The online retailer of hair color products introduced “Madi,” a bot on Facebook Messenger, that uses image recognition technology and an algorithm to match hair colors.

E-retailer Madison Reed Inc.’s chat bot is taking a good look at who it’s talking to.

The web-only retailer of hair coloring products launched an artificial intelligence bot this week for Facebook Inc.’s messaging app Messenger. The bot is programmed with image recognition software and an algorithm so it can accurately suggest products that coordinate with a person’s hair color, says Madison Reed CEO and Founder Amy Errett.

When a consumer visits Madison Reed’s website she takes a 12-question quiz to receive a product recommendation that fits what she is looking for and her hair’s natural color. The retailer has 40 colors from which to match a consumer’s hair. Since the retailer’s launch in July 2014, 1.4 million consumers have taken the color quiz, Errett says.

In addition to the matching quiz, a consumer can choose to chat with a hair color stylist via email or live chat about the best product for her. The customer service agents found that 15% of consumers were showing them photos of themselves in the chats either via uploading them or attaching them to the email to help describe their hair, which is why the retailer decided to program its color-matching algorithm into a bot and let it factor in pictures, Errett says.

Here’s how the messaging bot works: A consumer searches for “Madison Reed” in the Messenger app and the “Madi” bot will show up in search results. The consumer enters into a chat with “Madi”—the name of the retailer’s bot—in Facebook Messenger. The consumer takes a selfie with her smartphone’s camera and sends that picture within the chat to the bot. From the picture the bot is able to answer some of the questions it would have asked in the quiz, such as her current hair color and if her hair is curly or straight.


The bot asks the consumer if the hair color it recognized from the selfie is correct. Once the consumer says yes, the bot takes her through the same questions should would answer on the website, such as what percent of her hair color is gray. The bot displays possible answers, such as between 0-10% gray or 11-20% gray, so the consumer can tap the response buttons and not have to key in the answer.

After the bot finishes asking questions, it displays the best-matching hair color for the consumer, plus two alternatives. If the consumer wants more information about the result or wants to buy the product, she taps on the image the bot sent, and she is taken to that product detail page on Madison Reed’s mobile website. This is the same page a consumer would receive if she had taken the color quiz online.

If the consumer doesn’t like the options and types that into the message, the bot will ask if she would like to bring in one of the retailer’s 23 customer service agents, who are on hand during business hours. If she responds yes, the chat will then be handed off to a human—within Messenger—who will handle the consumer’s questions.

This “unhappy” path was a challenge for the retailer to program into the bot, Errett says. There are many ways a consumer can express that she doesn’t like the product result, and programming the bot to understand a consumer’s sentiment and how to respond took time, she says.

The retailer had more than 1,000 consumers interact with the bot in Messenger during a beta test before Madi’s official launch. Madison Reed had to train the bot to respond to specific questions and nontraditional conversational paths. For example, many consumers ask for “highlights” in their hair, however, Madison Reed does not offer this product. The retailer first had to teach the bot what “highlights” meant and then how to respond. Also, a consumer may take a selfie and the bot might not recognize her hair color because of a low-quality image. In such a case, the bot will suggest hints to the consumer such as, “make sure you are in a well-lit place.” If the consumer sends another image that isn’t good, the bot will forgo the image-recognition piece and take the consumer through the full-length quiz to get the information it needs, Errett says.


The retailer decided to deploy the bot on Facebook Messenger because many consumers find out about Madison Reed from the social network, Errett says. The retailer advertises on Facebook and now some of its ads will promote finding a perfect color match through the Facebook Messenger bot. If a consumer clicks on that ad, the Messenger app will open for a consumer to begin the interaction.

The retailer will track how many bot interactions result in a sale. Errett hopes the bot will have a similar conversion rate to its website after a consumer completes the color quiz, although she declined to reveal specifics. Errett would not disclose sales or growth of business.

The retailer created the bot in response to consumer feedback about sending pictures and to offer an alternative to get a color match while consumers are on their smartphone, which is how 55% of consumers interact with Madison Reed, Errett says. 90% of consumers who start the online quiz finish it, Errett says. She does not anticipate the bot will boost the quiz completion rate. Instead, the bot offers a consumer an interactive way to shop with Madison Reed but without having to call, Errett says.

The retailer built the bot from scratch mostly in house with about four or five people working periodically on the project for about a year, Errett says. Madison Reed used Facebook Messenger’s Application Program Interface for the bot to work, but they did not have to do anything special with the bot to launch it on the platform.

Now that the bot is live, Madison Reed will start promoting it on other marketing channels, including Facebook ads, Pandora Radio and TV commercials. Next, the retailer will launch the bot via text messaging. It will be the same interaction, except consumers can text replies to a number instead of using Facebook Messenger.


“Facebook Messenger was the best place to get our training wheels on and from there we’ll be able to go SMS,” Errett says. The SMS bot will be live before the end of the year, she says.