Video Advertising Budgets: How Much Before It Is Measurable

Why there's a real floor, not just a soft suggestion

Individual account returns inside our own DSP book, across 27 advertisers over a 31-day window this summer, ranged from 0.85x to 18.03x with a median of 4.30x — a portfolio ROAS blends that whole range into one number, but any single account's own result can sit far from the average purely on statistical noise at low volume. A campaign spending $3,000 a month generates too few impressions and too few outcomes to distinguish a genuinely underperforming line item from an unlucky small sample, on any advertising channel, not just this one. The floor isn't arbitrary caution — it's the point below which the data literally can't answer the question being asked of it.

The same logic applies to attribution windows, not just impression counts. A campaign that hasn't run long enough or spent enough to generate a meaningful volume of purchases will show a noisy, unstable ROAS from one week to the next regardless of the format — a thinly-funded campaign can look excellent one week and terrible the next purely from sample-size variance, with nothing about the underlying targeting or creative having changed at all. Reading that noise as a real trend, in either direction, is how under-funded tests produce both false confidence and false alarm in roughly equal measure.

Where the number comes from

Amazon's own published minimums are one input: a $10,000 recommendation for a self-serve DSP campaign, $50,000 required for managed service. Our own internal guidance, developed from watching first-time video accounts across the portfolio, sets the practical floor at roughly $10,000-$15,000 a month before a test reliably produces a trustworthy read — close to Amazon's own recommendation, arrived at independently from watching what actually separates a useful first month from an inconclusive one.

It's also worth being honest that these two numbers agreeing this closely is somewhat coincidental rather than because one was derived from the other — Amazon's figure reflects what its own auction and delivery systems need to function well; ours reflects what a testing window needs to produce a signal worth trusting. That they land in the same range is a useful confirmation, not a single source of truth either figure depends on.

A worked example

At $12,000 a month split across two mature supply sources — say $8,000 Prime Video and $4,000 Fire TV Channels — representative CPMs of $30 and $22 buy roughly 267,000 and 182,000 impressions respectively, comfortably enough scale to produce a completion-rate and branded-search read inside a month. At $3,000 spread the same way, the same math produces roughly 67,000 and 45,000 impressions — technically a campaign, but thin enough that a single unusual week can swing the reported result well outside what the true underlying performance actually is.

Extend the comparison one step further: at $25,000 a month, the same math produces roughly 556,000 and 379,000 impressions across the two supply sources — genuinely diversified scale that supports not just a completion-rate and branded-search read but a reasonably confident supply-source comparison between the two. The relationship isn't linear in value; doubling a budget that's already above the floor buys more confidence and more testing flexibility, while doubling a budget that starts below the floor mostly buys a slightly-less-noisy version of an already unreliable read.

What to do below the floor

If the available budget sits under $10,000-$15,000 a month, the honest options are: concentrate the whole budget on one supply source rather than splitting it across several, which at least gets one line item to a scale worth reading; move to lower-cost OLV, where the same dollars buy meaningfully more impressions and a trustworthy read is achievable at a smaller budget; or hold off on video entirely until the budget grows, rather than running a test destined to produce an inconclusive result that gets misread as a verdict on the format.

Whichever route is chosen, set the same expectation with any stakeholder footing the budget: a below-floor test is exploratory, not a verdict, and framing it that way up front avoids a small test's ambiguous result being read later as definitive proof the channel doesn't work.

The common mistake

The mistake is running a thin test, getting an ambiguous result, and treating that ambiguous result as a real answer about whether video works for the brand — when the actual problem was that the test was never funded at a scale capable of answering the question. It's one of the few honest answers in this category that can cost a sale: telling a prospect their budget is too small for a trustworthy streaming test, before taking their money for one anyway. reMKTR states this threshold plainly on every video sales call, because a client who later feels misled about what a small test could actually prove is worse for the relationship than losing the deal at the outset.