Speaking in Color

Read the campaign story

Sherwin-Williams Coil Coatings and Wunderman Thompson developed Speaking in Color, an interface that began with a spoken description rather than a paint number. The agency describes using language interpretation, image search and optical recognition to propose a palette. Users could then ask for adjustments. WPP dates the work to 2022; the present VML page carries the agency’s later identity. The intended professional context was architectural colour selection.

The creative premise connected a technical coatings business with the personal associations behind a design brief. A remembered place or atmosphere could become a starting point for exploration. The sources establish this experience, but do not establish that every output became a purchasable coating or a completed building specification. Nor do their descriptions independently validate claims about the size or predictive usefulness of the resulting colour data.

Editorial interpretation: the most useful shift is the order of the conversation. A conventional selector asks customers to adapt themselves to a supplier’s catalogue. This experience first accepts the customer’s language, then translates it toward the supplier’s domain. That can make a professional interaction feel less like administrative work. The customer still needs judgement, samples and technical advice, but the first step can become an invitation to articulate intent rather than an examination of existing expertise.

Editorial interpretation: a surprising interface creates a second obligation. People may assume that natural language means the system understands the same memories, materials and constraints that they do. A convincing demonstration should make the translation visible enough to correct. Showing several alternatives and inviting refinement is more honest than announcing a single objectively correct colour. For a business supplier, the eventual handoff to specification matters more than the novelty of the input method.

Design the translation, not just the input

Before adding speech, map the gap between customer language and the decisions your service actually supports. Ask practitioners to describe a desired result without using your catalogue vocabulary. Preserve their meaning when converting it into options. A text field and a knowledgeable person may be a better early prototype than a sophisticated voice interface if the underlying interpretation is still uncertain.

Keep inspiration separate from specification

Treat a digital palette as an exploratory artefact with a clear next step. Show what still requires checking, including material, lighting, finish and production constraints. Do not allow an attractive representation to masquerade as a validated technical recommendation. Give customers a way to keep their original description alongside the selected options so the intent survives the transition to specialists.

Assess whose language the system understands

A personal interface should be tested with different accents, descriptive styles and access needs. Offer a nonvoice alternative and an easy correction path. Evaluate whether unusual descriptions produce useful variety or merely confusing results. Avoid treating the words people contribute as unrestricted research data; explain retention and use, and separate participation in the experience from permission for later analysis.

A small test for another brief

Proposed test: a small materials supplier could offer a guided mood-to-sample session for one narrowly defined product range, using human interpretation before investing in automation.

Measure whether customers reach a suitable shortlist with fewer clarifying rounds, and whether a technical specialist agrees that the next step is workable. Record enjoyment separately from selection quality.

Stop if users mistake exploratory outputs for approved specifications, if access barriers exclude participants, or if language data cannot be handled transparently.

• Recruit designers with a live but nonconfidential brief and obtain consent.
• Compare a conventional catalogue task with a descriptive brief followed by curated options.
• Record revisions, sample requests and misunderstandings without retaining unnecessary personal descriptions.