Programmatic: What machines can (should) do better than humans

Ever since the advent of programmatic advertising, there have been repeated speculations in the market as to whether all buyers of advertising space will lose their jobs sooner or later as a result. The rapid development of machine learning for digital advertising is fueling this debate once again.

For Marius Rausch, VP & Managing Director Central Europe at AppNexus, the division of tasks is clear: "Humans and machines must each be able to play to their strengths. To achieve this, it must be made much easier for users of programmatic systems to define the really important goals and parameters so that algorithms can then effectively take over the micro-decisions."

While the intuition and experience of media buyers should still be at the heart of campaign planning, machine learning can take over a variety of data-dependent optimizations. For Marius Rausch, the following four areas should be left to programmatic algorithms and machine learning in the future:

1. pacing. Advertisers expect results. However, these are best distributed evenly over the campaign duration. Machine learning can ensure efficient, consistent targeting that takes into account time zones, market conditions for different types of inventory and many other trends and parameters. Manual optimization according to countless individual parameters, a subsequent analysis and finally the adjustment of the individual campaign modules still costs traders too much time.

2nd Discovery. From the very first moment of a campaign, machine learning can find the perfect balance between spending on delivering on inventory with proven good performance and allocating budget for exploring new domains. This is particularly helpful at the beginning of campaigns, when performance usually leaves a lot to be desired while traders develop their strategy with limited information.

3. supply path optimization. A major problem that has arisen in recent years as a result of header bidding is the flooding of the market with duplicate impressions. Individual advertising spaces are now being offered by different SSPs at the same time. This makes it much more difficult to find the optimal path to the desired impression. Even if a trader had the time to check the individual impressions manually, it would be impossible to find out which provider is ultimately the best. Machine learning can determine the most efficient path to a particular inventory, taking into account 3rd party costs and the quality of the inventory.

4. bid price optimization. Header bidding and the spread of different auction logics (first price vs. second price) have increased the complexity of the bidding strategy. Algorithms can implement the optimal strategy for each impression and react dynamically to the different auction types.

AppNexus has prepared a white paper entitled "The Future of Trading" to explain what the future of ad trading could look like and how the distribution of tasks between humans and machines is changing. Direct link (PDF)

AppNexus is an Internet technology company that enables and optimizes the real-time buying and selling of digital advertising.