Confidential. Grow is a codename for an unreleased Constant Contact product. Reading this case study constitutes agreement to keep its contents confidential. Participant names and identifying details have been removed or changed.

Grow brand study

Rebuilt a polluted brand test so it measured the brands, not the brand books, then audited the results and caught AI summaries that had dropped the audience that mattered most.

Spark — direction A · one agent
Crew — direction B · a crew of sub-agents

Two brand directions for Grow, rebuilt as matched, interactive brand books.

00 — Overview

Problem

Grow needed a brand. The team explored two distinct directions on purpose, to learn what each did best, but the first round of testing was polluted: the two brand books differed in length, structure and purpose, and only one had an interactive demo. People got lost, and we were measuring the books, not the brands.

Solution

I worked with the senior designer behind the second direction to rebuild both as matched brand books, then verified every result against the raw data before the team made a call.

Result: the best-supported finding, dropping the mascot characters, carried into the final brand, and the team had a verified record of what the data did and didn't support.

My Role

Principal Product Designer. Designed the Spark direction, co-designed the matched test with the senior designer who created Crew, and audited the analysis.

How we rebalanced it

  • Same sections in the same order: character, mark, color, type, motion, in the app
  • Matching interactive demos. Both characters animate through the same states, in light and dark mode
  • One deliberate difference, to answer a real product question: one agent (Spark) or a crew of sub-agents (Crew)?

Both books ship with light and dark modes.

Checking the analysis, not just the results

Three rounds ran with 30 participants: one with small business owners and two with general professionals, with viewing order reversed in one round. Before the team made a call, I verified every figure against the raw exports and found the earlier AI-generated summaries were wrong. They left out the small business round, dropped participants and invented demographic details. I then checked my own read against a second method: a rule-based classifier with no AI inference, which marks any ambiguous answer as unclear instead of guessing.

  • The population split mattered. General professionals clearly preferred Spark: 15 to 3 in my read of every transcript, and still 10 to 1 under the stricter classifier. Small business owners split 5 to 4 under both methods, statistically a coin flip, so a blended percentage hid the answer that mattered most.
  • One finding held everywhere. In all three rounds, illustrated mascot characters read as childish, with no reversals.
  • Each direction had a clear strength and a clear risk. Spark read as approachable and easy to understand. Crew read as more serious, sometimes cold, and to some like a design tool rather than a marketing tool. Its plain-language copy was praised, but its color system and animated dot mark caused the most confusion, especially among small business owners.
  • Some claims didn't hold up. "Spark is cluttered" reversed in one round, and an argument about legal and finance owners rested on three people.
  • Both books shared a gap. Neither made clear at first glance what the tool does, or proved the AI could do it. Real product screens with concrete numbers fixed that in a later round.

What changed

The original brief asked for delight, gamification and character. In every round, the research showed that characters undermined credibility for this audience, and that changed the brief. Design leadership chose a direction based largely on Crew, and the mascots were dropped. I flagged the remaining risks: the color system and pixel-style type, which confused or put off the most people.

Constraints worth naming

  • The small business round was 10 people, so its results are directional.
  • The sample skewed older: no one was under 35, and half were 45 to 54.

Back to the main case study: Grow →