โฆษณา
A typical Thai multi-platform marketplace seller running 3–5 active shops across Shopee, Lazada, and TikTok Shop allocates monthly ad budget by historical revenue share. The mental model is intuitive: Shopee produced 60% of last quarter's revenue, Lazada 25%, TikTok Shop 15%, so the ad budget for next month splits 60/25/15 across the three platforms. The split feels proportional, defensible, and operationally easy. It is also, in classical economic terms, the wrong allocation.
The operationally correct cross-platform ad allocation equalises marginal ROAS — the contribution profit produced by the next ad dollar on each platform — across all three platforms simultaneously. This is the equimarginal principle, the foundational result in economic resource allocation that goes back to Alfred Marshall's 1890 Principles of Economics: when a fixed budget can be deployed across multiple uses with diminishing marginal returns, total return is maximised when marginal return is equal across all uses. In ad-budget terms: the next dollar in your monthly ad spend should produce the same contribution profit regardless of whether it lands on Shopee, Lazada, or TikTok Shop. If it doesn't, you can increase total profit by moving budget from the lower-marginal-ROAS platform to the higher one — without spending an additional dollar.
This note documents the framework, walks the math with a chart of marginal-ROAS curves per platform, runs a worked example showing the gap between historical-share allocation and equalised allocation on a representative account, applies sensitivity to the inputs that move most, and ends with the operating procedure that applies the framework at production cadence rather than as a one-off quarterly exercise.
The next ad dollar should produce the same contribution profit regardless of which platform it lands on. If it doesn't, the allocation is wrong.
Thesis: in our data across ~150 Thai SEA-6 multi-platform accounts running active campaigns on at least two of Shopee / Lazada / TikTok Shop, the gap between historical-revenue-share allocation and marginal-ROAS-equalised allocation accounts for 4–7 percentage points of net contribution margin per quarter on average. The gap is structural: it persists across categories, across account sizes within the sample, and across rolling 90-day windows. The framework recovers it without changing total spend; the work is purely the reallocation.
The marginal-ROAS curve, per platform
Each platform's ad inventory has a finite supply of high-intent placements (high-quality keywords, top-decile recommendation surfaces, premium feed slots) and a much larger supply of low-intent placements that fill in as budget grows. The seller bidding into a platform consumes the high-intent inventory first; as budget rises, additional spend lands on progressively lower-intent placements with progressively lower conversion. This produces a marginal-ROAS curve that declines with budget — the next dollar always returns less than the previous dollar, and the rate of decline differs by platform.
Each line shows the marginal true ROAS — the contribution-profit return on the NEXT dollar of ad spend at that budget level — for a representative Thai mid-tier seller. Shopee's curve declines fastest because the auction is dense and broad-match expansion drives more low-intent traffic at scale; Lazada's curve is flatter because the ad cohort is smaller and intent-driven Sponsored Search dominates.
Read the chart: the two platforms cross around the THB 20K monthly spend level. Below that, Shopee's marginal ROAS is meaningfully higher — the next dollar on Shopee returns more than the next dollar on Lazada. Above THB 50K monthly spend per platform, the relationship inverts — Lazada's marginal ROAS becomes higher because Shopee's auction is saturating while Lazada's curve has flatter decline. The crossing point is the operating signal: at the crossing-point budget, marginal ROAS is equal across platforms, and the equimarginal principle is satisfied. Spending more on Shopee past the crossing point or less on Lazada past the crossing point both reduce total contribution profit even though total spend stays constant.
A worked example — historical share vs. marginal-ROAS-equalised
Take a representative multi-platform account with THB 200K monthly ad budget and the marginal-ROAS curves above. The account's historical revenue share is 60% Shopee / 30% Lazada / 10% TikTok Shop, so the default allocation is THB 120K Shopee / 60K Lazada / 20K TikTok Shop. The equimarginal allocation finds the budget split where marginal ROAS is equal across the three platforms.
Historical-share allocation:
Shopee: THB 120,000 marginal ROAS at THB 120K spend ≈ 1.9
Lazada: THB 60,000 marginal ROAS at THB 60K spend ≈ 4.7
TikTok: THB 20,000 marginal ROAS at THB 20K spend ≈ 4.1
Total contribution profit estimate: ~THB 540,000
Equimarginal allocation (marginal ROAS equalised at ~3.0):
Shopee: THB 72,000 marginal ROAS ≈ 3.0
Lazada: THB 98,000 marginal ROAS ≈ 3.0
TikTok: THB 30,000 marginal ROAS ≈ 3.0
Total contribution profit estimate: ~THB 615,000
Outcome: Same THB 200K total spend; ~THB 75K higher contribution profit per month.
Gap as % of revenue: ~5–6 percentage points.The historical-share allocation overspends Shopee past its high-marginal-ROAS region (the next THB 1 of Shopee spend at THB 120K monthly returns only ~1.9× — lower than the SKU break-even bar on most categories) and underspends Lazada at its high-marginal-ROAS region. Reallocating THB 48K from Shopee to Lazada — and a smaller THB 10K from Shopee to TikTok Shop — moves all three platforms onto the same marginal-ROAS bar. Total spend is unchanged; total contribution profit increases by ~THB 75K per month, ~14% lift on the same budget.
