Mainak Ghosh on taking a logical approach to digital advertising
We live in a fast-evolving world. Companies are continually trying to improve their various operating strategies in order to stay in business. One key business strategy is advertising. How a company advertises its products has a direct impact on the decision-making of the prospective customer, and thus on its revenues.
This becomes more relevant with the emergence of digital advertising, which presents the company with an opportunity to engage with customers in a much more effective manner. However, the online advertising space is perhaps even more fast-moving. An inefficient advertising campaign might lead to loss of prospective business to competitors, and can also cost significant amounts of money and effort. How can campaigns can be optimised to increase business at minimum cost?
Let’s say that ‘Actuari’ is an online health insurance provider. It has recently received approval to launch its new flagship product. The chief marketing officer has 10 advertisement video options to choose from in order to market the product on social media. She wants to figure out which video will have the highest click-through rate (CTR), but she is also concerned about the costs of exploration involved in showing each target social media user all 10 videos and finding out which one they click.
“An inefficient advertising campaign might lead to loss of prospective business to competitors”
One approach to solving this problem is using Thompson sampling, a reinforcement learning algorithm. Reinforcement learning is an area of machine learning that trains computers via a system of reward and punishment. A video based on the results of a simulation of success rates of each video is selected first, and then it is checked in reality whether or not the video is clicked by the user. The simulation is then refined using real-time observation. This process is repeated until the video selection converges on a particular video.
The idea of this process is to perform the steps in real time. Refining each step using the emerging experience in each round leads to the selection of the best video in terms of CTR. However, the emerging experience of whether the selected video is actually clicked on will not be available until the campaign goes live on social media platforms. To address this, the real-world outcome can be mimicked using simulation techniques, and in such a way that there is an inherent ‘winner’ video. The Thompson sampling algorithm will finally spit out the optimised video – the inherent winner – at the end.
What would happen if the chief marketing officer decided to pick videos randomly, without applying any particular logic? It is perhaps surprising that total video clicks will almost double if she chooses Thompson sampling over random selection.
Hopefully, this highlights just how important advertising and marketing are. The quality of a product needs to be demonstrated in order for it to be chosen by consumers, and health insurance is no different. In the modern world, large signs on billboards are no longer the norm.
The order of appearance in online search engines can be the key driver in growing a customer base. While the cost of moving from rank 2 to rank 1 in order appearance can be costly, a company may find that this difference increases its consumer base to such a high degree that the additional revenue dwarfs such increased costs.
Mainak Ghosh is guest student editor