DataGlass is an e-commerce profit optimization platform that combines financial intelligence, predictive advertising models, and AI-powered workflows to help businesses understand and improve profitability.
Most marketplace advertising dashboards focus on attributed sales and ROAS. DataGlass incorporates your actual product economics and predictive campaign response models to optimize for expected contribution profit.
DataGlass supports Shopee, Lazada and TikTok Shop. Available features depend on the marketplace and campaign type.
DataGlass combines transactional data with product costs, commissions, transaction fees, discounts, fulfillment expenses, and advertising costs to reconstruct contribution profit at the product, order, and campaign levels.
DataGlass can automate supported budget, pacing, and Target ROAS decisions, using predictive models and configurable optimization constraints. Available controls depend on the marketplace and campaign type.
DataGlass is designed to operate within defined constraints and supported approval settings, allowing you to retain control over advertising decisions.
Connect your supported marketplace account, set your product costs, and allow DataGlass to synchronize the data needed for analysis. Actual setup and initial data processing times depend on your shop.
DataGlass supports multi-shop management for eligible accounts. Available access controls, reporting, and workflow capabilities depend on the subscription and supported marketplace integrations.
DataGlass Agent helps you explore shop data, investigate performance, identify opportunities, and prepare recommended actions. Some agentic workflows may be available only through pilot access.
DataGlass offers subscription options for different business needs, including growing sellers, brands, and agencies.
Yes. Every client sits in one workspace, and you switch between them or view all shops at once.
You set who approves what. Changes wait for approval unless you allow them to run automatically within limits.
Shopee, Lazada and TikTok Shop. Some features are available on only some of them, and the docs show which.
From each campaign's own spend and sales history, refreshed as new data arrives.
Yes. Every recommendation can wait for approval; automation is opt-in, per campaign, within limits.
Supported controls depend on the marketplace and campaign type.
Yes, for supported campaigns, based on forecast hourly demand.
Every change shows its expected profit, and the simulator tries changes without deploying them.
No. DataGlass shows plain answers and suggests the next step; the maths stays under the hood.
Shopee, Lazada and TikTok Shop. Some features are available on only some of them, and the docs show which.
Upload a sheet or type them in; DataGlass fills gaps with an estimate until you do.
Probably not. The agency runs the campaigns; DataGlass tells you whether the campaigns are profitable. Many shops keep the agency for execution and use DataGlass for the profit-aware decision layer above the agency's work.
Some agencies do model COGS and full economics; that is great. The difference is that DataGlass is the system of record for the seller. Switching agencies, expanding to a new platform, or running a multi-agency setup leaves the data and the decision layer intact.
No. DataGlass is built to complement the ERP. The ERP remains the financial system of record; DataGlass becomes the daily marketplace decision layer.
Direct two-way ERP integration is on the roadmap. Today the workflow is export-driven: DataGlass surfaces the decisions, the operator approves and acts, and reconciled data can be exported back to ERP-friendly formats.
Most ERP marketplace modules focus on order ingestion and inventory sync. They do not surface True ROAS, ad waste, or per-SKU contribution margin. Those are what DataGlass adds.
No. Listings, ads, order management, and customer messaging stay in Seller Center. DataGlass is the analytics and decision layer that sits above all three platforms.
Selected actions, yes. Pricing changes and ad-spend adjustments are surfaced as one-click recommendations that write back to Shopee via the platform's authorized APIs. Other actions are read-only for now.
DataGlass does not run ads. It evaluates the campaigns Shopee is already running against your real margin and surfaces the budget moves that grow profit. Bid changes are deployed via Shopee's native ad system.
No. Most teams keep a planning spreadsheet for things DataGlass does not yet do. The bulk of the per-day reporting and decision work is what moves into DataGlass.
Yes. Every view supports CSV/XLSX export so the work that still belongs in a spreadsheet stays one click away.
The cleanest comparison is hours per week. A seller running three shops typically spends 4–8 hours a week joining and reconciling marketplace data. DataGlass takes that to under one. The remaining hours go into decisions, not data prep.
No. DataGlass does not receive orders, does not run a packing workflow, does not deduct stock across channels and does not do chat commerce — all of which are the defining jobs of a Thai OMS such as Shipnity, Zort, Page365 or BigSeller. DataGlass reads marketplace data and decides: which SKU is actually losing money after fees, vouchers and returns; what tomorrow's Shopee ad budget should be; which prices to move. Most shops that use DataGlass keep their OMS. The two write to different things, with one exception worth planning for — both can set a price, so choose which system owns that field.
E-commerce profit optimization is choosing the values of the decisions you control — ad target, ad budget, price, assortment, and stock — so as to maximise contribution after every real cost, subject to the constraints that bind each decision. It differs from e-commerce analytics in that analytics reports a state and optimization selects an action, and it differs from following platform recommendations in that marketplaces optimize gross merchandise value, which is the base they charge commission on, rather than your profit. In practice the hard part is not the optimization but the objective: contribution per order line has to be reconstructed from cost of goods, category commission, transaction and payment fees, seller-funded vouchers, the free-shipping cost share, fulfillment, and a returns reserve before any decision can be scored against it.
The same five decisions apply, with one Shopee-specific starting point: contribution per SKU after category commission, transaction and payment fees, vouchers, the free-shipping cost share and returns — which gives break-even ROAS as one divided by that margin. Treat break-even as the zero line rather than a floor, because a target set there permits a bid equal to the whole contribution of a click. The full Shopee playbook, with the worked example and the operator checklist, is the Increase Shopee profit pillar linked from this page.
Because they are optimal for a different objective. Marketplaces earn commission on gross merchandise value, so their ad systems, their campaign suggestions and their bid recommendations are tuned to maximise GMV — which is why the flagship Shopee campaign type is called GMV Max. Maximising GMV and maximising profit coincide only when contribution margin is constant across everything you sell, which it never is. A recommendation that lifts GMV by pushing volume into low-margin SKUs, or by lowering an ad target below break-even, is doing exactly what it was designed to do while making you poorer.
Only after the same product is one row rather than three. Shopee, Lazada and TikTok Shop each carry their own fee schedules, voucher mechanics, attribution windows and cost-share rules, so the identical unit can hold materially different contribution on each. Optimizing per platform in isolation therefore produces a portfolio that is locally sensible and globally wrong — most obviously in budget allocation, where the correct rule is to equalise marginal return across platforms rather than to hit a target on each. DataGlass builds a canonical catalogue across the three, computes contribution on each binding, and ranks moves against one objective.
Revenue and profit move together only while contribution margin is positive and roughly constant, and neither condition survives contact with a real marketplace catalogue. In the worked example on our blog, a campaign whose Target ROAS is set two points below break-even posts the largest GMV of any configuration tested — four and a half times the profit-optimal one — and loses THB 6,173 a day doing it. Every visible metric improves as the campaign walks off the cliff. That is the general shape of the problem: revenue is observable and profit is not, so optimizing what you can see is a reliable way to lose money slowly.