Same THB 200K total monthly budget; allocation differs. The gap is driven by Shopee's steeper marginal-ROAS decline at scale combined with Lazada's flatter curve — Shopee earns the budget at low spend, loses it at high spend, and the equimarginal allocation respects that geometry rather than the historical-revenue accident.
Sensitivity — what changes the optimal split
The marginal-ROAS curves shift with platform-side and seller-side conditions. The table below stress-tests the worked example against the inputs that move most.
| Scenario | Shopee allocation | Lazada allocation | TikTok allocation | Note |
|---|---|---|---|---|
| Baseline (worked example) | THB 72K (36%) | THB 98K (49%) | THB 30K (15%) | Reference equimarginal split |
| Shopee CPC inflation +20% (auction tightening) | THB 56K (28%) | THB 110K (55%) | THB 34K (17%) | Shopee curve shifts down → less budget |
| Lazada Sponsored bid +15% (competitive pressure) | THB 84K (42%) | THB 84K (42%) | THB 32K (16%) | Lazada curve shifts down → rebalance |
| TikTok Affiliate Plan default 25% (commission stack) | THB 76K (38%) | THB 102K (51%) | THB 22K (11%) | TikTok curve flattened by commission stack |
| Total budget 200K → 400K (scale-up) | THB 130K (33%) | THB 200K (50%) | THB 70K (17%) | Curves stay; allocation scales |
| Total budget 200K → 100K (scale-down) | THB 38K (38%) | THB 46K (46%) | THB 16K (16%) | Allocation shifts toward Shopee at low spend |
| Inventory constraint on Lazada hero SKUs | THB 90K (45%) | THB 70K (35%) | THB 40K (20%) | Operational cap overrides equimarginal optimum |
Each row holds all other inputs at the baseline. The framework is mechanical; the input values that drive it are not — auction conditions, commission structures, and inventory constraints all shift the optimal allocation, sometimes meaningfully. The audit cadence has to keep up.
The operating procedure
Methodology in principle is methodology in production only when applied at sufficient cadence. The procedure below is the minimum implementation that produces always-on equimarginal allocation across a multi-platform account.
- Compute marginal true ROAS per platform at the current spend level. Use rolling 28-day windows to smooth campaign-window noise; compute on the most-recent 7-day window for the marginal estimate.
- Identify the equilibrium marginal ROAS — the value where total ad budget equals the sum of per-platform spend at that marginal-ROAS level.
- Compute the per-platform reallocation. Shift budget from platforms below the equilibrium to platforms above it. Cap any single weekly reallocation at 20% of the source-platform budget to avoid disrupting the platforms' bidding-system learning windows.
- Validate against operational constraints. Inventory, fulfillment capacity, and platform-specific campaign-window commitments may cap the receiving platform; the equimarginal allocation is the unconstrained optimum, the operational allocation is the constrained one.
- Re-run weekly. The marginal-ROAS curves drift with platform-side conditions (auction prices, recommendation-system updates) and seller-side conditions (new SKUs, COGS shifts, campaign-window participation). Quarterly cadence is too slow; weekly is the practical floor.
Equimarginal allocation is not a one-off optimisation. It is a weekly rebalance against curves that drift.
Limitations and where this argument breaks
- Inventory and capacity constraints. The framework computes the unconstrained optimum. Reallocating budget into a platform whose listings cannot serve the additional demand wastes the marginal lift; the operational allocation respects inventory caps and fulfillment capacity.
- Campaign-window commitments. Pay Day, Mega Sale, 11.11, and 12.12 carry pre-committed campaign budgets and voucher tiers that override the marginal-ROAS equilibrium during the campaign window. The framework applies between campaign windows, not during them.
- Bidding-system learning windows. Shopee Target ROAS, Lazada Sponsored Search, and TikTok Shop ad bidding all have learning windows (typically 7–14 days) during which the platform's automation tunes against the seller's budget. Aggressive weekly reallocation (>20% per platform per week) disrupts these learning windows and compresses the marginal lift the framework is trying to capture.
- Sample-size lower bound. Marginal-ROAS estimation requires sufficient ad-attributed order volume to compute reliably. Below ~30 ad-attributed orders per platform per week, the marginal estimate is noisy and the framework should be applied with wider confidence bands.
- Internal-data scope. The 4–7 percentage-point margin recovery figure, the cost-input ranges in the worked example, and the typical curve shapes are aggregated across the SEA-6 Thai multi-platform accounts we model directly. They are not population claims about all SEA marketplace sellers; they exclude single-platform operators and very large enterprise accounts.
Methodology
Public-data citations are taken from the Shopee Ads Help Center (Target ROAS bidding documentation), the Lazada Sponsored Solutions seller portal, the TikTok Shop Seller University documentation, the Bain e-Conomy SEA 2025 commentary on retail-media inflation, Sea Limited's 4Q25 / 1Q26 investor disclosures on Shopee's ad-revenue trajectory, and Alfred Marshall's Principles of Economics for the foundational equimarginal principle.
Internal-data claims — the 4–7 percentage-point margin recovery figure, the typical marginal-ROAS curve shapes per platform, the cost-input ranges in the worked example — are aggregated across the Thai SEA-6 multi-platform marketplace seller accounts we model directly. The cross-platform subset is approximately 150 active accounts running campaigns on at least two of Shopee / Lazada / TikTok Shop, across the DataGlass research methodology sample frame (Jan 2024 – Apr 2026, 28-month observation window).