The ability to accurately predict customer behavior and intervene in a timely way to influence that behaviour relies on leveraging a mixture of data and analytics techniques, according to a new report

The report titled, The Shift to Predictive Engagement, recommends using predictive, real-time and reactive methods in conjunction. That’s because without historical data to compare against it is impossible to retrain algorithms.

Conducted by the Which-50 Digital Intelligence Unit in partnership with Genesys the report cites the use of analytics by full-service media buying agency, Audience Group, which implemented predictive engagement techniques last year.

In the Investigates report, director James McDonald told authors that only a year ago the agency was manually allocating its spend between advertising, SEO, premium video, data use, remarketing banners, and mobile. Each channel would have its own budget approval, and then the team would execute slavishly for those channels.

Audience Group now uses a predictive approach, which means staff spends less time channel planning and more audience planning.

“Now we can dynamically shift budgets across all those channels according to how they are performing,” says McDonald.

Many of the Audience Group’s statistical techniques rely on historical data to help build and train models.

“Without that, you cannot have a continuous predictive algorithm out there doing something for you. You’ve trained it on old data, and it’s making decisions based on real-time data,” he said.

“However, if something changes, you need to change the algorithm. If you don’t continue collecting historical data — which is reactive analytics, in a sense — you won’t be able to retrain those algorithms. You have to be able to work all the way through the continuum. There’s a place for each of them.”

Analysts and data experts the report’s authors spoke with warned of ethical considerations involved in the collection of training data for machine learning models, which can quickly become problematic when machines begin automating decisions without the adequate human oversight.

The report notes, using machine learning also requires new governance structures and ongoing monitoring to the models’ outputs for biases and fairness.

“We’ve all seen those algorithms that don’t work, that go rogue,” McDonald said. “Sentencing algorithms in the United States are biased towards older people who are white. Reoffender algorithms have cultural biases baked in. Your marketing and advertising — that’s even more so.”

“You have to be careful not to just rely on predictive or reactive engagement. You have to keep a watch over everything. Otherwise, you might not know if your algorithms are working or not. It’s important to have a check-up and look back every now and then.”

When it comes to implementing predictive technology, the report highlights the need to look beyond technology.

It recommends marketers develop processes to manage, share and validate data and the models it informs and provide continuous training for staff, ensuring they are equipped with the skills and information to maximise conversions, sales, and customer satisfaction.

Previous post

VMware Announces Intent to Acquire Avi Networks

Next post

Worldwide CRM Software Market Grew 15.6 Per Cent in 2018, Gartner Says