For every order, DataGlass deducts COGS, marketplace fees (Shopee commission fee, Lazada platform fee, TikTok Shop transaction fee), attributed ad cost, fulfillment, and any seller-funded vouchers from revenue. The result is a per-order profit breakdown that rolls up into contribution margin per SKU — the per-unit profit that funds the business.
Yes. Every Shopee, Lazada, and TikTok Shop platform fee is modeled per category and per program — commission fee, transaction fee, payment-processing fee, free-shipping subsidy, and seller-funded voucher. The profit breakdown shows each line so the math is auditable per SKU.
Yes. Upload what you have via CSV or XLSX with column auto-mapping. For SKUs without a COGS, DataGlass infers a starting estimate and flags it so you can refine it over time. The profit calculator still produces a directional profit insight while you fill the gaps.
Gross margin only deducts COGS from revenue. Contribution margin also deducts marketplace fees (commission, transaction, platform structure), ad cost per unit, fulfillment, and vouchers — it reflects real per-unit economics on a marketplace and is the number the DataGlass profit breakdown surfaces.
A Marketplace seller pays a commission of 8.50–10.50% by category (Mall 6.12–11.08%), a 2.24% transaction fee on the buyer’s total including shipping, a 5.60% Platform Shipping Fee capped at ₱100 per item (4.48% on Mall), a Seller Growth Support Fee of 1.00% or 1.50% by category since 11 May 2026, and a ₱5 order processing fee after the first 50 orders each month. All rates include 12% VAT. New sellers pay no commission, Seller Growth Support Fee or order fee for 90 days. On a ₱499 women’s clothing order that adds up to ₱101.51, about 20% of the price.
Because per-order margin is computed against the costs that attach to an order — platform fee, payment fee, cost of goods — while the costs that decide the month attach to the month. Seller-funded vouchers are a redemption rate across orders, returns remove orders from the denominator after the margin was computed, affiliate commission arrives as a separate settlement line, COGS drift is invisible until someone re-reads a supplier invoice, and ad spend is a monthly total that no single order carries. Each one is small per order and none of them appears on the order the calculator is pricing.
Work the supply-side cost stack in recovery order, not alphabetically. Refresh COGS monthly on your top-50 SKUs by GMV — drift is the largest and least-audited line, worth 1.5–3.0 percentage points of margin across a typical 87-day refresh gap. Then run a pre-flight check on every campaign that refuses Shop Voucher tiers above what the category margin can carry; Shopee documents Shop Vouchers as seller-funded marketing cost, so an accepted default tier is a cost decision you made by not deciding. Then review Mall vs non-Mall commission tier per category, the Free Shipping Program cost-share, the returns reserve against actual return rate, and a quarterly fulfilment RFQ. In our Thai SEA-6 cohort, that audit recovers 4–8 percentage points of net contribution margin in the first quarter on accounts that have not run one in six months, before any ad-side change.
Each additional unit sold during a scale-up carries a higher ad cost (the auction is bid up by the seller's own additional spend), a heavier voucher subsidy (Shopee's campaign tiers escalate with campaign size), and a marginally lower realised price (more orders at the discounted tier). The platform-reported ROAS does not see any of these compounds, so the dashboard reports a 60% revenue gain while net profit silently falls. The arithmetic is structural, not a mistake.
It depends on the SKU. LazMall delivers a measurable conversion lift — typically 5–10% on visible search placements — but it carries a higher commission tier (commonly 4–6 percentage points more than non-Mall), a mandatory deeper free-shipping subsidy, and stronger participation pressure on campaign days. For SKUs where the conversion lift exceeds the cost-per-order delta, LazMall is profitable; for SKUs where it does not, the badge actively compresses margin. The decision is per SKU, not per account, and it should be reconciled against contribution margin reconstructed from order-line data — not the Lazada P&L summary.
Reconstruct contribution margin per SKU at the order line, not from the Seller Centre income summary. Subtract from each order's revenue: COGS, the program-aware Sale Transaction Fee (varying by category, Mall status and small-seller status — 7.49–17.12% non-Mall and up to 19.26% Shopee Mall, VAT-inclusive, on the schedule effective 4 August 2026), the 3.21% payment transaction fee, the THB 1 per-order Platform Infrastructure Fee, the seller-funded portion of any Shop Voucher applied, the seller-share of the Free Shipping Program subsidy, the seller-funded portion of Shopee Coins / Cashback, attributed ad cost, packaging, and a category-specific returns reserve. Divide by revenue for the contribution margin rate. The Shopee in-dashboard income view typically overstates margin by 4–7 percentage points on a typical Thai SEA-6 account because it under-attributes the three middle lines (Shop Voucher seller-funded portion, Free Shipping Program seller-share, Shopee Coins seller-funded portion).
Reconstruct contribution margin per SKU at the order line, not from the Lazada Seller Centre income summary. Subtract from each order's revenue: COGS, program-aware commission (LazMall vs. standard store, varying by category — 1–4% standard, 3–10% Mall), the ~2% payment processing fee, the seller-funded portion of any Voucher Wallet contribution, the seller-share of the Free Shipping Logo subsidy, attributed Sponsored ad cost (use order-close attribution, not click-time), packaging, and a category-specific returns reserve. Divide by revenue for the contribution margin rate. The Lazada in-dashboard income view typically overstates margin by 4–7 percentage points on a typical Thai SEA-6 account because it under-attributes the two middle lines (Voucher Wallet seller-funded portion, Free Shipping Logo seller-share).
Compute contribution margin per SKU: selling price minus COGS, marketplace commission and transaction fees, seller-funded voucher cost, the seller-funded portion of the Free Shipping Program subsidy, packaging, and payment fees. Divide by selling price for the contribution margin rate. Then derive break-even ROAS = 1 / contribution margin rate. A SKU is profitable on its current campaigns only if its true ROAS (post-attribution contribution profit / ad spend) exceeds break-even ROAS — not if the platform-reported ROAS exceeds the account-wide target. The two answers disagree on roughly half the catalog of a typical Shopee account, which is why the audit produces a different ranking from the dashboard.
Compute true ROAS at the order line, not the campaign aggregate. Subtract platform commission, affiliate commission (if the order was attributed to an affiliate creator), the seller-funded voucher portion, the platform shipping subsidy, COGS, and a category-specific returns reserve. Divide by ad spend. The same campaign that reads ROAS 6.0 in the TikTok Shop dashboard typically lands at 1.4–1.8 true ROAS once the full stack is applied. Set break-even ROAS per SKU (1 / contribution margin rate) and pause every keyword whose true ROAS falls below the SKU's break-even bar — the bar varies meaningfully by category because affiliate commission and returns reserve do.
Yes. Every voucher and flash sale is reconciled against per-SKU contribution margin so you see whether it lifted profit, just lifted GMV, or destroyed margin.
Yes. The bundle-margin projector models contribution margin of each bundle composition before you publish, including the attach-SKU mix and the platform fee impact.
Not yet. DataGlass surfaces the recommended promotion structure and you publish via the platform native flow. Auto-publish is on the roadmap.
