Math Meets Marketing with Statistical Models
Companies that use social media advertising are also increasing their reliance on statistical modeling to determine which ads to serve on consumers’ pages, according to Colette Nataf, CEO of Lightning AI, a performance advertising software provider. Companies can use purchase data from before and after a Facebook campaign, for example, to determine the effectiveness of the ad. Even if people don’t click on the ad and make an immediate purchase, companies can still benefit from the additional brand awareness.
The models are also gaining use in helping companies optimize customer conversion and retention, Moorthy says. Historically, statistical modeling has been used more on the conversion side, but the retention side has grown.
To get it into these use cases, though, artificial intelligence has been the key, experts agree.
As AI becomes more integral in statistical modeling, it is enabling marketers to quickly analyze huge amounts of data to determine the effectiveness of email campaigns, online advertising, webinars, and other digital marketing efforts.
“Marketing today is supported by the big dataset,” says Esther Galantowicz, senior marketing manager at Saggezza, a provider of software and services for Big Data and analytics, business process optimization, network management, and service development.
“Taking a data-first approach and experimenting with ad groups to different demographics will give marketers a much more nuanced perspective on how different creative is resonating with different groups of people,” Ziech-Lopez adds. “Overall, combining these perspectives results in more targeted and effective ads for the marketer and a much better experience for consumers—something we can all appreciate.”
Galantowicz says that monitoring data with an eye to changes in campaign performance over different time frames enables digital marketers to quickly adapt their efforts. Similarly, she recommends that marketers carefully monitor the time consumers spend on their web pages and adjust accordingly. If a change to a website means someone is spending more time engaging with the brand, that is a good sign, unless that reflects the fact that they can’t find what they want, are getting hung up in the checkout process, or are encountering other sources of friction. In those instances, time is limited. Consumers might be willing to spend more time early on, but if the problem persists, they will likely abandon the company for a competitor offering a superior online experience.
KEEP MODELS DYNAMIC
Langers recommends integrating analytics for marketing, sales, and customer service to determine all components of customer journeys. Once a baseline is established, each step of the funnel can be examined and measured using statistical insights.
But to do so, statistical models need to be dynamic so that they evolve with consumers and technology, experts agree. The statistical model that produced excellent results a couple of years ago is probably no longer valid today, and the model that works today likely will not be so reliable a year from now, Langers says.
As evidence he points to the travel and hospitality industry, which has seen customers drop off dramatically in the past few months. By keeping a statistical model dynamic, marketers in these industries could have better forecast the fallout as soon as the COVID-19 pandemic started taking hold in the spring. By the same token, any model that would include pandemic-related data likely will not be relevant by 2022.
Similarly, Langers and others say statistical modeling should be evaluated carefully to look for surprising results that could be tied to specific one-time sales spikes or other anomalous underlying data. “You have to look at the details. Is there a fat finger?” Langers asks, raising the specter of data entry errors.
“The world is an evolving place,” Moorthy adds. “You need to constantly retrain your models.”
Langers also recommends discussing results of statistical modeling with management and other stakeholders before reacting to help ensure that surprising results do indeed reflect reality. “It takes collaboration.”
As noted, AI is the key here. Machine learning is becoming more integral for maintaining the accuracy of statistical models as technologies evolve, enabling models to continue to learn to produce more accurate results, Erande points out.
And that will only continue to be the case amid the quickly advancing technology landscape, experts agree. Adopting now will allow companies to get in on the ground floor and move up as the pace of innovation quickens.
Phillip Britt is a freelance writer based in the Chicago area. He can be reached at firstname.lastname@example.org.