Behr Paint - AI Color Recommendation Experience

Behr partnered with us to create an interactive AI ad experience using IBM Watson. The goal was to help people choose a paint color for a specific room while also giving practical guidance about interior paint projects — all within the constraints of an ad unit.

The hard part wasn’t recommending colors. It was dealing with how people actually talk about them.

Most users don’t think in SKUs or color families. They describe what they want in emotional terms — calm, cozy, bright, dramatic. The challenge was translating that kind of subjective language into something a system could reliably work with.

Outcome

The result was an AI-driven experience that felt more conversational than transactional, and more useful than a typical ad. It balanced emotional input with practical guidance while working within tight format constraints.

The initial version of the campaign won:

• Adweek’s 2019 Project Isaac Award (Digital Transformation)

• 2019 Internet Advertising Competition Awards, including Best of Show

The core system was later extended into multiple subsequent versions.

My Role

I designed the conversation flow and underlying behavior of the experience, including:

• Dialogue structure and user pathways

• A training matrix and Q&A set for the chatbot

• Language frameworks that mapped emotional intent to paint attributes

• Responses for edge cases and out-of-scope questions

Much of the training content was grounded in Behr’s existing materials, supplemented with large volumes of customer service logs to reflect real questions and phrasing.

Designing for Mood

To support tone analysis, I authored non-consumer-facing descriptions for over 100 paint colors. These descriptions weren’t meant to sell the color — they were written specifically so Watson could better understand the emotional character of each option.

I looked at how other industries describe color (including paint and nail polish) to understand how mood is commonly expressed, then adapted that language into a consistent internal system the model could use.

This created a layer where:

• Users interacted with simple mood words

• The system worked with richer semantic signals behind the scenes

Conversation Flow

The experience began when a user selected a room type. From there:

• Users chose mood-related words to describe what they were looking for

• Each word triggered backend sentences written to reinforce the intended emotional signal

• The system generated color recommendations aligned to that mood

Users could then:

• Expand recommendations into swatches

• Visit Behr’s site for deeper exploration

• Upload a photo of their space

• Save or share colors via email or social

Click each frame to see full image.

Recommendations expand into Color Swatches

Recommendations shared on social

Users ask questions about interior paint project

At any point, users could ask freeform questions about their paint project. I wrote responses for common paint-related questions as well as fallback replies for anything outside the experience’s scope, with the goal of staying helpful without overreaching.

A small sample of responses I wrote in the event someone asked something outside of this AI experience’s scope.

Interested in more? See a sample video here.