Price elasticity modeling estimates, for each product in a seller's catalog, how much unit demand changes in response to a price change. For sellers on Shopee, Lazada, and TikTok Shop, an elasticity model converts three inputs — historical sales data, platform promotional events, and a small amount of deliberately randomized price testing — into a per-SKU demand curve with confidence intervals. The model identifies which products are price-sensitive, which are not, and which Bundle Deal threshold, FlexiCombo configuration, or flash-sale depth maximizes profit under each seller's margin constraints.
Price elasticity in e-commerce is a numerical measure of how much unit demand for a product changes when its price changes. A high-elasticity product gains many additional units sold per percent of price reduction; a low-elasticity product gains few. Marketplace sellers use elasticity estimates to set discount depth, configure Bundle Deals, and target vouchers.
Past sales reflect both price changes and everything else that was happening at the same time — seasonality, ad spend, competitor moves, platform-wide campaigns. A simple regression of sales on price typically gives a biased answer because price is endogenous: sellers raise prices when demand is strong, so the naive correlation between price and demand is too flat. Modern elasticity modeling uses methods like double machine learning, hierarchical Bayesian regression, and deliberately randomized perturbation experiments to separate the causal price effect from confounders.
Platform dashboards report what happened — units sold, revenue, conversion rate. They do not estimate what would have happened under a different price. Elasticity modeling is causal: it produces a counterfactual demand curve so the seller can compare expected outcomes across alternative pricing decisions before committing.
Yes, but only if the model uses partial pooling — borrowing strength from similar SKUs in the same category. A new listing inherits a starting elasticity from its category prior, then updates as its own data accumulates. Models without partial pooling tend to fail on new and long-tail SKUs.
All three benefit, but the highest-leverage applications differ. Shopee's Bundle Deal is the cleanest single lever because it is a structurally well-defined price-discrimination mechanism. Lazada's voucher stack benefits most from elasticity-driven segmentation. TikTok Shop benefits most from short-window, live-stream pricing, where time pressure makes intuition-based decisions least reliable.
It is one of the foundations of dynamic pricing, but not the same thing. Elasticity modeling is the measurement layer; dynamic pricing is the automation layer that uses elasticity estimates to reset prices on a schedule or in response to events. A working elasticity model is necessary for any non-trivial dynamic-pricing system, but a seller can capture much of the contribution gain from elasticity modeling alone, by manually setting better discounts informed by the model.
For sellers migrating from gut-feel to elasticity-aware pricing, the discount mix typically flattens in month 1, sale-event configuration improves in month 2, and the cumulative margin improvement becomes visible by month 3. Published dynamic-pricing programs anchor the realistic company-level envelope at 5-10% margin growth, with higher pilot gains possible in promotion-heavy catalogs.
For a small catalog, no — a managed decision-intelligence platform can deliver elasticity-aware pricing as a service. For a larger catalog with cross-elasticities, category-specific dynamics, and integration into ad spend (where the same demand model feeds both price and budget decisions), the answer becomes "yes, or a partner who provides one."
Because price elasticity varies by SKU. A 10% discount might pull 30% more units on a fashion accessory (high elasticity) and only 3% more on a niche office cabinet (low elasticity). A flat 15% across the catalog over-cuts the inelastic items (margin given up for no volume) and under-cuts the elastic items (volume left on the table). Calibrated elasticity models adjust depth per SKU so each unit of margin given up earns the volume it's actually worth — published dynamic-pricing programs anchor the realistic range at 5–10% margin growth and 2–5% sales growth from doing this correctly.
It depends on per-SKU economics, not on the campaign as a whole. Run a pre-launch contribution-margin simulation for each SKU you would feature, using the campaign's eligibility discount (typically 18% off list at 11.11), the seller-funded voucher tier you would commit to (typically 10–15% during 11.11 windows for placement-weight-eligible slots), and the ad-auction multiple expected during the campaign window (typically 2–3× baseline CPC). For SKUs whose post-attribution contribution margin remains positive, participate at the appropriate voucher tier; for SKUs whose margin inverts, substitute into bundle attach slots only if the bundle's aggregate margin remains positive. Account-level "join everything" decisions during 11.11 are the largest single source of campaign-period margin compression in the accounts we model.
Compute the break-even volume lift for the discount: required volume lift ≈ price_cut% / (contribution_margin_rate − price_cut%). For a SKU with 30% contribution margin, a 10% discount requires ~50% volume lift just to maintain total contribution profit. Most categories see 15–30% volume lift on a 10% discount — i.e. the discount is structurally margin-negative on the typical SKU. The exceptions are highly-elastic categories (consumer electronics, commodity beauty) where elasticity ≤ −2.0 and discounts can be profit-positive at the right tier. Compute the break-even bar per SKU before the campaign goes live; if the SKU's historical volume response cannot clear that bar, the discount destroys margin.
Platform dashboards show attributed revenue. DataGlass shows profit after fees, COGS, vouchers, and fulfillment — the actual number that determines whether the campaign was worth running.
Yes, for the campaigns you put in an optimization group. On Shopee and on TikTok Shop GMV Max, DataGlass sets each campaign’s budget every night from expected profit, steps budgets through the day through each platform’s official API, and proposes a margin-based Target ROAS you apply in one click. Every change is logged with its reason, and you can pause a group or switch a shop to read-only at any time.
Profit-aware bid adjustments target campaigns and keywords that are losing money, not the ones driving organic ranking. DataGlass shows the trade-off explicitly so the decision is informed.
ROAS Protection is a Shopee Ads programme for GMV Max campaigns with a custom ROAS target. If a campaign delivers less than 80% of its target ROAS in a week (70% with Rapid Boost or Campaign Surge) and meets the minimum volume, Shopee pays a 1:1 ad credit of ads spend minus ads GMV ÷ (target × 0.8), the following week, valid for 180 days. The effect is to lift the campaign’s effective ROAS to exactly 80% of its target. That is a profit floor only if 80% of the target is at or above your break-even ROAS, which means setting the target at 1.25× break-even or higher. Changing the target manually ends protection for that period.
Start from contribution margin per SKU, not from the ROAS Shopee suggests. Break-even ROAS is 1 divided by that margin, and the Target ROAS should sit well above it, because break-even earns exactly zero and some attributed orders would have happened anyway. Then check which lever is binding: if a campaign reports a ROAS above the target you entered, the budget is setting the bid and changing the target does nothing. Size the budget from the most you can pay per order. Finally, treat time of day as its own decision: on our Thai panel a click cost twice as much just after midnight as in the evening, and no single target can tell those hours apart.
Start from the product’s contribution margin after TikTok Shop’s commission, Commerce Growth Fee, Transaction Fee, vouchers and cost of goods: break-even ROI is 1 divided by that margin, about 5.3 on a THB 550 womenswear item. Then account for affiliates, because GMV Max counts affiliate orders in its ROI and their commission comes out of the same margin; at 10% commission the same item needs about 10 to break even. Blend the two by your affiliate share, set the target above that line rather than at it, and remember that GMV Max also credits organic orders to itself, so its reported ROI overstates what the ads caused. If the campaign spends its whole budget, the budget is the lever to change.
