What is the difference between DCO and creative automation?

Creative automation produces finished assets in bulk before launch, such as every size and language of a campaign, and those can be reviewed as files. DCO assembles the ad at the moment of serving from approved parts, based on feeds and signals. DCO needs stricter component rules, because each final combination is never seen before it runs.

How do you review thousands of DCO combinations?

You do not review every one. Approve each component, write rules for the pairs that interact, bind dependent components together, and render the extreme cases: longest text, every language, smallest size, empty feed. Then monitor live delivery for feed errors. Structural review scales in a way that one-by-one review never can.

Does AI make DCO better?

It removes the component bottleneck. DCO used to be limited by how many images and lines a team could produce and approve, so pools stayed small and repetitive. Generation makes large, varied, on-brand pools practical. It does not fix weak templates, bad feeds or unclear rules, which still decide whether combinations make sense.

AI Dynamic Creative Optimisation (DCO)

Dynamic creative optimisation with AI-made components.

Dynamic creative optimisation assembles ads at serving time from components, such as headlines, images, offers and calls to action, chosen by rules or algorithms from feeds and signals like product, location or weather. AI makes components cheap to produce in quantity. The risk moves to assembly: every combination must be lawful, on brand and coherent, so the work lies in template design, component rules and QA across combinations.

Templates, slots and feeds

A DCO template is a layout with slots: background, product image, headline, supporting line, price, offer, call to action, logo, legal line. Each slot is filled at serving time from a pool of approved components or from a feed. Product feeds supply names, images and prices; context feeds supply location, weather, time of day or stock levels; audience data selects the message. Design the template so anything that can land in a slot fits it: a character limit per headline in every language, image crops that hold at every size in the media plan, and a price field that copes with any currency and long numbers.

Rules and decisioning

Something decides which components appear together, and that logic needs an owner. Simple rules are explicit: rain in the forecast shows the waterproof range, a returning visitor sees the product they viewed, a store's stock level decides whether an offer runs. Algorithmic decisioning tests combinations and shifts delivery towards those that perform. Both need written guardrails: components that may never appear together, offers restricted to certain markets or audiences, and what happens when a feed is empty or stale. Keep the rules readable by a marketer rather than buried in an ad server configuration that one person understands.

Quality control across combinations

Five headlines, four images, three offers and three calls to action already make 180 possible ads, and nobody reviews 180 ads. QA has to be structural. Approve each component on its own, then check the pairs that interact: a headline promising free delivery beside an offer with delivery charges, a claim separated from the disclaimer it depends on, one product's image beside another's price. Bind dependent components so they always travel together, such as a claim and its qualifier. Render the extremes for review: longest headline with longest price, every language, smallest size. Then monitor live feeds, because a broken price feed publishes wrong prices at scale.

Brand safety of automated assembly

When software assembles the ad, nobody sees the final version before a customer does. Pre-approval of components therefore becomes the real control. Each component must stand alone as on brand and lawful, and the rules must block unsafe pairings, such as alcohol imagery beside messaging that could appeal to under-age audiences, or a sale price next to a luxury line that never discounts. Keep a log of which combinations actually served, so a complaint can be traced to a specific assembly. The studio produces component pools from one brief with product-lock and brand rules applied, and each component passes through the same approval workflow as any other asset before a combination is formed.

Updated 25 September 2026 · General information, not legal advice. Rules change, so check the current text with your legal team before relying on it.