GMV Max, and What It Does With Your Budget

Every advertising platform ends up here. Manual campaigns give a marketer levers and ask them to be right about audiences, placements and bids; automated campaigns take the levers away and ask for a budget and an objective instead. GMV Max is TikTok's version of that trade. The argument for it is real. An algorithm testing thousands of combinations against live behaviour will out-perform a person adjusting settings weekly, provided it has something to learn from. That proviso is the whole article.

Conversion signal, in volume, recently. An automated campaign optimises towards outcomes it can observe, and a product with a handful of sales gives it almost nothing to observe. It will spend the budget regardless, because that is what it was asked to do, and the spending is what people mistake for working. That is why the same feature produces wildly different verdicts from different sellers. A brand with steady sales switches it on and sees efficiency improve. A brand launching a product switches it on and watches a budget disappear into an audience nobody has validated, and concludes the feature is broken. Neither is wrong about their own experience. The practical rule is that automation is something to graduate to. Establish that the product converts, at what cost, to whom, using campaigns you can read. Then hand that over. The sequence is the same one that applies to structuring a launch on any platform: learn first, scale second. Which decisions are worth handing over and which are worth keeping is a question every platform eventually forces, and the answer is more stable than the features are. It is worked through for the older of the two channels in what to automate and what to keep human, and almost all of it carries across.

Less than with manual campaigns and more than nothing, and the remaining controls matter more precisely because there are fewer of them. The budget, which is the only hard limit. An automated campaign will spend what it is given and the ceiling is the decision. The target, which is the objective it optimises against. Asking for volume and asking for efficiency produce different campaigns from identical inputs, and a target set unrealistically tight will starve the campaign of the exploration it needs to find anything. The creative, which is the largest lever left and the one most neglected. Automation decides who sees the video; it cannot make the video better. On a platform where the content is the advertisement, that means creative is now most of the performance rather than part of it. And the catalogue: which products are in it, at what price, with what stock. An algorithm pushing traffic at a product that runs out has wasted the learning as well as the budget.

Not against the campaign it replaced, over the fortnight after the switch. Automated campaigns have a learning period during which the numbers are genuinely unrepresentative, and judging inside it produces a decision made on noise. Give it long enough to leave that period, then compare total sales and blended advertising cost of sales at account level rather than the campaign's own reported return. The account level figure is the one that cannot be flattered by an automated campaign taking credit for demand that already existed. The same caution applies everywhere and is set out in what a reasonable return actually looks like. And watch what happens to organic. A campaign driving genuine incremental demand shows up as more of everything; one harvesting existing demand shows as advertising growing while organic falls by a similar amount, which is a transfer rather than a gain.

Automation optimises the distribution of a budget. It does not touch anything upstream of that, and most disappointing results are upstream. A product priced above what the market will pay does not become affordable because the algorithm found a better audience. A listing that fails to answer the obvious question loses the same proportion of traffic however that traffic arrived. A margin too thin to support any advertising at all stays too thin, and an efficient campaign on a product that cannot afford advertising simply loses money more precisely. This matters because automation is often bought as a fix for poor results, and poor results usually have a cause the campaign type cannot reach. The honest sequence is to establish that the product sells when people see it, then worry about how efficiently people are made to see it. Reversing those two is the most common way to spend a quarter's budget learning something the reviews would have told you. There is also a reporting effect worth naming. An automated campaign optimising towards conversions will find the people most likely to convert, and the people most likely to convert are frequently the ones who were going to buy anyway. That produces excellent campaign level numbers and unchanged total sales, which is the single most common way an automated campaign looks like it is working while nothing has actually improved. The defence is the account level comparison rather than the campaign report, and it is the only one that works. Everything else is a number the platform calculated about its own performance.

During a launch, for the reason above. While testing a new creative direction, because you need to know which video worked rather than that the set of them did. And on any product whose margin is tight enough that a period of unrepresentative spending is not affordable. Running both is normal and is the arrangement most accounts settle into: automation carrying the products with history, manual campaigns doing the learning on everything new, and the graduation from one to the other happening per product rather than as a policy. Deciding when a product has earned that graduation is a judgement about data volume rather than about time, and getting it wrong in either direction is expensive. That call sits inside TikTok Shop management alongside the creative and the catalogue it depends on.