Shopee GMV Max is an auto-bid campaign type: you set a daily budget and a Target ROAS, and the platform places the per-click bid on your behalf. Underneath, both of your instructions collapse into a single scalar that acts as a threshold on value-to-cost — the campaign bids on an auction in proportion to the value the platform estimates it will get from winning, and wins whenever estimated value divided by clearing cost exceeds that threshold. This means a seller controls how far down the platform’s own value ranking the campaign buys, not which auctions it prefers. The threshold is also the campaign’s marginal ROAS: the return on the next baht of spend, which is always lower than the average ROAS the dashboard reports.
Almost always because the daily budget is binding, not because the campaign is outperforming. The operative bid is the lesser of two ceilings — the price your Target ROAS permits, and the price at which spend would exhaust your budget. When the budget is the smaller of the two, the campaign bids below its permitted price and therefore reports a return above its target. The ratio is exact: entered target divided by reported ROAS equals the fraction of the permitted price the campaign is actually bidding. A campaign entered at 5.00 and reporting 6.32 is bidding 79% of its ceiling. Reading that overshoot as headroom and raising the budget is the standard mistake — it releases the campaign toward a ceiling that, at a break-even target, returns exactly zero.
Change whichever one is currently binding, because the other is inert. If the campaign spends its full daily budget, it is budget-bound and the Target ROAS is doing nothing — every target below the crossover produces an identical campaign. If it underspends, it is ROAS-bound and the target is the live lever while the budget is doing nothing. The test is one division: entered target ÷ reported ROAS. A ratio near 1.0 means the target is binding; a ratio well below 1.0 means the budget is. Changing the inert lever produces no effect, which sellers then misread as the lever being weak rather than switched off.
Not your break-even ROAS. Break-even — one divided by your contribution margin — is the target at which the entire margin on a click goes to the auction, so profit is exactly zero regardless of how much volume you buy. The profit-maximizing target is break-even multiplied by three markups: attribution inflation, because some credited GMV would have happened anyway; one plus the reciprocal of your click-supply elasticity, because winning more clicks raises the price of the clicks you were already winning; and one over one minus the value-decay rate, because your average click is worth more than your marginal one. On mid-market parameters those compound to roughly 3.3x, which moves the correct target on a 20%-margin product from 5.00 to about 16.7. The practical starting point is to set the target well above break-even, then lower it only while the marginal return — not the reported average — stays above your true contribution.
Because the Target ROAS is enforced on the campaign average while profit is decided at the margin, and because attributed GMV is not incremental GMV. A campaign reporting 5.00 against an entered 5.00 has done exactly what you asked. If the target was set at break-even, what you asked for was a campaign that returns its own cost and nothing more — before accounting for the fact that some of the credited revenue would have arrived without the ad. In the worked example in this post, a campaign reporting 5.00 is earning 2.10 on its marginal click and losing THB 827 a day on a true incremental basis, while every metric in Seller Centre looks correct.
No — it is the zero line, and the curve around it is asymmetric. Above break-even a mis-set target forgoes a bounded amount of profit: the most you can lose is what you were going to make. Below break-even the permitted bid is reciprocal in the target, so lowering it raises the bid without limit and the loss with it. In the worked example, the best this campaign can do is THB 800 a day; setting the target two points below break-even loses THB 6,173 a day, nearly eight times that, while posting the largest GMV of any row in the table. Treating break-even as a floor puts you on the steep side of the curve with every visible metric telling you it is working.
One divided by your contribution margin ratio — and the margin is per SKU, not per shop. On a fashion SKU paying 17.12% Sale Transaction Fee, 3.21% Payment Transaction Fee and a flat per-order fee, with a 51% cost of goods, a shop voucher and an absorbed shipping share, contribution margin ratio is about 21.6% and break-even ROAS is 4.64. If only 60% of the orders Shopee attributes to the ad were actually caused by it, the dashboard has to read 7.73 for the campaign to be at break-even. The same SKU with a higher landed cost needs 11.31. There is no shop-wide answer.
Under cost-per-click billing, no — not directly. Your cost per click is the price the auction cleared, and click-through rate does not appear in it: a better creative buys the same number of clicks from fewer impressions, so at a fixed daily budget your cost per click, your attributed revenue and your profit are all unchanged. What a better creative does buy under CPC is delivery — the auction ranks on expected revenue per impression, so a higher click-through rate wins more impressions at the same bid and lets you spend more. Under cost-per-impression billing the relationship changes completely: effective cost per click becomes CPM divided by (1000 x CTR), so a click-through gain of g cuts the price of every click by g/(1+g) — a 25 per cent gain makes clicks 20 per cent cheaper. Profit responds more than proportionally, by gross contribution divided by profit, which grows without bound as a campaign approaches break-even.
ROAS stands for Return on Ad Spend — the amount of sales you get back for every THB 1 you spend on ads. The formula is ROAS = sales ÷ ad spend, so if THB 1 of ad spend returns THB 20 of sales, your ROAS is 20. It is measured in sales, before COGS, platform fees, payment fees, and vouchers are subtracted — so a high ROAS shows your ads are efficient per baht, not that you are necessarily profitable.
ROAS = sales ÷ ad spend. Take the sales attributed to a campaign and divide by what you spent on it: THB 5,000 of sales from THB 100 of ad spend is a ROAS of 50; THB 200,000 from THB 10,000 is a ROAS of 20. To turn ROAS into real profit, multiply it by your margin after all costs — a 15% margin at a ROAS of 5 returns only THB 0.75 of contribution per ad baht, which is a loss on every baht spent.
There is no single "good" ROAS number, because it depends on your product's margin. A high-margin product can tolerate a lower ROAS, while a low-margin product needs a much higher ROAS just to stay profitable. What you should really watch is your ROAS after all costs are subtracted — cost of goods, GP and platform fees, payment fees, coupons, shipping and packing, returns, and ad cost — not the ROAS shown on the dashboard. The real question is always: at this ROAS, is there any profit left?
No. A high ROAS means a high return per baht of ad spend, not the highest total sales. A shop spending THB 100 to make THB 5,000 has a ROAS of 50, while a shop spending THB 10,000 to make THB 200,000 has a ROAS of only 20 — yet the second shop sells 40× more. ROAS measures efficiency per baht; total sales depend on how much you can profitably spend, and the two often move in opposite directions as you scale.
Not necessarily. The ROAS a platform reports is based on sales, not profit — it does not subtract your cost of goods, GP and platform fees, payment fees, vouchers, shipping and packing, or returns. After those costs, a healthy-looking ROAS can leave very little real profit, or even a loss. The number to watch is ROAS after costs: a thin-margin product needs a much higher ROAS than a high-margin one just to break even.
Target ROAS (target return on ad spend) is the revenue-to-ad-spend ratio you set as the goal for an automated campaign such as Shopee GMV Max — a Target ROAS of 5 tells the system to keep bidding as long as every THB 1 of ad spend brings back at least THB 5 of sales. It is measured in gross sales, before COGS, commission, payment fees, and vouchers, so a healthy-looking Target ROAS can still lose money. To turn it into profit, multiply it by your margin — a 15% margin at a Target ROAS of 5 returns only THB 0.75 per ad baht — and never set it below your break-even ROAS of 1 ÷ margin.
