The repetition is_machine work

01

What is automated

The machine has two separate jobs in a campaign. Producing versions is automation: DCO & Templates produces versions from an approved template in the supported formats, and when a feed is connected and maintained, the versions update with it. Ordinary feed automation follows agreed rules, so this page does not call it AI.

The AI in a campaign is mainly the ad platforms' machine learning. It decides who sees the ad, where and at what price, based on the objective it is given. The Assets page explains how asset jobs are split.

02

Where the machine gets its objective

The platform optimises for whatever event it is given as the objective. We choose an event that reflects the business result and occurs often enough to learn from: a purchase, a lead or an earlier step such as a product page view. If the event is rare, the platform gets little signal.

Learning takes time. In our campaigns the first days' figures have not shown the campaign's real level, and a flight of only a few days has not had time to learn at all.* So budget moves are decided on an agreed review period and the objective, and a short campaign is planned so that there is time to learn.

03

How results are checked

Optimisation is judged against the objective that was set. If the objective is purchases or leads, we judge purchases or leads, and the click rate is a supporting metric. A platform can favour the asset or audience that delivers the target event most cheaply, which is not always the best result for the business. In our campaigns the weakest asset has taken most of the budget, even when another version earned more clicks.* Likewise, the platform has given most impressions to the youngest age groups while the oldest group clicked the most.* So we look at results by asset and by audience and limit delivery or budget where needed.

In our campaigns, when the same audience sees the same ad again and again, the click rate has fallen during the flight, and a single asset has tired quickly in both native and social.* So we make several versions from the start, watch the results and swap versions in time. Versions made from a template make the swap possible without designing each version from scratch.

04

What people approve

People decide the message, the objective, the limits and the brand voice. The template and its content rules are approved before use, and we agree which changes can go live without separate approval, such as a price updated from the feed. Product and audience data are put in order before the automation starts. The wrong message in a hundred versions is still the wrong message.

05

When to bring in the machine

Automation is worth it when more versions are needed than can be made by hand, the data behind them can be trusted and the objective is clear.

The use of generative AI in assets is agreed separately: what it is used for, who checks the output and what is never published without human approval.

* NØRR3's own campaign bank, campaign reports 2024–2026: each lesson recurs in campaigns across at least two industries and two advertisers.