Promotions remain one of the most widely used tools to lift demand. While on the surface, promotions may seem a relatively simple exercise, if you really want to make the process of promotions effective and precise, there is a long list of nuances, like:

• Context of promotion.
• Time of promotion
• Articulating the promotion by using appropriate semiotics and imagery.
• Context of promotion (like an upgrade, upsell, conversion etc.).
• Language and neurological triggers to convince consumers.
• Pricing and discounting based on brand loyalty.
• Best combination of pricing and language to switch a consumer 
• Parameter tweaking based on consumer and predicted volume.
• Optimizing price based on conversion.

While the list above may seem long, it is not even exhaustive. The strategy of promotions goes much beyond the template and words embedded therein that you see in your email or flier. 

Algorithms to the rescue

Traditionally, AI and other mathematical models have been used to address individual challenges. Examples of methodologies used to address nuances of promotions are:

• NLP for choice of language, metaphor, semiotics, imagery, and of the promotion
• Context scoring in offering a promotion
• Dynamic pricing
• Clustering and classification of discounts based on success
• Factor analysis of parameters of a discount

The challenges in translating some aspects into models can best be illustrated using the brand loyalty parameter. To leverage brand loyalty data in your promotions, you need to translate the data for each customer (or segment) into the customer’s personality traits. This can be done by leveraging frameworks like the five-factor personality model, but it needs a good depth of human expertise.

So, the gist is that even if the algorithms can help, a broad basket of algorithms needs to be leveraged. But the most critical aspect is to thread the output from all these models into one cohesive picture that can be used to define the strategy. And this is where deep learning can help.

Deep learning as the master algorithm

As mentioned in another article, Deep Learning for Dynamic Pricing, The traditional basket of methodologies to fine-tune promotions leveraged today have their roots in decades-old methodologies (except for the NLP methods). The process of leveraging multiple algorithms and then threading an analysis together is from a time when practical applications of neural networks were far from feasible. 

Most of these solutions could not capture all the factors because it was impossible. The problem is they still do not capture it all in an age where it is possible to incorporate most of these factors precisely. So, like the approach mentioned in that article, we can use the master child approach of using the models mentioned above as child models to quantify factors and then feed them into the parent model.

But if you design the algorithms hierarchy optimally, you don’t have to worry about running the child algorithms yourself. You hear all the jazz around LLMs, but when it comes to applications, you will see the applications centered around automation. When I used to publish a video series around Generative AI, in one of those videos, I suggested an approach where you could use LLMs to actually fine-tune and optimize a portfolio of models. This is a perfect example of the application of that approach.

The deep learning model can not only help fine-tune the models but can also be trained to evaluate the efficacy and weightage of the outputs from each child model so that it can thread together a better promotions strategy.

The opportunities to change the paradigm of marketing analytics are knocking. You just need to keep your house ready so that you can let them in.

References:

• Promotional Marketing, Mullin, Routledge
• Event Marketing, Preston, Wiley
• Sales Promotions Decision Making, Barnes, Business Expert Press
• Marketing Metrics, Farris, Pearson