Start from your margin before ad spend — selling price minus COGS, Shopee commission and payment fees, seller-funded vouchers, and a packing, shipping, and returns allowance. Your break-even Target ROAS is 1 ÷ that margin. Use the current rates: on the Shopee schedule effective 4 August 2026 (4 ส.ค. 2569) the Sale Transaction Fee is 7.49–17.12% non-Mall and up to 19.26% for Shopee Mall, VAT-inclusive, plus 3.21% on payment — so a THB 1,000 product with THB 600 of COGS and an 8% voucher has 7.7% margin before ads and breaks even at a Target ROAS of 13.0, not the 6.67 an out-of-date fee assumption would give you. To actually keep profit, set Target ROAS = 1 ÷ (margin before ads − the profit you want to keep); a 15% margin targeting 10% net needs a Target ROAS of 20. And do not use one number for everything: a 40%-margin product can run 3–4, while a 15%-margin product may need 7–10 or more.
Because the ROAS your dashboard shows is "advertising-attributed revenue divided by ad spend" — a gross sales ratio, not a profit one. It is calculated before commission, transaction fees, seller-funded vouchers, the free-shipping cost-share, and the cost of the goods are subtracted, and it still counts orders that were later returned or cancelled. A campaign reporting a 5.0 ROAS nets out to roughly 1.05 in true terms on the 50%-COGS profile worked in this post, once the 4 August 2026 fee schedule, the seller-funded lines and the returns reversal are removed — at break-even, not above it. True ROAS subtracts the full variable cost stack and the returns reversal, then divides by ad spend, so it tells you whether the ad actually made money. Compare each campaign to its own SKU's break-even ROAS rather than to a single account-wide target, because a 50%-COGS SKU and a 25%-COGS SKU break even at very different numbers.
Apply three levers in order: visibility (win impressions cheaply on long-tail keywords whose auction has not been bid up yet), conversion (fix the single worst gate per SKU — image, review block, price anchor, or checkout friction — not all four at once), and repeat (Shopee Chat outreach 7 days post-delivery on top-margin SKUs, three-SKU cross-sell graph per primary product). Track win rate (sales per impression), not GMV rate, to verify the lift was earned not bought. Run the cadence on Shopee's own campaign calendar — 21-day pre-event ramp, 3-day in-event window, 30-day re-engagement, 36-day steady state — rather than constant-on voucher pressure between events.
Allocate by marginal ROAS equalisation, not by historical revenue share. The next ad dollar on each platform should produce the same contribution profit; if it doesn't, shift budget from the platform with the lowest marginal ROAS to the platform with the highest until they match. Compute marginal ROAS at the current spend level on each platform — it is the contribution-profit return on the next dollar, not the average ROAS the dashboard reports. On a typical Thai multi-platform account, the equalised allocation differs from the historical-share allocation by a meaningful amount; in our data, applying the framework recovers 4–7 percentage points of net contribution margin per quarter without changing total spend.
Because the metric you scaled against does not contain the costs that scale with you. Shopee reports ROAS as attributed gross revenue divided by ad spend, before commission, transaction fees, seller-funded vouchers, shipping subsidy, and COGS. As budget rises, the auction is bid up by your own additional spend, campaign voucher tiers escalate, and more of the volume lands at the discounted tier — so the marginal order costs more and returns less than the average order the dashboard is reporting. The diagnostic is to compute true ROAS per SKU (attributed revenue minus COGS, fees, vouchers and shipping, divided by ad spend) and compare it to that SKU's break-even bar of 1 ÷ (1 − variable cost share). Keywords below the bar are where the profit went.
No. Shopee's in-platform ROAS, per the Shopee Ads Help Center documentation, is ad-attributed revenue divided by ad spend. It is a gross-revenue metric, not a profit metric. The numerator excludes COGS, marketplace commission, transaction fees, seller-funded vouchers, the Free Shipping Program subsidy, packaging, and returns provisioning — the entire variable-cost stack that compounds at scale. True ROAS is the same formula with contribution profit before ads (revenue minus all of the above) substituted for gross revenue. On the 55%-COGS profile worked in this post, a campaign reading 8.0 platform ROAS lands at 0.70 true ROAS under the Shopee fee schedule effective 4 August 2026 — below 1.0, which means the ads are destroying contribution rather than buying it. The two numbers are different by a structural factor, and the platform dashboard does not display the difference.
Compute true ROAS per placement, not per campaign. For Sponsored Search: aggregate ad-attributed revenue and ad spend per keyword × SKU, subtract LazMall (or standard-store) commission, payment fee, Voucher Wallet seller-funded portion, Free Shipping Logo seller-share, COGS, and a category returns reserve from the revenue. Divide by ad spend. For Sponsored Discovery: same arithmetic at the audience × SKU resolution. Compare the result to the underlying SKU's break-even ROAS (1 / contribution margin rate) rather than to the account-wide target. Placements where true ROAS sits below break-even are the structural waste; the platform-ROAS dashboard cannot distinguish them from winners. In our data, this audit recovers 20–30% of monthly Lazada ad spend in the first 30 days without losing meaningful attributed revenue.
Yes. The stockout-day model accounts for scheduled campaign windows (11.11, 12.12, Pay Day, Lazada flash sales) and TikTok Shop live events that historically lift demand on specific SKUs.
Yes. The canonical product catalog binds the same product across Shopee shops and platforms so inventory state is one number, not per-listing. Opt-in writeback keeps Shopee total_stock in sync.
Not yet. DataGlass surfaces a ranked restock list with the recommended quantity and lead time; PO creation is on the roadmap.
On our panel of 80+ Thai Shopee shops running single-product campaigns between February and August 2026, effective cost per click ranged from THB 3.36 in the 20:00–23:59 window to THB 6.75 in the 00:00–01:59 window; outside the midnight window the whole day sits between THB 3.36 and THB 4.45. The daypart spread is 2.01x, which is larger than the spread on any behavioural input we measure. Treat these as a reference band rather than a target: they are averages across a panel weighted toward mid-market advertisers, they exclude platform campaign days, and your own CPC depends on your category, your competitors and how saturated your audience already is.
On this panel the 20:00–23:59 window carries both the lowest CPC (THB 3.36) and the lowest CPM (THB 89.16), and it has the highest profit on ad spend of any window at 3.63. The most expensive window is 00:00–01:59, at THB 6.75 per click and THB 200.15 per thousand impressions — the platform’s own checkout peak. But this is an observational ranking of spend that already happened, not a measured effect of moving spend. The honest status of any gain from reallocating budget across the clock is unmeasured, and we have deliberately not published one.
