A UGC Testing Budget That Fits a $200K Ad Spend

Most DTC brands at $200K monthly paid social spend allocate between 15 and 25 percent of budget to creative testing. Rough math: $30K to $50K a month. That sounds like enough. It rarely is, because the money gets spent wrong. Every new UGC concept gets the same $500 test, gets killed or scaled on day four based on whatever CPA landed, and the operator calls that a testing program. It is not a testing program. It is lottery tickets.

This is field notes on a three-tier structure that actually produces readable results. The numbers here are calibrated to $200K monthly paid social spend, a $120-150 AOV, and a $30-40K monthly testing allocation.

What testing is actually for

Testing exists to identify winning concepts with enough confidence that you are willing to put real budget behind them. The output of a test is not "this ad's CPA was $42." The output is "this concept is worth scaling to $2K/day" or "this concept is not worth scaling, here is why."

The most common mistake is treating the test itself as the outcome. Brands spend $500 on a creative, see a CPA of $38 against a benchmark of $45, and call it a winner. Two weeks later the creative has a CPA of $58 at full budget. Nothing was wrong with the test, except the test was testing the wrong thing. $500 of spend at a tight audience produces noisy data. The CPA that landed was inside the confidence interval of the test, but the test did not have enough data to pick a signal out of the noise.

The three-tier structure

Different concepts deserve different budgets. Treating every concept the same wastes money on low-confidence experiments and starves high-confidence iterations.

Tier 1: High confidence. $800-1,200 test budget

Tier 1 is for concepts where you already have strong evidence the pattern works. You are iterating on a proven winner. A creator whose last four ads in this format all beat benchmark. A new hook on a product angle that has produced two hits already. The test is not "will this work at all" but "does this specific iteration work better than the previous iteration."

Budget: $800-1,200 over 48 hours. At this budget, you get enough purchase events to see a real CPA signal against a tight benchmark. Decision rule: if CPA lands within 30 percent of benchmark, promote to main ad set. If it beats benchmark, scale aggressively.

Tier 2: Medium confidence. $400-600 test budget

Tier 2 is for concepts where you have partial evidence. Proven creator, new format. Or new creator with strong organic performance. Or proven format with a new product angle you have not tested before. You expect some of these to fail but your hit rate should be 30-50 percent.

Budget: $400-600 over 4 days. Longer test window because CPA is noisier at this budget level. Decision rule: promote if CPA lands within 50 percent of benchmark. Kill if worse than that. The 50 percent tolerance is because you are paying partly for learning, not just for performance.

Tier 3: Low confidence. $150-250 test budget

Tier 3 is genuine experiments. Unproven creator, unproven format, or both. Your hit rate expectation should be 10-20 percent. Most of these will fail. That is the point of Tier 3: keep the downside small while maintaining exploration.

Budget: $150-250 over 7 days. The longer window matters because you have so little data. Decision rule: promote only if CPA beats benchmark. Anything worse, kill. The bar is higher because the confidence interval at this budget is wide enough that "within benchmark" does not tell you much.

The budget allocation math

At $200K monthly paid social spend with a 15 percent testing allocation ($30K/month), the tiers divide roughly as follows:

• Tier 1: 40 percent of testing budget. $12K/month. Supports 10-15 Tier 1 tests per month.
• Tier 2: 40 percent of testing budget. $12K/month. Supports 20-30 Tier 2 tests per month.
• Tier 3: 20 percent of testing budget. $6K/month. Supports 24-40 Tier 3 tests per month.

That is 50-85 concepts tested per month. Realistic if the creator pipeline and production operation can keep up. If it cannot, the testing program is production-constrained, not budget-constrained, and the fix is upstream.

The testing campaign structure

Do not test creatives by spawning new ad sets inside your prospecting campaign. This fragments the prospecting campaign and creates the learning phase fragmentation problem. Instead, run a dedicated ABO testing campaign with its own budget, its own audience structure (usually one broad prospecting audience), and its own reporting.

Winning concepts graduate into the main prospecting campaign as new creative inside existing ad sets, not as new ad sets. This keeps the prospecting campaign structure stable while allowing creative variety to flow in.

The scoring sheet

Every test gets a scoring sheet. Five data points recorded before the test launches: tier, budget, expected CPA benchmark, decision threshold, decision date. Four data points recorded at decision: actual CPA, thumbstop ratio, save/share count, decision (promote, kill, or extend). One field for qualitative notes.

The scoring sheet is not a bureaucratic artifact. It is the mechanism that lets you look back at 50 tests from the past quarter and identify which creator relationships, which formats, and which hook patterns are producing wins. Without it, you have a vague memory that "that one creator worked well" and no systematic learning.

The Spark Ads wrinkle

Spark Ads introduce their own testing structure. The organic performance of a creator's video is itself signal. A video with strong organic retention and engagement earns Tier 1 treatment straight to a higher budget. A video with mixed organic signals drops to Tier 2. A video with weak organic signals probably does not get Sparked at all, because weak organic does not improve when boosted.

This means the Spark Ads pipeline has implicit tiering that does not map directly onto the generic three-tier structure. Track Spark Ads separately in your scoring sheet so the data does not contaminate your generic UGC benchmarks.

Where the program breaks

Breakdown 1: the operator overrides decision rules. A creator the founder likes personally keeps getting extended tests because "it might still work." The decision rule is violated for relationship reasons. Results: budget drain, distorted benchmarks, no learning.

Breakdown 2: the tier assignment is sloppy. Every concept gets Tier 2 because that feels safest. The actual confidence levels are ignored. Results: Tier 1 winners never get enough budget to scale properly, Tier 3 experiments get more budget than they deserve.

Breakdown 3: the scoring sheet is not maintained. After three months, the sheet is out of date. Nobody can reconstruct which tests were Tier 1 and which were Tier 3. The quarterly review becomes vibes. Results: no systematic learning, same mistakes repeated.

The program is only as useful as the discipline to run it. If the team will not keep a spreadsheet honest, reduce the complexity. Run two tiers. Run one tier. Better a simple program run honestly than a three-tier structure nobody maintains.

Related reading

This field note is part of the paid social for DTC operators hub. The upstream piece on why you cannot read test results in a fragmented ad account is the learning phase explainer. The downstream piece on detecting when a winner is fatiguing is creative fatigue signals. For the CAPI and measurement plumbing that makes CPA benchmarks trustworthy, the DTC Stack Audit covers the stack.