Monthly Creative Tests or One Flagship Ad? A CNY 180,000 Model

Video Production Budget Decisions 04/10

Consider an established home-product brand preparing a 28-day TikTok paid-media cycle in the United States. The question for this month is not whether the team can make one attractive commercial. It is which entry point deserves further investment next month: a problem-led opening, a result-first demonstration, a feature explanation, or a talent-led testimonial. The available cash is CNY 180,000. Samples and the landing page are ready, but the team has no reliable historical creative win rate.

For that situation, spending most of the budget on one flagship film is usually a poor fit. A flagship film can be valuable, but it buys one resolved expression and may leave too little media budget to compare the underlying hypotheses. When the primary job is learning, protect the traffic, duration, and measurement conditions first, then calculate how much production can consume. When the job is a launch moment, a committed placement, or a long-lived brand asset, the flagship route may move to the front.

Every CNY figure below is a bottom-up hypothetical input. It is not a market average, a public supplier quote, or an ONCE rate card. A real buyer should replace each figure with actual quotations and the forecast from the live ad account.

Define a test unit before counting deliverable files

One test unit should answer one question. For example: with the same product, audience, offer, and landing page, does showing the result first outperform describing the problem first for completed purchases? The A and B ads keep the audience, bid, offer, observation period, and product facts stable. Only the opening logic changes. That is one test.

Exporting the same timeline in 9:16, 1:1, and 16:9 does not create three creative hypotheses. Neither does changing a subtitle colour, thumbnail, or filename. Those versions may be required deliverables, but they should not inflate the learning count. The standard option below produces eight ads, representing four hypothesis pairs rather than eight independent ideas.

TikTok Ads Manager's split-testing overview, updated in January 2026, explains that the audience is divided into exclusive groups and a winner is verified only when the result is statistically significant. Its best-practices page recommends defining a hypothesis, running for at least seven days, using a sufficiently large audience, and targeting an estimated power of at least 80%. TikTok does not publish one universal minimum spend for every account. CNY 20,000 per pair is therefore a scenario input here and must be replaced with the forecast in the advertiser's own account.

Three budgets purchase different capabilities

All three routes start from the same CNY 180,000 maximum cash pool, but they do not serve the same job.

The first numeric column still needs to be itemised. These are not packaged offers.

Lean testing: CNY 105,000

• Test strategy and scripts: CNY 5,000
• Existing-footage review and necessary pickups: CNY 6,000
• Six test edits at an assumed CNY 2,500 each: CNY 15,000
• Captions, music, and usage review: CNY 4,000
• First media allocation: CNY 60,000
• Measurement record and review: CNY 5,000
• One iteration reserve: CNY 10,000

The total is CNY 105,000. It works only if the source footage truthfully shows the product, the intended paid use is covered by the relevant talent, music, and footage permissions, and three hypotheses can answer the current decision. If the product has changed or a decisive demonstration is missing, cheaper editing simply distributes the wrong evidence more efficiently.

Standard modular production: CNY 180,000

• Strategy, hypotheses, and test matrix: CNY 8,000
• One modular live-action shoot: CNY 28,000
• Talent, location, and sample preparation: CNY 12,000
• Eight edits at an assumed CNY 2,500 each: CNY 20,000
• Music, talent, captions, and agreed market usage: CNY 8,000
• Media across two waves: CNY 80,000
• Measurement records and review: CNY 6,000
• Re-edit, delivery, and uncertainty reserve: CNY 18,000

The total is CNY 180,000. The four pairs should not all compete for attention at once. Run two pairs in the first wave; enter the second wave only if estimated power is sufficient, the setup is clean, and the result can inform the next decision. The CNY 20,000 per pair and the two-wave design are model assumptions, not a TikTok minimum-spend rule.

Flagship first: also CNY 180,000

• Creative development: CNY 15,000
• Pre-production and boards: CNY 15,000
• Shoot, talent, and location: CNY 50,000
• Hero-film post-production and sound: CNY 35,000
• Talent and music rights for the agreed use: CNY 15,000
• Platform versions: CNY 10,000
• Media: CNY 35,000
• Measurement and review: CNY 5,000

The total is also CNY 180,000. It can deliver a more unified brand expression and basic distribution, but it cannot also fund four adequately powered creative comparisons in this scenario. Calling it a performance-testing plan would misstate what was purchased. It becomes sensible when the hero film itself must serve a launch event, homepage, retail partner, or committed media placement.

Work backward from the media floor to the production ceiling

Media should not be whatever remains after production. In this scenario, the account forecast assigns at least CNY 80,000 to two waves and four pairs. The buyer also protects CNY 8,000 for rights and captions, CNY 6,000 for measurement, and a CNY 10,000 minimum reserve. The production ceiling is:

CNY 180,000 - CNY 80,000 - CNY 8,000 - CNY 6,000 - CNY 10,000 = CNY 76,000

The standard route's strategy, shoot, preparation, and editing total CNY 68,000, leaving CNY 8,000 of additional reserve within the ceiling. The flagship's creative, pre-production, shoot, post, and versions total CNY 125,000, exceeding the production ceiling by CNY 49,000. Unless the flagship scope is reduced, the money has to come out of test media. The film may be completed while the original learning objective remains unanswered.