POAS is contribution divided by ad spend, so it breaks even at 1.0 on every product — unlike ROAS, whose pass mark moves with margin. On this panel of 80+ Thai Shopee shops the six dayparts ranged from 2.25 to 3.63 on an attributed basis, with a click-weighted blend of 3.11. Read those as reference points and not as targets: they are averages over spend that already happened, they credit sales the ads may not have caused, and the decision to spend one more baht depends on the marginal return rather than the average, which is always lower. A campaign averaging POAS 3.11 can still be losing money on its last baht.
E-commerce optimization is the use of data and mathematical optimization — the same operations-research methods used to plan production and route logistics — to decide the next best action for an online store, measured by true profit rather than sales. On Shopee, Lazada, and TikTok Shop it answers questions a sales dashboard cannot: where to add or cut ad budget by true ROAS per SKU, which promotions are eating contribution margin, which products to push, and where to reprice. The distinction that matters is profit versus revenue — a campaign can grow GMV while shrinking profit once marketplace fees, vouchers, shipping, and ad cost are subtracted. Optimization makes that hidden arithmetic the basis of the decision, instead of the surprise at quarter-close.
Decision intelligence in e-commerce is the discipline of turning marketplace and operating data into ranked, auditable next-actions across pricing, demand forecasting, inventory, promotions, and personalization — and integrating those decisions into a single workflow rather than handing them off as charts. Mathematically it sits at the intersection of statistical learning, causal inference, optimization under uncertainty, and reinforcement learning; operationally it sits behind a feature store, an experimentation system, and a human-in-the-loop UI. The bottleneck for most sellers in 2026 is no longer choosing between algorithms — those are documented in textbooks and shipped in open-source software. It is the integration layer that turns those algorithms into a decision a merchandiser will actually approve.
Production dynamic pricing combines three pieces. First, an elasticity-aware demand model — typically a hierarchical Bayesian model or a structural choice model — that yields per-SKU expected demand at candidate prices. Second, an optimization layer that picks the price that maximises expected contribution after returns, ad cost, and platform fees, subject to margin floors, competitor-aware bounds, and category guardrails. Third, an experimentation system (A/B or off-policy evaluation) that closes the learning loop so the elasticity estimates do not drift. Repricers without an elasticity model behave like rule engines and tend to chase the floor; the literature is clear that elasticity-aware pricing recovers materially more contribution at scale.
The honest answer: a combination, not a single winner. Hierarchical reconciliation is used to keep store/region/SKU forecasts coherent. Intermittent-demand methods (Croston, ADIDA, TSB) handle the long tail of slow-moving SKUs. Deep methods — DeepAR, N-BEATS, Temporal Fusion Transformers — earn their keep on dense, regular series and on cross-sectional learning across many similar SKUs. Quantile forecasts, not point forecasts, feed downstream inventory and promotion decisions because the tail risk is what costs money. Stacking and exogenous-feature ingestion (price, promotion, holiday, weather) are standard. The differentiator is rarely the model class — it is the data hygiene and the quantile-aware optimization on top.
The newsvendor model is the right starting point: a single SKU, a single period, a known demand distribution, an explicit critical fractile that trades off underage cost (lost sale, lost margin, lost rank) against overage cost (carrying cost, markdown, storage). It is also where most production systems still live, augmented with quantile demand forecasts and stochastic lead times. Multi-echelon models matter when inventory is held in more than one location and shipments cascade — typical of marketplace sellers running fulfilment-by-marketplace plus their own warehouse. In that case Clark–Scarf base-stock policies, or their robust-optimization variants under demand ambiguity, replace single-echelon newsvendor logic. Both are cheap to implement; the gain comes from honest service-level targets, not exotic policy structure.
Marketplace sellers run far more decisions than they can A/B test. Every campaign, every voucher, every category-level price move would, in a pure-A/B world, need its own randomised exposure — and platforms increasingly do not allow it at the granularity sellers need. Uplift models (causal forests, X-learner, R-learner) estimate the per-unit effect of an intervention from observational or partially-randomised data, so promotions can be targeted at customers and SKUs where the lift is highest rather than at the overall average. Off-policy evaluation (doubly robust estimation, IPS, weighted importance sampling) lets the seller score new policies against historical data before deployment. Together they extend the reach of experimentation to the long tail of decisions that A/B testing alone cannot cover.
A working deployment needs six components in production: data connectors that ingest order-line, ad-spend, inventory, fee, and return data with reliable joining keys; a feature store that serves the same features to training and inference at SKU and customer grain; a model layer that produces calibrated forecasts, elasticities, uplifts, and personalisation scores; an optimization layer that combines those signals into ranked decisions subject to operating constraints; an experimentation system covering A/B, multi-armed bandits, and off-policy evaluation; and a human-in-the-loop UI that surfaces the math behind every recommendation so the merchandiser can approve, override, or escalate. The moat in 2026 is not which algorithms are used — those are commoditised. It is whether all six components are wired into a single workflow that a merchandiser uses every week.
Build a three-stage decision loop, not a tactic checklist. Stage one: reconstruct contribution margin per order — commission, transaction fee, seller-funded vouchers, Free Shipping Program subsidy, COGS, returns reserve, all per order, not as category-mean assumptions. Stage two: apply per-SKU rules — break-even ROAS = 1 / contribution margin rate — instead of an account-wide ROAS target; pause campaigns whose true ROAS is below their SKU's break-even bar regardless of dashboard signals. Stage three: run the audit on a cadence (rolling 14-day window) rather than reacting to big-campaign weeks. The leverage is in the loop's tightness, not in the tactic count. Tactics inherited from a different account underperform decisions made about your own.
Shopee fees come in six layers, not two. Layer 1 is the Sale Transaction Fee — 7.49% to 17.12% non-Mall and up to 19.26% for Shopee Mall, VAT-inclusive, on the schedule effective 4 August 2026 (4 ส.ค. 2569). Layer 2 is the Payment Transaction Fee, 3.21% VAT-inclusive on standard payment methods since 7 April 2026. Those two are the published figures, and together they already take about 20 baht out of every 100. Layer 3 is the seller-funded portion of Shop Vouchers, escalating to 10–15% on big-campaign tiers. Layer 4 is the seller-funded portion of the Free Shipping Program subsidy, quasi-mandatory because non-participation costs ranking weight. Layer 5 is the Mall premium — 0 to 3.21 commission points depending on the category band, plus stricter program obligations. Layer 6 is the payment-method spread (up to 7.49% on a 10-month instalment) and the THB 1 flat Platform Infrastructure Fee. On a Mall apparel order with a 10% voucher and FSP participation, the all-in take-rate works out to roughly 31%, against the 19.26% the headline rate implies.
Yes. Shopee Thailand's Sale Transaction Fee schedule effective 4 August 2026 (4 ส.ค. 2569) raised 1,382 of 1,437 leaf categories by 2 percentage points before VAT — 2.14 points VAT-inclusive — with 20 categories up 1 point and 35 unchanged. Non-Mall rates now run 7.49% to 17.12% and Shopee Mall up to 19.26%, VAT-inclusive as published. Phones, tablets, laptops and desktop computers were among the categories that did not move, staying at 7.49%. The revision also removed the discounted-commission programme (PRIME Seller) rate column, so shops that had been on the programme took a 3.21-point increase rather than 2.14 — half as large again as everyone else's.
