Dynamic Video Ad Production: What Brands Should Commission

A price expires. Inventory disappears. A CTA changes with the buyer’s stage. Yet many brands still burn all three into the most expensive frames they produce.

That made sense when a video ad was a finished file. It makes less sense when the serving system can choose product information and interactive elements at impression time. In that environment, the useful deliverable is not merely an MP4. It is a controlled set of runtime-ready video assets.

This is not an argument for turning every brand film into a template. It is a procurement question: what should be locked, what may vary, and what must the production partner deliver so the combinations remain accurate?

Three platform moves expose the same production gap

On May 11, 2026, Amazon Ads introduced Dynamic TV Creative for Prime Video. Amazon says the capability can adjust an Interactive Video Ad’s format, CTA, headline and product details at impression time, using signals that include shopping and browsing activity, Prime Video engagement, product availability and geography. The initial release was for selected US advertisers, with broader expansion planned.

On May 20, Google said retailers would be able to upload videos to Merchant Center and have them distributed across Demand Gen according to real-time user interest. Adobe’s 2026 Advertising Creative documentation describes a more explicit assembly workflow: a video template, a product catalog and a feed template that maps catalog fields to creative elements. Its release notes say a feed can process up to 500 rows containing video assets and support structures such as start cards, end cards and overlays.

These products do different jobs, and vendor announcements do not prove that the whole market has already shifted. They do, however, reveal a mismatch: media systems can make decisions at runtime while many production scopes still describe only finished edits.

The failure mode is not “dynamic creative.” It is making the wrong things dynamic

A product price can come from a catalog. Product geometry should not. A CTA may change with the journey. A required claim qualifier should not quietly disappear. A SKU image may swap. The hand action demonstrating that SKU still has to be true.

Before production, separate two kinds of information:

Stable narrative: the brand promise, real product behavior, physical details, approved claims, performance, voice and story logic;
Volatile data: price, offer, inventory, SKU, location, language, product image, CTA, landing page and expiration date.

ONCE’s production rule is simple: spend live-action money on stable narrative and product truth. Put volatile data into a governed variable layer with explicit fallbacks. A price update should not require a full re-edit, and a bad feed row should never rewrite what the product is.

The ONCE Narrative Dependency × Data Volatility Matrix

Score each element on two dimensions: how much it depends on continuous performance and storytelling, and how likely its information is to change during the campaign.

Condition	Best production form	Typical content
High narrative dependency, low volatility	Locked hero master	Real use, brand idea, product proof, performance and core sound
High narrative dependency, high volatility	Locked story with configurable overlays or end cards	One story serving different markets, SKUs, offers or CTAs
Low narrative dependency, high volatility	Dynamic template plus product feed	Product selection, price, stock status, local offers and landing pages
Low narrative dependency, low volatility	Fixed cutdowns or modular edits	Evergreen benefits, simple feature cards and stable packaging

The matrix prevents two expensive mistakes: fragmenting an emotional brand film that needs continuity, or baking short-lived commercial data into a costly master. The first weakens the story. The second turns every offer change into post-production.

Replace the file count with a four-layer Runtime-Ready Video Contract

A scope for dynamic video ad production should define four layers, not just the number of 16:9, 9:16 and 1:1 exports.

Locked master layer: approved live action, product behavior, narrative structure, voice and non-editable claims. Give every master an asset ID, version, eligible SKUs and rights window.
Variable-slot layer: define where product imagery, headline, price, offer, CTA, language and end card may change. Specify character limits, duration, safe zones, opacity, transitions and default states.
Feed contract: document SKU ID, asset ID, field types, allowed values, currency, date format, URL, inventory state, market, rights expiry and prohibited combinations when data is missing.
Fallback and QA layer: provide default creative and state what happens when an item is out of stock, an offer expires, an image fails, a price is blank or a URL breaks. Test devices, aspect ratios, captions, audio and visual overlap.

This structure also assigns accountability. Production owns the masters, slots and visual QA. Commerce and media teams own feed accuracy and updates. Brand and legal own claims, rights and approval rules. If a layer has no owner, runtime scale simply multiplies the error.

Shoot for recombination, not only for today’s edit

Runtime-ready production does not lower the bar for cinematography. It adds requirements for composition and separation:

leave intentional negative space for headlines, product cards and CTAs;
protect the action across horizontal, vertical and square crops;
capture longer action handles so modules can connect without jump cuts;
separate product, shadow, reflection and environmental sound where practical;
deliver voice, music, ambience and captions as controllable layers, not only a mixed track;
slate shots with asset and SKU IDs so similar packages cannot be misassigned later.

This is why we treat the creative workflow itself as a client deliverable. The asset must keep working when the offer, market or SKU changes. Its source, permissions and expiration should also sit inside a traceable provenance and rights record.

Approve runnable combinations, not theoretical variants

Dynamic systems produce seductive multiplication: five templates times 100 products becomes “500 ads.” That number is meaningless if half the combinations overflow, link to the wrong page or display an expired offer.

Use operational acceptance metrics instead:

Approved runnable combinations: variants that pass brand, product, layout, URL and rights checks;
Invalid-combination rate: the share blocked by stockouts, expiry, empty fields, wrong images or rule conflicts;
Cost per approved combination: production, templates, feed preparation, QA and review divided by cleared variants;
Swap latency: elapsed time from a price, SKU or CTA change to an approved live replacement;
Fallback success rate: whether the default creative remains accurate when source data fails.

Start with a full-combination pilot, not a huge rollout: three SKUs, two end cards and two aspect ratios create 12 combinations. Inspect every one for overlap, captions, sound, price, stock, link and rights. Scale only after the exception rules work.

When a fixed film is still the right purchase

Brand manifestos, complex performances, continuous emotion, precise comic timing and messages that require an intact legal context usually deserve a fixed master. Dynamic layers should not alter product form, functional demonstrations, approved claims or the evidence supporting them. Availability also varies by account, market and format, so confirm the platform workflow before building around it.

The point of runtime video is not to abolish finished films. It is to put change in the correct layer. On the next brief, do not begin with “How many videos?” Label every element as locked, replaceable, condition-triggered or fallback. Then decide what deserves a shoot and what belongs in the feed.

Your most expensive frames should carry your most durable value—not your shortest-lived price.