Work02 · Generative content toolCreative tool
Plume
On-brand content, generated in minutes — not gimmicks
A generative studio for small businesses that can't afford a content team: type one idea and it produces a week of social posts — captions and graphics — locked to your brand. The hard part was never the generation; it was trust and control.
Try the product
Generate a post with Plume
Give it a business and a goal — it drafts an on-brand caption the way the real generator would.
A working agent, running live — not a mockup
Plume
Content generator · live demo
Plume
on-brand contentGenerates on-brand, never off-brand
Interactive prototype — click around, it actually works
Strategy deliverable
The journey we mapped before designing anything.
Every stage from first spark to loyal return — what the customer does, feels, and exactly where the experience leaks. The dip in the curve is the moment we built the work around.
Customer journey map · concept
Sana, 41 — keeping the shop's feed alive
Owns a candle shop, knows she should post but has no time, no team, and no content skills.
Customer journey map for Sana, 41 (keeping the shop's feed alive), across 6 stages. Sentiment dips lowest at the Trial stage (the biggest drop-off) and peaks at the Habit stage.
- Awareness: The feed's been dead for weeks. The customer Knows she should be posting but has neither the time nor the ideas. Pain point: No team, no ideas, no time — so the feed just sits there looking abandoned. Our move: A tool that turns a single idea into a week of posts in minutes.
- Trial: Tried a generic AI writer once. The customer Pastes a prompt into a generic AI and gets bland, off-brand text. Pain point: Generic output is obvious and off-brand, so she posts nothing rather than something embarrassing. Our move: Brand-locked generation, so the output sounds like her shop — not a robot.
- First generate: Wait — that sounds like us. The customer Sets her brand once, types 'weekend candle sale,' gets on-brand options. Pain point: Trusting AI output enough to actually put your name on it. Our move: Options, not an answer: she picks and tweaks, always in control.
- Edit & trust: Changed two words, done. The customer Lightly edits a draft and regenerates the one she didn't love. Pain point: The fear that AI will post something wrong in her name. Our move: Nothing auto-posts; edit and regenerate make every draft unmistakably hers.
- Habit: A week of posts in ten minutes. The customer Schedules a week of on-brand posts before she opens the shop. Pain point: Consistency was simply impossible for a one-person shop. Our move: Fast, on-brand, scheduled — the dreaded chore becomes a trivial habit.
- Advocacy: Customers think we hired someone. The customer The feed looks professional; she tells other small-shop owners. Pain point: Small shops look amateur online next to the chains. Our move: A consistent, on-brand presence that punches well above her size.
Scroll the map sideways →
Engagement
Discovery + Sprint
Timeline
6 weeks
Surfaces
Desktop app, brand kit, generation canvas
Agent role
On-brand caption & graphic generation
The problem
Small businesses are told to 'post consistently' with no one to do it. Generic AI writers spit out off-brand, soulless content that's obvious and a little embarrassing — so owners either post nothing or paste raw model output that doesn't sound like them. The blocker isn't whether AI can write; it's whether an owner can trust it to stay on-brand and stay in control.
The approach
We treated this as an AI product-design problem, not a prompt. The hard part of a generative tool is the interface around the model: how a non-designer locks their brand, judges the options, and stays in control without learning to 'prompt.'
The brand kit is the guardrail. Colors, type, and voice are set once and applied to everything the tool makes, so output is on-brand by construction — the owner picks a voice, not a temperature slider.
Every generation produces options, not an answer, and each one is editable, regenerable, and obviously a draft. Nothing posts without a human tap. We designed hardest for the moment the AI gets it slightly wrong, because that moment decides whether people keep using it.
Key moves
01
Brand-locked by construction
The brand kit constrains every output — colors, type, and voice are applied automatically, so 'off-brand' isn't even an option the tool can produce.
02
Options, never an answer
Each prompt yields a set of drafts to choose, regenerate, and edit. Choice plus control is what makes a person actually trust a generative tool.
03
Designed for the wrong answer
Confidence, edit, and regenerate states are designed deliberately — because how a generative feature fails is most of its real UX.
Design targets
- to a week of on-brand posts
- 1 prompt
- off-brand output, by construction
- 0
- every draft — nothing auto-posts
- Editable
A self-initiated concept study. Plume is a fictional tool; the prototype above generates from a fixed set, and the agent below writes live.
Next study · 03 Multi-location clinic
Verra →