Shopee's Sale Transaction Fee per order is the category rate — conditional on category, Mall status and small-seller status — multiplied by the buyer-paid amount, not the listed price. Buyer-paid is list price minus the seller-funded discount. On the schedule effective 4 August 2026 (4 ส.ค. 2569) the rate runs 7.49% to 17.12% for non-Mall listings and up to 19.26% for Shopee Mall, VAT-inclusive as Shopee publishes it — apparel, beauty and FMCG at the top, phones, tablets, laptops and desktop computers at the 7.49% floor. Do not add 7% VAT on top: it is already inside the published figure. Commission is one of three platform-take lines on the order; the other two are the Payment Transaction Fee (3.21% VAT-inclusive on standard methods, up to 7.49% on a 10-month credit-card instalment) and the THB 1 Platform Infrastructure Fee per completed order. A complete calculation needs all three, not just the commission rate × list price that simple calculators return.
At minimum, seven surfaces: orders (Shopee Seller Centre or Open Platform API), product catalog with SKU + variation data, ad campaigns (keyword- and product-level performance from Shopee Ads), COGS (uploaded or AI-inferred per SKU), marketplace fees (commission, transaction fee, voucher seller-funded portion, Free Shipping Program seller-share), inventory state (current stock + lead time), and returns log (post-attribution reconciliation). The canonical-entity model that makes this surface decision-ready is narrow — SKU, Order, Campaign, Ad spend, Cost, Inventory, Price, Promotion, Channel — nine entities; everything else is a derived view. Connecting Shopee Open Platform + uploading COGS should take ~5 minutes for a single-shop operator; for multi-shop operators, the canonical product catalog (binding the same physical SKU across shops) is the unlock that makes per-shop integration time scale linearly rather than quadratically.
On a typical Shopee SKU with 6+ months of clean order-line history, a feature-engineered gradient-boosted-tree model with stockout-period censoring achieves ~13% sMAPE on the 28-day horizon — vs. ~38% sMAPE for a naive moving-average baseline. The single highest-leverage step is censoring stockout-period sales (treating them as missing rather than observed); the model-architecture upgrade contributes a smaller share of the accuracy gain. New SKUs (under 6 months of history) and SKUs moved by viral content have meaningfully wider confidence intervals, and the operating procedure should reflect that — high-confidence forecasts feed automated reorder, low-confidence forecasts feed human review.
A stockout is not one cost; it is five compounding costs. (1) Lost contribution profit on the unit not sold — the visible cost most accounts count. (2) Wasted ad spend during the stockout window — clicks land on the unavailable product page and the platform charges for them anyway. (3) Ranking demotion — Shopee's algorithm tends to penalise out-of-stock listings during high-traffic windows and the recovery takes 30–60 days. (4) Repeat-buyer trust erosion — buyers who hit an empty product page once may forgive it; twice usually loses the relationship. (5) Distorted future forecasting — stockout-period sales data is censored, and naive forecasts learn an under-demand pattern that increases the next stockout's likelihood. The full multi-line cost is typically 3.5–5× the lost-contribution-profit-only estimate over the 60-day window following the stockout.
A dashboard is descriptive: it surfaces what happened (campaign sales, SKU units, attributed revenue, ROAS). A decision engine is prescriptive: it surfaces what to do next, ranked by projected profit lift, with the math behind every recommendation. The five-layer architecture (ingestion → normalisation → metric calculation → pattern detection → recommendation) is what produces the prescriptive output. The seller still owns the decision; the engine eliminates the analysis tax of getting to it. For SEA marketplace sellers in 2026 the descriptive-only model has stopped being sufficient because the platform-side optimisation cadence is faster than monthly P&L review can keep up with — Sea Limited's documented AI investment cycle, the Google–Sea agentic shopping prototype, and the Bain-documented retail-media inflation all push the seller-side decision frequency upward.
Below roughly THB 200K in monthly revenue, the operational overhead of running the architecture exceeds the margin recovered — simpler heuristics (a per-category margin lookup, a manual weekly reconciliation) outperform. The engine also struggles when order-line data is materially incomplete, when SKUs cannot be mapped reliably across shops (Layer 2 failure), or when the decision frame is genuinely novel (a new campaign mechanic, a one-off platform policy change) where pattern detection has nothing to anchor on. In those cases, human judgment still wins.
An e-commerce decision engine is a software stack that ingests marketplace data — orders, ads, fees, inventory, costs, returns — and outputs ranked, auditable next-actions for the seller, attached to specific entities (campaigns, SKUs, prices, reorders). It differs from a dashboard in that the dashboard answers "what happened?" while the decision engine answers "what should I do next, and why?" The seller still owns the decision; the engine eliminates the analysis tax of getting to one.
Layer 1 — Ingestion: pulls data from marketplaces, ad platforms, inventory, costs, pricing history. Layer 2 — Normalisation: maps SKUs, variations, campaigns, costs, and orders into one canonical structure across shops and platforms. Layer 3 — Metric calculation: contribution margin, true ROAS, break-even ROAS, ad waste, stockout risk, days-of-supply, price-elasticity signals. Layer 4 — Pattern detection: recognises failure modes (high revenue with low profit, ads on stockout-risk SKUs, discount-driven revenue, deteriorating margin). Layer 5 — Recommendation: outputs a ranked list of suggested actions attached to specific entities, with the math attached for audit. Most spreadsheet-based seller setups fail at Layer 2.
Order-line data from Shopee, Lazada, and TikTok Shop (so contribution margin can be reconstructed per order, not estimated per category); ad-platform data from each marketplace's ad surface; inventory state and lead times; per-SKU COGS and supplier costs; pricing history; fee schedules including transaction fees, commission, voucher mechanics, and platform-funded promotions; and settlement/payout schedules so working capital is in view alongside P&L. Without all of these joined at the canonical-SKU level, the recommendation layer can't be honest.
No, and that is the point. The engine eliminates the analysis tax of getting to a defensible decision; the seller still approves the action and owns context the engine doesn't see (a supplier delay, a product launch, a reputational reason to keep a campaign live). Good decision engines surface confidence and the math behind every recommendation so the seller can override with a known reason. A decision engine that overclaims confidence will burn trust faster than no recommendations at all.
Compute per-channel contribution profit (not just revenue) per SKU from order-line data. The same physical SKU often produces meaningfully different margin on Shopee vs Lazada vs TikTok Shop because the fee structure, voucher mechanics, ad-auction characteristics, and customer-return profile vary per platform. Allocate inventory and ad budget toward the channel that produces the highest contribution profit per SKU — not the channel with the highest GMV. The two answers diverge on a typical multi-shop account; in our data, applying per-channel profit reconstruction reallocates 12–20% of ad budget across platforms in the first 60 days and lifts annual contribution margin yield by 5–8 percentage points without measurable revenue loss.
