What a Streaming TV Media Plan Looks Like at $50K a Month

The split, and why it isn't even

A reasonable first-month allocation: $22,000 to Prime Video (the most mature, best-measured supply source), $12,000 to Fire TV Channels (broad, cost-efficient reach, including a device-level slice for Feature Rotator and sponsored tiles), $6,000 to Twitch (a smaller, deliberate test against a specific category-fit hypothesis, not an even fourth share), and $10,000 to Disney's DRAX bundle, split further into $6,000 Hulu and $4,000 Disney+ given Hulu's larger ad-supported base. Notably absent from a first month at this budget: Netflix and live sports. Both are genuinely valuable, but a $50,000 budget spread across six or seven supply sources instead of four produces too little scale per line item to read any of them with confidence in month one.

The logic behind weighting Prime Video heaviest isn't brand loyalty to Amazon's own inventory — it's that Prime Video carries the deepest, most mature measurement tooling of any supply source in this plan, including pause ads and interactive formats with published engagement benchmarks. Putting the largest share of a first serious month's budget behind the best-measured inventory means the plan's early lessons come from the line item most capable of teaching them clearly.

The impression math

At representative CPMs — $30 for Prime Video, $22 for Fire TV Channels, $28 for Hulu, $35 for Disney+, and $18 for Twitch's blended display-and-video mix — the $50,000 buys roughly 733,000 Prime Video impressions, 545,000 Fire TV impressions, 214,000 Hulu impressions, 114,000 Disney+ impressions and 333,000 Twitch impressions: just under 2 million gross impressions before deduplication and frequency capping trim that toward a genuine unique-household reach figure through Amazon's authenticated graph, which reaches over 90% of US households deterministically.

These CPM assumptions are planning inputs, not guarantees — Amazon publishes no fixed rate card for any of this inventory, and actual delivery will move with auction dynamics, seasonality and how competitive your specific bids are. Treat the impression totals above as the order of magnitude to expect, and revise the whole model against real in-platform delivery once the campaign has run for at least a full week.

What to expect, honestly

Set expectations against our own book's actual pattern rather than an optimistic guess. Across 27 advertisers in our own Amazon DSP book over a 31-day window this summer, connected-TV and streaming inventory overall returned 0.77x on strict last-click attribution, with the highest new-to-brand rate on the report at 56.3%, and streaming specifically ran at a $15.51 blended CPM. A $50,000 first month at those benchmarks should not be judged on same-week ROAS — the honest scoreboard for month one is completion rate by supply source, branded search movement, and whether frequency stayed within the cap you set, not attributed sales in the DSP dashboard.

What to do when the month doesn't go as planned

If one supply source dramatically underdelivers against its allocated share, check bid competitiveness and market eligibility before reallocating — a newer or third-party supply source competing against Amazon's own well-established inventory in the same auction can simply be outbid rather than genuinely audience-mismatched. If completion rate is weak across the board rather than on one specific line, that's more likely a creative problem than a targeting one — check length and hook against the platform's non-skippable, sound-on viewing context before touching audience settings. Resist the urge to declare the whole test a failure inside the first two or three weeks; Amazon's own audience models are still resolving during that window, and pulling spend early throws away the learning along with the budget.

If frequency reporting shows unexpected spikes on any single supply source, check whether cross-publisher capping through the authenticated graph is actually enabled for the campaign — a plan running four supply sources under independent per-platform caps rather than one household-level cap can quietly over-serve the same viewers across the mix, which shows up first as a completion-rate dip rather than as an obvious error.

The common mistake

The mistake is splitting $50,000 evenly across every named supply source available, on the theory that even coverage is the safest choice. It isn't — an even split across six or seven line items at this budget produces too little scale on any single one to separate a genuine signal from noise, and it usually means every line item gets cut or kept based on too little data to trust either decision. reMKTR builds a first-month streaming plan around three or four supply sources with genuine scale, adds the rest in month two once the first set has produced a real read, and resists the instinct to test everything available at once just because the DSP makes it technically possible.