The power of creative analytics in data-scarce scenarios
The use of data can be more intuitive when it is in abundance, but what can one do when it is scarce, such as in B2B Media Sales
An abundance of data does not necessarily translate to deep insights or better business outcomes. But it usually does create some low-hanging fruits. B2C businesses generally fall in this category and have abundant data. An e-commerce platform for example will have a wealth of customer purchase and activity tracking data. The use cases are also more intuitive, e.g. product recommendation, marketing campaign optimization, etc. But what about B2B businesses, where both the quantity and scale of customer data are low? These are tough nuts to crack! Some advanced and creative techniques can help here and create an edge.
B2B Movies and TV Series licensing is such an example. When you see a movie on a TV channel or on an SVOD platform that has usually been licensed by the content owner to the platform. While licensing can happen on an individual title basis, usually it is done as a package of multiple titles. The buyer is trying to maximize the overall quality of the package to maximize performance on their platform and the seller is trying to maximize the revenue from licensing. Generally, there will be certain titles that overperform and compensate for lower-performing titles, but, it is still a game of averages and not about a single title. The analytics implication of this is that the typical recommendation engines that work for end viewers do not work for B2B sales. But it is a B2B2C sale eventually and most of the information about the performance and behavior is coming from the “C”.
This creates an information asymmetry and places more power in the hands of the buyers as they are closer to the end consumer and have much richer data. The negotiating power in the hands of sellers is usually when they have an extremely high-performing title such as The Avengers and the buyer is willing to buy several other titles in the package just to have The Avengers be included. From a standpoint of being able to leverage analytics though, the balance of power is still in the favor of the buyer.
The last decade was also the golden era of content production, the amount of money that was spent on new content creation was unprecedented. The pendulum now is swinging back with the new economic environment and content spend is declining. In this scenario, the library content that has been produced over the last few decades will again become very important. And generally, there are thousands of library titles with most large sellers, and being able to utilize analytics to package the right titles for the buyers and price them correctly will be very important. There is often a negotiation power-play in the dealmaking for the B2B sale, however, there is an opportunity to create a true partnership. This can be done by curating title packages and pricing in a way that maximizes value for both sides. Using certain data and analytics techniques that are not often talked about, and likely the buyer isn’t thinking of them either, can create this opportunity. In this article, we will explore three themes and some examples of specific techniques that I have used in the past to help create a balance of negotiating power and a win-win situation for both the buyer and the seller
Get ‘statistically’ creative
Use hybrid metrics. There are two key analytics challenges in B2B Movie sales. First, there is a high affinity towards using only certain metrics such as TV ratings and Box Office. Second, while the emotional response of viewers was a big factor in a movie’s success, incorporation of that in the analysis is usually non-existent. For the first challenge, a good approach is to identify which metrics truly matter and then create a weighted measure. For example, the number of raters on IMDb is a far higher predictor of the movie’s success rather than the IMDb rating. So a 6.5 rated movie with 200,000 raters would perform better than a 7.0 rated movie with 50,000 raters for most buyers’ platforms. One of the drivers for this is that certain genres of movies have a higher skew of ratings on IMDb, for example, action movies are usually rated a lot higher than romantic comedies. And hence the number of raters is a more universal measure than the rating for predicting performance. For the most impactful variables, a weighted performance score can be created. For the second challenge, affinity scores based on the synopsis and genre of the movies are quite effective. This affinity score can then be combined with the performance score and created a hybrid metric that combines both subjective and objective information.
Leverage innovations from creative players. The affinity scores from the synopsis and the genre are a valuable addition to the objective metrics but not nuanced enough. They do not capture the end viewers’ emotions and responses. One of the best-performing indicators in this area are the Nanogenres. Even within a specific genre of movies, there can be wide variations in terms of specific themes, story arcs, and emotions generated. For example, an action movie can be a mindless comedy or a tense gripping movie or a wild stunts movie etc. Nanogenres provide this deep information and that creates a significant lift in performance prediction. In many instances, these rich hybrid performance scores help prove to the buyer that a certain package of movies, on the whole, is worth buying and not just a few high-performing titles. While mass solutions are usually the most visible ones in every industry certain niche players create innovative and specific techniques such as Nanogenres that are worth finding and leveraging.
Go back to basics
Higher order measures. Artificial intelligence is all the rage right now and certainly has a lot of potential. But for a vast number of business applications, traditional statistical measures can be very powerful when used correctly. For example, to analyze the worldwide performance of a movie there could be hundreds of data points from different platforms across the world. And all the performance measures would be on different scales and distribution patterns. The goal is to identify movies that are likely high performers based on these data points. As a first step, the performance measures can be normalized through simple indexing to the average or the median of a specific cut of data. But then it gets interesting. Using measures such as the average of indices would exclude movies that had very high performance in certain geographies and very low in others as the average would be mediocre. The machine learning algorithms would also fail because there isn’t enough of a pattern to pick. But Kurtosis works! It can help identify hidden gems that work very well in certain situations but also fail in others. I have often found that such second and third-degree statistical measures are sometimes more powerful and quicker to yield results.
Use probabilities. Another example of a powerful long-standing statistical technique is Monte Carlo simulations. As I had mentioned previously that movies are often bought as a package by many buyers, so what happens if one wants to estimate the value of a certain subset of those movies that the buyer was willing to pay? This is a perfect example of having an imperfect information set. Monte Carlo simulations were a perfect tool for filling the information gaps and balancing out the asymmetry between the buyer and the seller.
Be the anti-current
Challenge the existing ways of thinking. Every industry and company whether old or new has certain preconceived ways of working. Even if a wholesale change is not warranted there is always an opportunity to improve upon these. In B2B media sales there is a strong market preference for transacting on titles that have worked in the past, especially in old media such as linear TV channels. The ‘repeated sales’ often become a self-fulfilling prophecy. The buyers would buy the same titles repeatedly, and the sellers would also want to only sell those titles, and because viewers of the linear channels did not have much of a choice and they would keep watching some of these titles and hence creating an illusion that the end viewers do not want to try new titles but want to repeat watch tried and tested stuff. And the only new movies they would watch are the new releases. This cycle made for easy sales for a salesperson, however, created a situation of having lower overall revenue generation from the inventory the content creators had. Along with some progressive salespersons, we were able to use rigorous analytics to identify new opportunities and revenue streams. Data teams are in the business of serving our stakeholders, which also means helping them learn new ways and challenging existing ways of working where it makes sense.
Think averages in the world of hyper-personalization. In many B2B businesses, the concepts of hyper-personalization are not very effective. A lot of the focus of advanced analytics in the world of business is around improving offer targeting, better pricing promotion, anomaly detection etc. This means that algorithms are focused on hyper-personalization. In B2B sales however buyers are usually buying a package of products and hence they are buying an average and not an individually targeted proposition. This means that the optimization eventually has to be on the greatest good for the population as a whole and not narrowly defined segments of customers. Ensemble models that reach global maxima in this case compared to local maxima. The approach is to use a human-machine teaming. Where the models generate segment-specific recommendations but a human weighs the individual outputs subjectively and creates a final recommendation.
We discussed some very specific themes and specific examples in this article that applied to B2B media sales. While each specific industry and situation would warrant specific approaches and techniques; I have often found analytics to be most fun and rewarding in data-scare scenarios.
