Most global production scopes still separate deliverables by placement: one 30-second landscape film, several vertical cutdowns, a set of display units, and perhaps an audio spot commissioned elsewhere. Media buying is beginning to collapse those boxes. If production continues to price only by file count, a brand can receive dozens of assets that technically serve but no longer feel like the same brand.
On September 29, 2026, Amazon Ads announced DVA+, a campaign type designed to reach streaming TV, online video, display and audio inventory from one workflow. Advertisers can begin with an objective, products or a brand URL, review AI-generated creative, set a budget and then add more direct controls when needed. The important limitation is timing: DVA+ was announced in a limited U.S. phase, open beta was expected to begin in late October 2026, international expansion was planned, and audio availability was still rolling out. This is evidence of Amazon's product direction—not proof that every channel or market has already converged.
The production brief now begins before the media format
In a traditional workflow, the team selects placements and then commissions files for each one. DVA+ reverses part of that sequence: the advertiser supplies an objective, product or brand source, creative options and budget, while the system can choose formats and inventory. The completion test can no longer be “all sizes delivered.” The useful question is whether the promise survives when a film becomes vertical, plays without sound, compresses to six seconds, freezes into one frame or disappears into audio only.
The format differences are material. Amazon's online video specifications cover 16:9 and selected 9:16 environments and recommend several durations from six to 60 seconds. Its streaming TV specifications describe full-screen, non-skippable 16:9 inventory with supply-specific durations. Standard audio advertising must name the brand, communicate the main benefit and end with one clear action without relying on a picture. A master film cannot solve viewing distance, silent comprehension, glance speed and audio-only storytelling through mechanical crops alone.
The ONCE Four-Environment Anchor Map
Before scripts are locked, map what must remain recognizable in four different experiences. This is not another size matrix. It is a creative continuity map: one promise travels across environments, while each environment uses the senses available to it.
| Environment | How attention works | Anchor to preserve | Production component |
|---|---|---|---|
| Streaming TV | Lean-back, full sequence, usually sound-on | Brand world, cause and effect, sonic memory | Wide composition, legible product action, dedicated mix |
| Online video | Short, sometimes muted, horizontal or vertical | First-two-second recognition and silent benefit | Early product reveal, motion hook, protected vertical framing |
| Display | One-glance decision | Product silhouette, one reason, one action | Key stills, clean backgrounds, separable layers |
| Audio | No screen and divided attention | Brand name, usage context, one action | Sonic cue, voice versions, split music and effects |
Across all four, define five invariants: one core promise, one recognizable product action, a visual recognition kit, a sonic recognition kit, and a controlled set of claims and disclosures. The platform may change length, layout and pacing. It should not be asked to invent those five foundations.
Make sound a pre-production decision
Many product brands have detailed visual rules but no reusable sonic system. That becomes a production gap when the campaign enters living rooms and audio inventory. A useful scope should specify brand-name pronunciation, voice character, a repeatable sonic cue, product-operation sounds, music permissions, market-specific language rules, and separate voice, music and effects stems. Sound is not decoration added to the television cut; it is an asset that must also carry the idea when the screen is absent.
Give automation components it can combine safely
When creative can be generated from a product listing or brand URL, weak source material becomes a scaling problem. Production should therefore capture clean product views, isolated use actions, short single-benefit clips, backgrounds without baked-in copy, multiple protected compositions, approved caption lines and approved calls to action. Our earlier analysis of Amazon's component-based creative assembly focused on listings, display and short video. DVA+ extends the procurement question into television and audio.
Choose the production route by the cost of being wrong
The decision is not whether AI is “good enough” in the abstract. It is what happens when a generated variation is wrong.
- Platform-led variation: useful for stable product facts, simple benefits, long-tail SKUs and secondary versions supported by clear brand rules. Human review of claims, images, pronunciation and destination still remains necessary.
- Hybrid production: appropriate when a team needs test volume but cannot allow product proof, performance, visual recognition or voice to drift. Produce master components and anchors, then let the platform expand lengths, ratios and combinations.
- Authored production: appropriate for premium launches, regulated categories, technical demonstrations, licensed talent and emotionally specific storytelling. The platform can optimize distribution without becoming the origin of the central idea.
Contracts need to follow the same logic. Instead of specifying only a number of films, state whether likeness, voice, music, typefaces and third-party assets may be generated, cropped, translated, dubbed and distributed across media. Name the approver for new variants, the owner of claim updates and the rules for retiring outdated source assets.
Compare bids with cost per coherent environment
Cost per finished file rewards easy delivery and often hides weak reuse. ONCE uses a different comparison:
Cost per coherent environment = (master production + visual and sonic anchors + adaptations + review) ÷ environments that pass all four tests
The four tests are brand recognition, promise clarity, format compliance and a correct action path. A low-cost package may deliver 20 files yet work convincingly only in display and landscape video. A smaller master-asset package that supports all four environments can have the lower coherent-environment cost. This model does not pretend to forecast media return. It measures whether production has created usable, recognizable coverage.
Run the four-state test before the shoot
- Write the idea as a 30-second sound-on sequence and check the complete cause-and-effect story.
- Reduce it to six seconds, mute it, and confirm that the product appears early and the benefit remains understandable.
- Freeze it into one frame and check whether a glance reveals the product, one reason and one action.
- Turn off the screen and retain 15 seconds of audio. Confirm that the brand, context, benefit and action still make sense.
If any state fails, adding another export size will not fix the system. Return to the anchor map and change the promise, shot design or sound asset. Cross-format production is not one shoot generating more files. It is one shoot creating a controlled system that platforms can assemble without erasing the brand's judgment.
The action for the next budget meeting
Move the deliverables table off page one of the production brief. Replace it with the Four-Environment Anchor Map, the platform's allowed variables, the brand's invariants, generation rights and named approvers. Then add formats. Whether DVA+ reaches your market on Amazon's current timeline or another buying platform adopts a similar workflow, the brand will own portable production assets instead of waiting for a new interface to reveal what is missing.
Primary sources: Amazon Ads, “Reach new audiences using DVA+” (September 29, 2026); Amazon Ads, “Introducing Amazon Ads Agent” (September 29, 2026); Amazon Ads online video, streaming TV and audio specifications, accessed October 8, 2026. The Four-Environment Anchor Map, routing method and cost model are ONCE analysis, not Amazon guidance.
