Harnessing the Power of Now to Predict the Future

Guest Post By William Rand

Editor’s note: William Rand, an assistant professor of business management at N.C. State University’s Poole College of Management, first introduced the term “nowcasting” — the idea that by using modern telecommunications, the internet of things, and social media it is possible to automatically, and in near-real-time, observe what millions of people are currently thinking, doing or believing–to the CIMS community at the organization’s 2016 annual meeting. This article, which is excerpted from the September/October issue of the CIMS Innovation Management Report, is based on Bill’s findings on the relationship between social media use and donations to the Red Cross during two recent weather disasters.

Imagine that you plan to launch a new product or service online in a few weeks—a new massively multiplayer online videogame, for instance—but first you want to get a sense of how many servers you will need up and running to support all new users.

You could commission a large consumer survey to estimate how many new users will purchase the game on the release date, and you can use data from pre-game sales to forecast the number of initial users, but these methods have their limitations. Alternatively, you can examine the frequency with which the game is being discussed on social media, and based on models built upon past launches predict the sales of the new product. This is an example of “nowcasting” in that you are predicting overall interest in the videogame right now.

Nowcasting Made Easier

We have known since the earliest days of marketing research that what consumers say to one another has orders of magnitude more effect on purchase decisions than anything an advertiser can do. However, for most of marketing history. it has been difficult and expensive to obtain insight into what consumers are discussing among themselves, which made nowcasting of consumer beliefs nearly impossible. The advent of the Internet, though, has made nowcasting much easier.

In the Internet’s early days, consumers needed a certain level of technical sophistication in order to write a blog post, or even write a review on epinions for that matter; however, when social media came along in the early 2000s, channels such as Facebook, Twitter, Instagram, and YouTube made it increasingly easy for consumers to make their opinions widely known.

Many marketing managers worry about social media because it seems uncontrollable, seeing only the downside to consumers commenting on and rating their brands. However, consumers were having all these conversations before the advent of social media, but on private channels that were unobservable by brands.

The growth of social media has not only given the average consumer the ability to express their opinions to much larger audiences, but it also gives brands the ability to nowcast those opinions and observe how other consumers react.

Hurricane Tweets

With Chen Wang, one of my students, and Cornell University professor Shawn Mankad, we recently worked with the American Red Cross and Teradata on a similar nowcasting project: the relationship between tweets about the major hurricanes Irene and Sandy and text-message donations the Red Cross received during those hurricanes.

Text message donations are made by texting a keyword to a certain number that applies a fixed sum (usually $5-$10) to the consumer’s bill and sends the donation in their name to a charity, in this case the Red Cross.

We learned that such tweets were in fact a good prediction of the donations the Red Cross would receive. This is because the primary driver of donations to the Red Cross and tweets about hurricanes is the same, namely the level of the population’s awareness of and interest in the disaster. Consequently, Twitter serves as a monitor that allows us to detect this level of awareness and interest.

However, we also found that the Red Cross’ own marketing had the most powerful effect on donations. Roughly twice per major disaster, the Red Cross texts donors who have previous donated to another disaster relief program to ask for a donation to the current relief program. The donations received as a result of these texts were much larger than the baseline donation activity we observed. We also observed that the Red Cross received more donations when it sent these text messages during periods of high Twitter interest in the disaster.

We used this information to build a model of the increased level of donations the Red Cross could expect to receive during any hour of the day. This model would first predict what the interest on Twitter about the disaster would be during the next 24 hours based on the past Twitter data. It would then predict the level of text message donations that could be expected as a result of that level of interest on Twitter. This predictive model could then be used to determine when to send the text message to coincide more closely with peak levels of interest in the disaster.

Put simply, imagine a green light and a red light: The model flashes the green light only when it believes the donations the Red Cross is likely to receive in the next 24 hours will be at or near a high level. In fact, we can build in a control knob that affects how often the green light flashes, based on how confident the model is in its own prediction.

If the control knob is set at a higher level, then the green light flashes less often. This means the model is very confident in its predictions, but it also has the possibility of missing a few peak times, i.e., it increases the number of false negatives. However, in the end this gives managers the crucial ability to control the number of times the green light will flash, because if they send out too many text messages during a disaster they will essentially be spamming their donors and are likely to incur a lot of “unsubs,” or users who ask to stop receiving text messages.

Learning More

This is just the first step in how we can use social media to improve marketing. Although in the current project our scope is limited to the overall volume of activity, we could also look at the content of the tweets being shared.

We made an initial investigation into this and found that geography had a huge effect on the topics being discussed. For instance, during Hurricane Sandy, around New York City people were concerned about stocking up on food and water and running out of clean water. However, in the Baltimore area, which was also affected by the storm, there was much more concern about wind damage and the destruction of housing.

Given this information, imagine that we change the language of a text message we send to people in different parts of the country. For instance, for donors near New York City we can send a text message that asks for a donation to assist with food and water, while we can text donors near Baltimore to contribute to temporary shelter of displaced victims of the storm.

Past marketing research says that if we can align advertising with consumer beliefs, we can obtain a higher conversion rate; that is, donors are more likely to give when the message resonates with them.

This is just one case study of how social media can be a powerful aid to making optimal marketing decisions. In fact, as we improve our models and abilities in this space, we can move toward a future of “real-time marketing”—marketing that is constantly being informed by what consumers are thinking and doing at that moment, and not based on surveys and traditional methods of research, which always lag actual promotions and advertising by weeks or months.

In other words, social media in the future may not just improve marketing but optimize it.


Comments are closed.