Only on the commission line, and only below about THB 3,720 an order it is actually more expensive. For Thai fashion, TikTok Shop charges 11.77% commission against Shopee's 17.12%, but adds a Commerce Growth Fee of 8.03% including VAT (capped at THB 199 per item) and a 3.21% Transaction Fee on top, so a THB 550 order gives up 23.2% on TikTok Shop against 20.5% on Shopee. Because the Commerce Growth Fee is capped and Shopee's rate is not, the two cross at roughly THB 3,720 of order value, and above that TikTok Shop is the cheaper of the two. Lazada's Fashion schedule is published as a range rather than a rate, so it can be either.
Because the cost of buying demand on Shopee has been rising faster than the cost of repeating it. Sea Limited's 4Q25 disclosure documents the operating-cost rise driven by AI investment in search, recommendations, and advertising; Bain's e-Conomy SEA 2025 documents the consequent retail-media CPC inflation. A discount today retrains the buyer to wait for the next campaign — pulling demand from the future at compounding cost. The cheap, repeatable lift comes from converting impressions you already have at a higher rate, and from bringing buyers back without paying the auction again.
Per Momentum Works, SEA platform e-commerce reached US$157.6 billion in 2025 (up 22.8% YoY); platforms now account for ~85% of regional e-commerce GMV; the top three platform groups (Shopee, Lazada, TikTok Shop / Tokopedia) control ~98.8%. The base-case 2026 projection in this report is US$186–190B platform GMV, with upside to US$195–200B if cross-border flows hold and content-led conversion deepens. But scale is not the operating answer. Per Sea's 4Q25 disclosure, Shopee's marketplace take rate rose from 12.4% to 13.0% and Q4 ad revenue grew >70% YoY — monetization is rising faster than GMV. The 2026 question for sellers is not "how big is the market?" but "who controls the decision loop?" The winning seller owns one data layer, one SKU taxonomy, and one weekly decision cadence over pricing, inventory, ads, logistics, and compliance — not the highest GMV ranking.
Sea's 4Q25 disclosure shows ad revenue grew >70% YoY while ad-paying sellers grew 20% and average ad spend per ad-paying seller grew 45% — i.e. the platform is monetising sellers faster than it is adding them. The marketplace take rate also rose from 12.4% in 2024 to 13.0% in 2025. On the working-capital side, Mall sellers are paid roughly 15 days after delivery (after the return window). Stack ad intensity, take-rate creep, fee layering, refund abuse (57% of merchants per the Merchant Risk Council), and slower payouts, and contribution margin compresses even as top-line GMV expands.
Content commerce — short video, livestream, affiliate-mediated discovery — is no longer peripheral. Per Momentum Works it accounts for ~US$49.7 billion or 32% of regional platform GMV in 2025. Indonesia alone saw video-commerce sellers up 75% YoY to 800,000 and annual video-commerce transactions up 90% to 2.6 billion. For sellers that means demand is increasingly retention- and content-mediated rather than search-led; ad budget allocation, creative production, and affiliate economics now compete for the same SKU-level margin envelope as conventional Sponsored Search.
Sea Limited reported Shopee served around 400 million active buyers and 20 million sellers in 2025; Shopee's 2025 GMV grew 27%. Top-three platforms (Shopee, Lazada, TikTok Shop / Tokopedia) now control approximately 98.8% of regional platform GMV. Country-level Shopee share of platform GMV in this report ranges from ~49% (Philippines) to ~58% (Vietnam), with a regional average near 53%. The Shopee-addressable market across the region in 2025 is estimated at roughly US$83 billion — large, but increasingly contested at the platform-monetisation layer.
GMV is a useful scale measure but a poor steering metric on its own. A seller can grow GMV 30% while their contribution margin shrinks if growth is bought through ads at unprofitable ROAS, discounts that overshoot the volume-lift breakeven, or campaigns where the platform-funded promotion mechanics push spend beyond the SKU's true contribution capacity. Operationally honest sellers measure profit per SKU after ads, fees, returns, and funding cost — not GMV ranking. The 2026 question is not "how big is the market?" but "who controls the decision loop?"
Five operating metrics matter more than GMV: contribution margin per SKU after ads, fees, vouchers, and returns; true ROAS per campaign against the SKU break-even ROAS; days-of-supply against lead time and the campaign demand window; payout lag and working-capital tied up between order and credit; and content-commerce mix as a share of demand because that segment now represents ~32% of platform GMV. Together these tell the seller which products earned growth, which products bought growth, and which products quietly destroyed cash.
Three structural changes documented in the platforms' own filings and seller documentation. (1) Ad auctions matured — Sea Limited's 2025 disclosures document a multi-year AI investment in search, recommendations, and advertising; the seller-side effect is that maintaining the same impression share typically requires 12–18% of revenue on ads in 2026 vs. 4–6% in 2022. (2) Promotions shifted from platform-funded to seller-funded — Shopee Help Center, Lazada Open Platform, and TikTok Shop Seller University all document explicit seller funding for vouchers, free-shipping subsidies, and affiliate commissions. (3) Cross-platform price arbitrage closed — comparison shopping happens in seconds across all three platforms, and competing sellers automate price-mirroring within minutes. The aggregate effect is the same regional growth rate at materially lower seller-side margin per unit on the increment.
Because the unit economics worsen at scale. Each additional unit sold during a scaling push carries a higher ad cost (auction price moves with the seller's own bid pressure), a heavier voucher subsidy (campaign voucher tiers escalate), and a marginally lower realised price (more orders at the discounted tier). On the increment, the new units commonly clear at single-digit or negative contribution margin. The dashboard reports the GMV and hides the marginal economics, so the seller sees growth where the books show compression.
Most multi-shop sellers operate each shop and platform as a separate system — separate logins, separate ads dashboards, separate fee reconciliation, separate COGS spreadsheets — and reconcile manually at week or month close. The dominant pattern in our data is 10–15 weekly hours spent on cross-shop reconciliation work that adds no direct revenue. The fix that recovers the most time is a canonical product catalog: bind the same physical product across shops and platforms once, then operate decisions (contribution margin, stock, ad attribution, reorder) at the product level rather than the listing level. After that, each new platform integration adds minutes of setup rather than weeks of spreadsheet plumbing.
Each platform runs ML models on every product impression — ranking in search, ranking in recommendations, ad auction matching, and (on TikTok Shop) creator-affiliate scoring. Sea Limited's 2025 disclosures and the Reuters reporting in March 2026 documented Sea's AI infrastructure investment for Shopee's search, recommendations, and advertising auctions. The Google–Sea partnership announced in February 2026 extends the direction with an agentic shopping prototype. Alibaba and ByteDance have made parallel investments. The seller-side implication is that every product impression won is the output of an optimization system the seller cannot see; the operating fix is to run a per-SKU optimization layer on top — true ROAS per campaign, contribution margin per SKU, pre-launch margin simulation, post-window reconciliation against margin not GMV.
See it on your shop
Book a 30-minute demo. We walk through DataGlass on your own Shopee, Lazada and TikTok Shop data, and you get 14 more days to try DataGlass free.
Get your shop report
Our team prepares a free report on your shop's unit economics and ads.
View a sample reportStart for free
Connect your shops and act on every decision DataGlass ranks for you.
