Pieter van Geel
Head of Analytics, Conversion & Data Science
AI & Automation
Artificial Intelligence (AI) & automation are trending topics in most industries these days, and marketing is no exception. Both terms are being hyped in nearly every marketing story or presentation. We are overloaded with articles suggesting that nearly every agency should use AI, and that machines are currently taking over all of our jobs.
Although I surely won’t deny the global trend and huge potential of AI and automation, I would like to provide a more nuanced interpretation of how media and marketing companies should implement AI and automation.
The terms “artificial intelligence” and “automation” are often used interchangeably. These terms are used to describe tasks executed by “robots” and other machines that allow us to operate more efficiently and effectively. However, there are some pretty big differences between automated systems and AI machines.
Chiefly: Automation is software that follows pre-programmed “rules,” while artificial intelligence is designed to simulate human thinking.
Despite these differences, both AI and automation start with data. Data, of course, is an essential part of the strategy of most companies these days, and most organizations are rightly convinced of the potential of data for their business.
So how should a company begin? In order to exploit AI & automation, data needs to be organized and available, preferably in a central place within an organization; call it a Data Lake or Data Warehouse. To fill the Data Lake, fully-automated data ingestion and transfer processes should be in place in your organization. Unfortunately, this can be easier said than done, and many companies struggle with the development of creating a complete Data Lake, leading to some of the first hiccups a company might experience when seeking to exploit AI and automation.
Digital marketing data is rather easy to ingest into Data Lake, but when it comes to traditional media or client data, it becomes rather more difficult. However, in order to get a full overview of marketing performance, clients expect a single point of truth for all data. Unfortunately, there’s currently no one turnkey solution available to connect all of a company’s data sources, as market demand exceeds the current speed of development that organizations need order to become fully data driven.
Despite of these development challenges, a Data Lake is still crucial for any organization, especially since GDPR became effective as of May 2018. In order for organizations to become GDPR compliant, a Data Lake should be in place to be able to properly manage large sets of data, especially personal information.
The next step beyond data is automation. As automation follows a pre-programmed set of rules, it is advised to first automate everything possible before moving towards more advanced tasks like AI. All recurring tasks should be considered when setting the scope for automation to increase efficiency. Dashboarding and reporting tasks are especially ripe targets for automation in marketing organizations.
Organizations need to set up a team to manage the following tasks on the road to automation: data storage & data quality; accessibility and infrastructure; maintenance; roadmap and prioritization; and finally, visualization.
Make sure to build flexibility into these teams and structures: in order to develop in automation, experimentation is key to developing new insights and innovating quickly.
Regarding dashboarding and reporting, we advise the follow the following steps: First, start with descriptive dashboards based on historical data. Then, once these dashboards are operational, move toward building predictive dashboards. Once these are in place and used, follow with prescriptive dashboards. These will be the foundation for your automation efforts.
As artificial intelligence is designed to simulate human thinking, it should be the driver to make current tasks smarter and even disrupt our current business to increase effectiveness. In order for a machine to act as a human, it needs to be trained properly with the correct data.
The quality of the training data set needs to be complete and unbiased to create an algorithm that is capable of simulating a human task. For example, in order for an algorithm to identify the gender of a person on a photo, it needs sufficient input and data from photos from both males and females.
For more complex algorithms, it is important to define clear objectives. As humans are able to make complex decisions taking multiple objectives into account, algorithms needs clear KPI’s to optimize toward, or else they won’t deliver logical results on easy tasks.
Therefore, most organization should start with AI in a predefined part of the business, and with algorithms generated in silos. Currently, humans combine the results of these algorithms in order to exceed the insights generated by any single channels, and to get a holistic view of the business.
A final note: algorithms need to be transparent. As algorithms will take over our current tasks, we need to make sure that the quality of this task performance is equal or preferably better than the current manual results. Therefore, it is important to verify the results of the algorithms continuously over time.
Summarizing the above, AI & automation definitely have huge potential and might in the future take over (certain aspects of) our jobs. In the current stage of the development, however, they will create more work and more jobs to do! So we should not fear AI & automation, but rather embrace them in order to unlock their huge potential to make our lives more interesting and easy in the future.