ad account cannot make it on its own.

The events you can optimize toward

The tension runs down the table. Earlier events are plentiful and fast but loosely tied to money. Later events are the money but arrive late and in small numbers. The practical answer is usually to optimize toward the trial start or an activation step, and make the value carry the information the event itself lacks.

Attach a value, and make it honest

A trial start is worth its chance of converting times what a paying subscriber is worth. If trials convert at 30% and a subscriber brings about $75 of contribution in the first year, a trial is worth about $22.50. Sending that value turns a trial count into a signal the bidding system can weigh.

The value gets more useful when it varies. Trials that choose the annual plan, trials that complete onboarding, and trials from markets with higher prices are not worth the same. If your history shows how each group converts, the value can reflect it, and value-based bidding will lean toward the trials most likely to pay. Keep the rules simple enough to explain and stable enough to learn from.

Mind the window

Meta’s longest standard click window is seven days. A fourteen-day trial converts after that window closes, so a purchase event sent at the first charge will often arrive too late to be credited or learned from. That is the main reason to optimize Meta toward an earlier event with a value, rather than toward the first payment.

Google gives you more room. Conversion windows for Google Ads can run far longer, and conversions can be imported after the fact from your own records. That makes it practical to send the trial start with an expected value at the time, then import the actual first payment when it happens, or adjust the reported value when a trial converts, refunds, or cancels.

Trial length is a media decision too

Seven, fourteen, and thirty days are usually treated as a product question. They are also a measurement question, because the trial length sets how long it takes for a paid outcome to exist at all.

A longer trial is not wrong. It means the account has to run on a predicted value for longer, so the value rules matter more and the cohort review has to wait. If the product genuinely needs thirty days, keep them, and invest in the proxy: the in-trial behavior that best predicts who pays.

Apps sold through the stores

When the subscription is bought inside an iOS or Android app, the billing record belongs to the app store, and the path from ad click to payment is harder to observe. Apple’s privacy framework limits and delays what ad platforms receive about installs and in-app events. Most subscription apps rely on a measurement partner to pass trial and purchase events to the platforms, and accept that the signal is thinner and later than it would be on the web.

Some move part of their acquisition to a web checkout they control, where the full chain from click to trial to first payment can be measured and sent back. Whether that is worth it depends on the product and the store rules in each market, but it changes the measurement picture enough that it belongs in the same conversation as trial design.

How to test one design against another

1. Run the two designs as separate flows with separate campaigns, so each has its own events and its own learning, rather than switching the design mid-flight in one campaign.
2. Give them comparable budgets, audiences, and creative, and run them over the same dates, so seasonality affects both equally.
3. Fix the success measure before starting: cost per paying subscriber, plus retention at the first renewal.
4. Wait for the cohorts to mature. That means the trial length, plus time for the first payment to clear, plus the first renewal before reading retention.
5. Read the result in billing, not in the ad accounts, which will each tell you their own version of what happened.

The plumbing, in order

1. Fire the chosen event from the browser or app and from your server, with a shared event ID so the platform can deduplicate the pair.
2. Include hashed email or phone where you have consent to, so the platform can match the event to the person who clicked.
3. Send the expected value with the event, in the account currency, using rules written down somewhere the whole team can see.
4. Import first payments, refunds, and cancellations back to the platforms that accept them, keyed to the original click or user.
5. Check the match weekly: platform-reported trials against your own records, and platform-reported subscriptions against billing.

How to tell it is working

The first sign is usually counterintuitive. Cost per trial rises a little, because the account is no longer buying the cheapest trials it can find. Trial-to-paid rate rises by more. Cost per paying subscriber falls. If you only watch the first number, you will switch the change off before the other two arrive.

Give it at least one full trial length plus a week before judging, and judge on the cohort: the trials that started after the change, followed to paid. Everything else about subscription acquisition, including what you can afford per subscriber, rests on that number being honest. The wider picture is in the guide to paid acquisition for subscriptions.