Madison Reed Inc.’s product recommendation engine factors in at least a billion data points before suggesting a product to a shopper.

Artificial intelligence makes this possible for the hair color products retailer. The machine-learning algorithm factors in the answers to a 20-question hair coloring quiz more than 4 million shoppers have taken, along with more than 24,0000 product reviews of the retailer’s 50 SKUs. It also factors in a shopper’s repeat purchase rate, Net Promoter Score and hundreds of thousands of shopper and customer service agent interactions, says Dave King, chief technology officer at Madison Reed.

With all of this historical information, Madison Reed’s algorithm is the ultimate master colorist that can recommend hair coloring products, King says.

When Madison Reed first launched its product recommendation engine, it was coded by humans and did not factor in all of the data points that it does today. For machine learning and artificial intelligence to work, there needs to be enough information fed into the system to interpret new information and adjust its algorithm, King says.

In early 2016, after more than 1 million consumers took the hair color quiz and Madison Reed had their “hair profile,” the retailer upgraded its human-written code to a recommendation engine fueled by artificial intelligence, King says.

The algorithm will update and improve its recommendations even if it only benefits less than 1% of the retailer’s total customer base, King says. This helps to ensure that all shoppers will receive the best recommendations even if they don’t fall into a majority group.

And it works. 90% of consumers who start the hair color quiz finish it. And shoppers who go on to buy the recommended product have a customer satisfaction score that is on average 30-50% higher than a shopper who buys a product off of Madison-Reed.com and doesn’t take the quiz or buy the recommended product, King says.

As more shoppers take the quiz, write a product review and interact with the retailer, the algorithm may want to recommend changing the product it recommends to a certain group of shoppers. If this is the case, one of Madison Reed’s master colorists has to approve the change, King says. Colorists review the changes in “batch mode” about once a week.

Madison Reed also updates its quiz questions or adds questions based on shopper feedback. For example, the retailer recently added a question to the quiz asking what problems she has had in the past when coloring her hair, such as the color not covering all of her gray hairs or her hair color getting too brassy.

This information will help Madison Reed recommend a better product, a complementary product, an application tool or signal to Madison Reed to provide the shopper with additional instructions for application, King says.

This feedback also has spurred Madison Reed to develop new products, such as its Knockout color line that is formulated to be better at covering gray, he says. The retailer will now recommend this product when appropriate, such as shoppers with a certain hair color who have had problems covering gray.

Madison Reed built its recommendation engine in house, and the retailer currently has about 12 engineers and data scientists.

In addition to its online coloring quiz, the retailer also has an SMS and Facebook Messenger chat bot called Madi that takes the shopper through the quiz by asking her questions, sending her options and using image recognition to interpret a photo of the shopper’s hair.

In April 2017, Madison Reed began selling it products at Ulta Beauty stores and on Ulta.com. In Ulta stores, Madison Reed promotes its chat bot via signage and encourages shoppers to text a selfie to Madison Reed to find their hair match. Ulta is No. 119 in the Internet Retailer 2017 Top 500.

About 90% of consumers fill out the quiz on the Madison Reed e-commerce site. The remaining 10% answer the questions between SMS and the Facebook Messenger chat bot, with SMS generating the majority, King says.