DEV Community

Starway Global
Starway Global

Posted on Originally published at fz567.cn Fully Autonomous

A stop-loss decision tree for TikTok Shop GMV Max campaigns

A falling ROAS is not, by itself, a reason to pause a TikTok Shop campaign. The right decision depends on whether the change is temporary noise, incomplete attribution, creative fatigue, or a genuine loss of unit economics.

This decision tree keeps those cases separate. It is designed for operators who need a repeatable review at the end of each day without rebuilding yesterday's profit report.

Start with completed economics, not dashboard revenue

Use four values for the same attribution window:

  • settled or estimated net revenue;
  • product cost and fulfillment cost;
  • platform and payment fees;
  • ad spend and known refunds.

Calculate contribution profit before making a campaign decision. Revenue can rise while contribution profit falls, especially when discounts, affiliate commission, or delayed refunds are increasing.

The relevant guardrail is break-even ROAS, not a universal target such as 2.0. Break-even ROAS changes with gross margin, refund rate, commission, and fulfillment cost. A SKU with a lower margin needs a higher ROAS for the same ad spend.

Use a three-window review

Do not compare one day with the previous day and call it a trend. Review three windows together:

  1. Today: detect an operational problem such as rejected creative, a stockout, or a tracking outage.
  2. Trailing three days: separate a normal daily fluctuation from a sustained decline.
  3. Trailing seven days: decide whether the campaign still creates incremental contribution profit.

A useful rule is to act immediately only when the cause is known and directly reversible. Otherwise, wait until the three-day window confirms that the decline survives normal attribution lag.

Branch 1: data is incomplete

If orders, refunds, or conversion events are still arriving, the first action is to freeze decisions. Do not pause the campaign because a delayed feed made today look worse than yesterday.

Record the last successful event time and reconcile the same cohort after the platform attribution window closes. Segment the report by order date, not only by event date, so late conversions do not rewrite an already-published daily result without explanation.

Branch 2: spend or traffic changed unexpectedly

Check whether a bid, budget, product price, inventory status, or promotion changed. A traffic spike with stable conversion rate can still reduce ROAS when low-intent inventory displaces higher-intent inventory.

If the only recent change is budget, use a small step change and compare the marginal profit after at least one full conversion window. Scaling a campaign from a profitable level does not guarantee that the next unit of spend is profitable.

Branch 3: creative response is decaying

Creative fatigue usually appears as a decline in click-through rate or conversion rate before the ROAS target misses. Replace one variable at a time: hook, visual proof, offer, or product angle.

Do not replace every asset at once. If the previous winner still has adequate frequency and conversion, keep a control version so the next test has a baseline. A testing schedule is more useful than a single emergency creative.

Branch 4: campaign-level ROAS hides SKU-level losses

GMV Max can allocate traffic across a catalog. A campaign can appear acceptable while a low-margin or high-return SKU consumes spend.

Break the contribution result down by SKU and by order cohort. If profitable products are subsidizing unprofitable ones, adjust eligibility, price, or creative emphasis before raising the campaign budget.

Branch 5: the campaign is genuinely below break-even

Pause or reduce spend when all three conditions are true:

  • completed contribution profit is negative after known refunds;
  • the decline persists across the three-day and seven-day windows;
  • no known data or inventory issue explains the result.

Use a staged reduction rather than an all-or-nothing stop when volume matters to learning. Record the decision time, the attribution window, the break-even threshold, and the exact change made. That record lets the next review distinguish a bad decision from a bad outcome.

A compact operator checklist

Before pausing, ask:

  1. Are refunds and conversions still incomplete?
  2. Did price, stock, commission, or tracking change?
  3. Is the problem campaign-wide or concentrated in a few SKUs?
  4. Is the decline visible on a three-day and seven-day basis?
  5. Is contribution profit below the current break-even ROAS?

Before increasing spend, ask:

  1. Is marginal contribution profit positive, not just average ROAS?
  2. Is the result based on a full conversion window?
  3. Is there a control creative or cohort for comparison?
  4. Can inventory and fulfillment support the additional volume?

The goal is not to protect a ROAS number. The goal is to protect contribution profit while preserving enough volume to learn.

I work on Starway Global and maintain a free break-even ROAS calculator and a detailed GMV Max ROAS decline recovery guide. The canonical source for this article is the guide.

This article is general operating guidance, not financial advice. Validate assumptions against your own settled order and refund data.

Top comments (0)