GMV Max Not Spending Your Budget? That's a Signal, Not a Malfunction

You set a daily budget. GMV Max spends a fifth of it. The instinct is to raise the budget, or to conclude that your account is being limited. Neither is right, and both waste a week.

GMV Max only spends when it predicts your ROI target can be met. Low utilization is not the machine failing to deliver; it is the machine declining auctions it does not believe will clear your bar. That makes an unspent budget one of the most informative numbers in the dashboard, because it tells you the math does not close somewhere. The job is to find where.

The throttle: your ROI target behaves like a price cap

Think of the ROI target as a filter on every auction the system considers. Set it near what your shop has historically achieved and the system finds plenty of opportunities it can accept. Set it far above, and it sits out most of the day, which prints as low utilization.

Two details make this bite harder than sellers expect:

• The recommended target comes from your own history, roughly your product GMV against the ad spend behind it. It is not a generic industry number. Overriding it upward means demanding a return your own track record does not yet support.
• If you are comparing against your old shopping-ad ROAS, the windows differ. Older formats often credit purchases across multiple days; GMV Max counts a one-day window (we break down the full attribution model in why your creatives show zero orders). The same real performance prints a lower number under a shorter window, so a target carried over from a longer-window format is silently inflated.

The diagnosis tree

Work these in order. Each level is cheaper to fix than the one after it.

1. Target versus recommendation. If your target sits above the recommended value, that alone explains low spend. Move toward the recommendation. If you cannot accept the recommended ROI, the honest fixes are on the product side: pricing, bundle, offer. A target the system can believe in is the entry fee; everything below assumes you have paid it.

2. Creative pool size. A hand-picked pool of five videos gives the system almost nothing to explore. Switch to auto-select, which explores all existing and future videos with no cap; authorize every TikTok account that holds usable posts; add strong videos via their codes. TikTok's best practices are unambiguous here: more product-anchored videos, greater GMV potential.

3. Creative fatigue. Watch daily GMV at the video level. If your top creatives' contribution is decaying week over week, the pool is stale, and the system responds by spending less. As a working cadence, plan on introducing fresh variants weekly rather than waiting for the decay to show up in campaign totals, because by then you have already paid for the stale week. Fresh variants restore delivery; identical re-uploads do not, since the system recognizes them as the same creative with the same tired signal.

4. Product competitiveness. If the target is sane, the pool is deep, and the creatives are fresh, but spend still will not move, the auction is telling you the offer loses to what is next to it. No amount of creative volume outruns a losing price.

Cold start on fresh accounts: a two-step playbook

New marketing accounts hit a wall the tree above does not fully explain: their creatives have no performance history, so in a shared campaign they race veteran creatives that arrive with months of accumulated signal, and they lose the auction before spending a cent. From working with sellers in this exact spot, the pattern that works is separation, then promotion.

Step one: a dedicated test campaign. Put the new accounts' creatives in their own GMV Max campaign with its own budget, so they compete against each other instead of against a thousand veterans. Set the ROI target meaningfully below your main campaign, as a working rule around 70 to 80 percent of it, because the goal of this pool is spend and learning, not efficiency. Run it for about a week and let every creative touch real traffic. Before any of this, it helps to publish a handful of organic posts from the account first, so it enters the auction as something other than a blank.

Step two: promote the winners. Move the creatives that found traction into your main campaign, where their week of accumulated signal lets them hold their own. Amplify one or two high-conviction creatives at a time rather than the whole batch, and let each amplification run its full course before judging it: a half-day of data on a half-spent test tells you nothing, and cutting it early wastes the spend that produced the partial signal.

Every conversion the test pool generates is banked into those creatives' history, which is precisely what makes step two work. The same logic cuts the other way, which is why amplifying a creative you do not believe in is worse than not amplifying at all: poor results are banked too.

One production note that costs nothing: open on the product. A creative that shows the product and states the core claim in the first three seconds gets classified into the right category pool faster, because ad systems label a video from what they can see and hear, not from the caption. Category-mismatched creatives leak spend into irrelevant auctions, which drags the very ROI number the throttle watches.

Every branch ends at the same place

Look back at the tree. Fixing the target buys you delivery this week. But fatigue wants fresh variants continuously, the auction rewards pools of roughly 20 varied videos per product, and every cold start wants disposable test volume that most teams cannot afford to film. Whatever branch you entered from, you exit at the same door: the system wants more distinct, fresher creatives than a filming schedule can supply.

That is the gap Riffkit exists for. You start from a short video that already won, and Riffkit rebuilds its formula, the hook, the pacing, the beats, around your own product: new footage, product anchored on screen, no filming. One proven structure becomes a test pool, because each re-run can change the angle, the character, the opening, the language, or the ratio, and each render takes minutes. The test-campaign playbook above assumes you can afford to treat creatives as ammunition; this is how sellers afford it. If you work in an agent like Claude Code, the Riffkit skill runs the whole loop from one sentence per variant.

Signing up is free and includes enough free seconds for a first video, so the first creative in your test pool costs nothing to produce.

The short version

An unspent budget is the system telling you it cannot clear your bar. Check the target against the recommendation, then pool size, then freshness, then price, in that order. Separate fresh accounts into their own test campaign before asking them to race veterans, and promote winners with their signal attached. And build a production pipeline that treats 20 varied creatives per product as a starting line, because that is the real constraint the dashboard has been pointing at all along.