Your TikTok app campaign looks worse. What should you change first?

A rising cost per purchase can make a campaign feel like an easy cut. But the same chart can point to very different problems: more expensive traffic, weaker onboarding, delayed purchases, or a reporting change.

If you manage acquisition yourself, sorting those out is part of the job. Changing the budget is quick. Finding the right thing to change takes a closer look.

Before you pause the campaign or rebuild its creative, work through four checks. Each should end with a decision you can explain.

Start with the number that triggered concern. Was it purchases recorded yesterday, or purchases made by people acquired yesterday? Those are different groups.

TikTok documents several reasons its reporting can differ from a mobile measurement partner, or MMP. These include attribution methods, lookback windows, time zones, and whether conversions appear against the date of the ad interaction or the event itself. A discrepancy alone does not establish a tracking fault. TikTok’s reporting guide

Before comparing totals, write down:

• The app, operating system, country, and campaign included
• The event definition, such as first successful payment rather than checkout started
• The reporting time zone and date basis
• Which clicks and views can receive credit, and for how long
• Whether the count represents events or unique people

The difference between events and unique people matters. Ten payments could come from ten new customers or one repeat customer. Only the first supports a claim that you acquired ten paying users.

For iOS, also establish which measurement view you are reading. Apple’s privacy-preserving attribution, through AdAttributionKit and SKAdNetwork, can withhold conversion values from attribution postbacks when privacy thresholds are not met. An unavailable value cannot safely be interpreted as a user doing nothing. Apple’s attribution overview

If the definitions differ, reconcile the reports before making a performance judgment. If the event itself is missing or duplicated, investigate measurement and limit exposure according to your test budget. Waiting for more traffic will not repair a broken event.

A cohort is simply a group of users acquired during a defined period. Comparing cohorts at the same age helps answer a useful question: what had each group done after the same amount of time?

AppsFlyer’s cohort reporting distinguishes partial periods from completed ones. Recent users may not yet have reached the day you are trying to measure. Use the report’s completed periods and check how it defines a cohort day; calendar days and rolling 24-hour periods are not interchangeable. AppsFlyer’s cohort guide

Consider this illustrative example. These are invented numbers, not BerryGood campaign results. Both cohorts use the same event definitions and reporting method:

• An older cohort cost $1,200, brought in 600 installs, and produced 40 unique paying users by its completed day 7. That is $30 per payer.
• A newer cohort also cost $1,200 and brought in 600 installs. It has 20 payers, but has only reached day 2. Its current cost per payer is $60.

It is tempting to call the newer cohort twice as expensive. That comparison gives one group five more days to buy.

The newer cohort might finish day 7 with 25 payers, costing $48 each, or 50 payers, costing $24 each. Neither outcome is a forecast. The point is that the early result cannot settle the comparison.

Choose a review window that fits your product’s purchase cycle. A seven-day trial may need longer than seven days to reveal paid conversion. While waiting, watch earlier signals you have reason to trust, alongside the agreed spending limit. Cohort maturity is a reason to postpone a verdict, not a reason to spend indefinitely.

Once the comparison is fair, break the journey into a few measurable steps:

Install → first useful action → first payment → repeat use or renewal

Use your own product’s definitions. For a language app, the first useful action might be a completed lesson. For a subscription tool, it might be completing the task that prompted the download.

Compare counts as well as rates at each step. Keep the same cohort and observation window throughout, counting each person once per milestone. Then follow the first meaningful change:

• Installs become more expensive, but later conversion rates hold: inspect delivery costs, audience mix, creative, and the store experience. The acquisition cost has changed; the available evidence does not yet show weaker users.
• Install cost holds, but fewer people reach the first useful action: compare the ad’s promise with the actual first session. Check app versions and onboarding changes before blaming traffic quality.
• Activation holds, but payment falls: inspect checkout, pricing, trial timing, and purchase measurement. For subscriptions, separate starting a trial from becoming a paying subscriber.
• First payments hold, but later value falls: examine repeat purchases, renewals, cancellations, or refunds where relevant. Cheap first purchases can still be a poor result for the business.

These are investigation routes, not diagnoses. A campaign may also look worse because it now reaches a different mix of countries or devices. Check those splits before concluding that every part of it deteriorated.

At this point, the next action should be specific.

If reporting is unreliable: repair or reconcile the measurement issue. State which results remain usable and what evidence will confirm the repair.

If the cohort is too young: set the next review for the first comparable completed period. Keep a spending boundary and an earlier stop condition for obvious delivery or product failures.

If one funnel step weakened: test a change aimed at that step. Name the metric that should improve and the downstream outcome that must not deteriorate. Fixing a checkout error is a different job from testing a new opening hook.

If mature results miss your economics: reduce or stop the test according to the limit agreed in advance. There is no universal acceptable acquisition cost; it depends on what a customer is worth to your business and when that value arrives.

Check the campaign’s optimization goal too. TikTok’s App Event Optimization can target actions beyond installs, including Purchase and Subscribe events. That makes the chosen event an important setup decision, but changing it is no guarantee of better economics. TikTok’s AEO overview

Avoid rewriting the whole campaign in response to one bad day. TikTok notes that performance can fluctuate during learning and that some edits affect that process. Record material changes so the next review has context. TikTok’s learning phase guide

Bring four things: a clearly defined outcome, comparable cohorts, the first funnel step that changed, and a decision with a review condition.

That gives you a useful answer to “why are we changing this?” It also makes the work easier to share with a partner: everyone can see the evidence, the uncertainty, and what success would look like next.