Paid acquisition: what app code reveals

One purchase, five signals

What does an ad network learn from a single sale?

For a single sale, Dicero, published by Habby, sends several purchase events to its trackers, including variants of purchase_m_1, fire_purchase_low_1, fire_purchase_high_1 and fire_purchase_change_1.

Each network and campaign optimizes for something different, which is why one revenue becomes five signals. A low-value purchase signal optimizes for volume, a high-value signal for margin. A change signal separates a first purchase from a reactivation.

The network never sees your P&L. It sees the version of your economics you chose to show it.

For a smaller app, define an event taxonomy that matches your decisions: install, activation, first payment, recurring payment, uninstall. Send those events server-side, so they depend neither on the SDK nor on attribution's good mood.

Value is predicted before the purchase

Can you know what a user is worth before they pay?

Block Blast estimates what a player will be worth before the purchase, from phone model, country, ad history and settings such as vibration turned off. The model runs locally in XGBoost, then drives difficulty, rewards and offers.

That is acquisition in the broad sense: the app spends attention and difficulty to maximize the value of an audience it already has. Media buying follows the same logic. If a user from a given channel, device and country is worth three times another, you also know how much you can pay to acquire them.

Subway Surfers goes further into inventory. A 158-rule engine decides which ad placement to show, when and to which profile, and logs every impression with network, revenue and test version. The team can compare the revenue of a rewarded ad against an interstitial for the same user, then move inventory toward what pays.

The same data ranks and prices

Does the data that sorts users also set their price?

Tinder shows the most explicit crossover. In Charles Liu's second Tinder teardown, the app sends device information to its server (model, memory, connection speed), and those signals feed ranking and personalization. The same teardown shows a price set by algorithm from account activity, device and location.

What the user sees and what the user pays come from the same data. Every ad marketplace works this way: the audience is segmented, and the price follows the segment.

The practical question for a founder is which signals you are allowed to collect, and which actually predict anything. Phone model and country both stand in for purchasing power; connection speed does so less obviously. Before adding a signal, check that it improves prediction without creating a compliance problem.

Creative: one format, not fifty

How many creative formats do you need to test?

The teardowns cover the tracking side of acquisition. Creative production is the rest.

In Charles Liu's playbook, a founder who grew an app to 200,000 dollars a month describes three steps: create a free account, study the formats already working in the feed, then post videos of your app in those formats. Most of them fail, and the goal is to find the one format that works and repeat it.

Liu also gives a grid for judging a UGC creator before signing: viral potential, repeatability and ability to convert. Conversion is the criterion that pays.

Followers	Rate per 30-second video
10,000 to 100,000	Up to about $1,000
100,000 to 600,000	$300 to $5,000
Several million	$3,000 to $25,000
Above 10 million	$20,000 to $250,000

Treat these ranges as orders of magnitude: a UGC program costs real money. Ask what a user who stays costs, rather than what a video costs.

What to set up before spending

What must be in place before the first dollar of media?

1. A stable event taxonomy. Five to ten events, defined once, sent server-side.
2. Net revenue per cohort. Refunds, fees and support deducted from gross.
3. Acquisition cost per channel and per creative. Without it, you optimize blind.
4. One creative format at a time. Test one variable: hook, demo or social proof.
5. A weekly cadence. One hypothesis, one creative, one result, one decision each week.

Do not fund acquisition before retention is stable. A successful campaign on a product that loses its users buys churn faster.

The checklist

• List the events your networks receive today. Check that they match your economics.
• Move critical events server-side.
• Calculate 30-day value per channel, not first-day revenue.
• Pick one creative format and produce it in series.
• Measure creator conversion, not just reach.
• Cut a channel as soon as its cost exceeds value, without waiting for month end.

Sources

• Ayo Omolewa, Dicero (Habby) teardown: multiple purchase signals, rating prompt after a win — instagram.com/reel/DeQ5k57R21Z
• Ayo Omolewa, Block Blast teardown: ad value predicted from device, country and history — instagram.com/reel/DecfvksvHiW
• Ayo Omolewa, Subway Surfers teardown: 158-rule engine, impression and test logging — instagram.com/reel/DenaJb2xyW-
• Charles Liu, second Tinder teardown: device data, personalized pricing — instagram.com/reel/Demh9pux3vf
• Charles Liu, playbook for a 200,000-dollar-a-month app: distribution, UGC, one format — instagram.com/reel/DdZRxaRRH7p
• Charles Liu, creator rates and UGC criteria — instagram.com/reel/DMgK1yUSmuq and instagram.com/reel/DKiz-hexY6B