Making changes

Cosabella optimizes digital designs with artificial intelligence

Retailers are tweaking everything from landing pages to images and fonts in their efforts to increase web conversion rates. Yet with hundreds of variables producing millions of combinations, it can be time-consuming and difficult to find the right match to produce optimal results.

While retailers have been able to use analytics to measure the performance of such campaigns, it has often taken time to obtain results. Luxury lingerie retailer Cosabella has been using a new system that offers near real-time insights on conversion rate optimization and allows users to make on-the-fly changes to campaigns to improve their effectiveness.

Cosabella, which has design headquarters in Miami and production headquarters in Italy, makes intimates, swimwear, shapewear and lounge wear for a global market. Marketing Director Courtney Connell says while the company recently expanded its online direct-to-consumer business, the industry has often struggled with how to bring emotion to the digital experience.

“Conversion optimization … [is] one of the most important things for any company selling online.”

Courtney Connell

Cosabella has been trying to revitalize the digital side of the brand by becoming more innovative in the colors they use and how they present the products. Discussions about optimization in the past couple of years started as a way to try to find some of the lowest hanging fruit to increase sales with little investment.

“Conversion optimization seemed like a no brainer to me,” Connell says, “and it’s one of the most important things for any company selling online.”

‘Automated’ thinking

Cosabella had used another third-party system in the past, but Connell says there was a growing need for more immediate results and feedback due to the rapidly changing environment. With a growing collection of 80 separate colors, it was becoming challenging to determine which combinations of colors and products could optimize digital marketing efforts.

“I needed to test all variables at one time,” Connell says. Ascend, a system from Sentient Technologies, “could combine all these into many different candidates without me having to make that decision.” Founded by data scientists who previously worked on what became Apple’s Siri, Sentient launched in 2014 with an “evolutionary computation” technology that brings artificial intelligence to retail. Ascend can help marketers accelerate testing speeds by as much as 100 percent and pick winning designs faster.

John Epstein, chief marketing officer at Sentient, says Ascend uses evolutionary algorithms and limits natural selection to give marketers an opportunity to test more campaigns and view their performance. He says as it becomes harder to “squeeze” more value out of customer acquisition, savvy marketers are turning to site and landing page optimization to increase the value of site visits. Traditional A/B testing can be time-consuming and doesn’t always qualify interactions with subtle changes, Epstein says. Ascend makes marketing teams more efficient by enabling them to test multiple ideas at once.

“We’ve really seen that whether it’s making changes to colors or headlines or page layouts or flow, you can easily achieve some significant increases in some of the core” business performance indicators, he says. Ascend’s artificial intelligence discovers not only which changes are most helpful but which combinations of changes can achieve the best results; Epstein says this can run into “millions” of potential design combinations for some companies.

Testing can all be done on a single page or even across multiple pages in a “funnel.” Each page is represented as a genome and simulates genetic operators like crossover and mutation to determine if parent variables are optimal. It then runs through the other combinations, testing each page only to the extent that it is possible to decide whether or not it is promising.

The technology enables thousands of pages of products, colors, subheads, fonts, images and layouts to be tested in a very short time. Ascend also learns over time which combinations of certain elements are effective and gradually focuses the search around the most promising designs, making its process even faster.

Epstein says these machine learning capabilities create “automated” thinking that can help marketers make the best decisions as quickly as possible. “It continually tests designs and learns, essentially testing thousands or millions, without having to test them all. It compresses the time to test by a dramatic amount,” says Epstein.

Ascend Cosabella
The technology enables thousands of pages of products, colors, subheads, fonts, images and layouts to be tested in a very short time.

Increasing conversions

Ascend users access a portal that allows them to set up and edit all changes themselves. Users outline the experiments, create aspects of the variables and then deploy it to obtain test results. Marketers send their designs through the portal, which then generates a report to show how candidates fare relative to the control. Epstein says users can even test how the placement of images or buttons can impact interaction and conversation rates.

Connell says that one month into the program, Cosabella has been able to select top candidates that deliver sales that are 35 percent better than the control. “In the end the final candidate would typically be higher,” she says. “We still don’t have the final data to be able to build those numbers yet but we’re seeing some awesome things.”

Digital marking firm Condorly recently employed Ascend to optimize the website of fast-growing fashion brand FlyPolar. The company ultimately delivered a 50-page report on changes that could deliver immediate impact, but there was too little time to run so many tests. Using Ascend to handle the administration and testing, they were able to implement just a few changes on each page in FlyPolar’s funnel and deliver 16 percent more checkout page users and 13 percent higher order value in only three weeks.

The system automates testing from end to end by evaluating the changes, interpreting the results and allowing marketing departments to spend more time focusing on the ideas that work. “We’re entering into an era of intelligent commerce,” says Andy Narayanan, vice president of intelligent commerce at Sentient, “where companies need smarter solutions to increase conversations across the entire user journey, from first click to checkout.”

Connell says Cosabella has been able to test many things beyond conversion rates and found some interesting revelations about other aspects of the business. The company recently began testing promo headers and found that of its messages of “Made in Italy” and “family-owned since 1983,” the latter was the most popular with customers.

Cosabella also plans to use Ascend to more aggressively optimize its mobile site and newsletter. Connell says the company will be able to continually learn how marketing efforts perform and find improvements in many aspects of the branding.

“I think we’ll be using it at all times,” she says. “We don’t see it as something that will just produce a few candidate [designs] and end there. We see ourselves constantly playing with new items.”

This article was published in the December 2016 issue of STORES Magazine.