With less cash, do not cut every line equally. Reduce four hypotheses to two, use verified existing footage for a lean test, or defer testing until the account has enough traffic. Keeping four learning questions, a complete live-action shoot, flagship craft, sufficient media, and unchanged rights while squeezing every line item is not a fourth viable route.

What hurdle must the standard upgrade clear?

The standard route uses CNY 75,000 more cash than the lean route. CNY 20,000 of that difference is additional media. If the brand's per-order contribution already subtracts the necessary marginal acquisition cost, counting that media again in the fixed hurdle would double-count it. The incremental non-media project cost is therefore:

CNY 75,000 - CNY 20,000 = CNY 55,000

Contribution per incremental order is not the retail price. Here it means net selling price minus product cost, platform and payment fees, fulfilment, expected returns, discounts, and the necessary marginal acquisition cost. Shopify's explanation of contribution margin also starts by subtracting variable expenses from revenue. The brand's finance or operating owner must replace the scenario assumptions with its own definition.

These are hurdles, not sales forecasts. A higher click-through rate or platform ROAS does not prove that 250 orders were incremental to the video. Google Ads' current Experiment Center overview distinguishes campaign experiments from lift studies: A/B tests compare tactics or assets, while lift studies use exposed and unexposed groups to estimate incremental effects. Availability, eligibility, sample, and method vary by account and market; one dashboard metric is not universal causal proof.

With insufficient traffic, “undecided” is a valid result

No winner does not mean the two ads are equally effective. The difference may be too small, the sample too limited, conversion events too rare, the groups modified during learning, or the window too short. TikTok recommends at least seven days and avoiding mid-test changes. Google Ads says an undecided or unavailable video experiment may need four to six weeks, more budget, or more time. Those recommendations belong to their respective platforms and should not be merged into one universal schedule.

Write the rules before spending:

1. If estimated power is below 80%, do not run four pairs at once. Keep the two largest differences, broaden the eligible audience, or postpone.
2. If the landing page fails, inventory is interrupted, or the setup changes, do not use that wave to declare a winner. Log the failure instead of covering it with extra polished versions.
3. Move into wave two only when one result reaches the platform's significance threshold and the core business metric points in the same direction. More views or clicks with worse valid-order economics is not an upgrade signal.
4. If both waves are inconclusive, stop making more variations of the same idea. Revisit the proposition, audience, offer, or landing page.
5. If incremental contribution is zero or negative, do not use sales return to approve more production. If the content still performs a launch, retail, or sales-training job, evaluate that job directly.

Wistia's 2026 State of Video combines a survey of more than 900 professionals with analysis of over 13 million videos on its platform. Its sample suggests that video demand is rising while almost half of teams are holding budgets flat, with blended in-house, outsourced, and AI-assisted workflows becoming more common. That supports the need for production leverage; it does not establish an ideal CNY budget for a particular DTC brand. AI can help with script drafts, previsualisation, captions, and version preparation, but generated output count is not experimental evidence. Product truth, talent and music rights, and version consistency still require review.

Put inputs, ownership, and completion into the scope

The brand owns product facts, market, offer, landing page, inventory, contribution definition, account forecast, and final approval. Production turns each hypothesis into assets with clear differences and delivers masters, versions, captions, audio, and asset records. The growth or media owner controls experiment setup, audience, budget, events, incident logs, and interpretation. Talent, music, third-party assets, and platform compliance should be confirmed by the responsible rights or professional owner.

ScreenSkills' current production-coordinator responsibilities connect schedules, resources, rights clearances, documents, and budget tracking. A DTC ad project is smaller than a film or television production, but the procurement point carries over: if a production quote omits source review, rights, version naming, the test matrix, or change ownership, the apparent saving may simply move necessary work to the launch deadline.

A test cycle is complete when:

• each pair changes only the agreed variable while the product facts and offer remain stable;
• every file maps to a real ad group, period, audience, and primary metric;
• talent, music, footage, market, and paid-use rights fit the agreed scope;
• the result is labelled significant, not significant, contaminated, or underpowered;
• the reason to continue, stop, or change the question is recorded; and
• masters, clean versions, captions, naming, and revisions meet the contract.

For a broader production service view, see ONCE's short-form video production solution. If the team is deciding how a hero asset should feed an ongoing calendar, the guide to deriving monthly content from one hero video explains the asset relationship. This article answers the narrower budget question: how production and media together create a test capable of answering one decision.

A fillable budget and decision sheet

Calculate the production floor for the necessary task, then the production ceiling after protecting media and operating cash. If they do not overlap, reduce the number of hypotheses, reuse verified footage, stage the shoot, or wait. Do not fill today's cash gap with sales that have not happened.

If you are preparing the next DTC creative cycle, share the product, target market, primary channel, existing assets, sample status, 28-day cash limit, non-sensitive account forecast, and contribution assumptions with ONCE. We can separate the flagship job from the learning job, map hypotheses, footage, production, rights, media, and stop rules, then decide between a lean test, modular production, or a flagship film for a confirmed launch use. Quotation, availability, and outcomes depend on the confirmed scope and real operating conditions.

Source note: checked on October 9, 2026: TikTok Ads Manager Split Testing, Split Test Best Practices, and Experiment Manager; Google Ads Experiments and Experiment Center; Wistia's 2026 State of Video; Shopify's contribution-margin explanation; and ScreenSkills production-coordinator responsibilities. All CNY figures are disclosed scenario assumptions.