CH · 01 · PROBLEM FRAMING章节 · 01 · 问题诊断
The current flow feels like two separate tools when it should feel like one assistant.现在的流程让人感觉是 两个分开的工具, 但它本应该感觉像 一个会帮你做事的助手。
Atria’s product sells itself as turning ad performance data into creative decisions. But the current flow ships those two halves separately: an analytics dashboard on one side, an image generator on the other, no glue between them. Here are ten specific places where it breaks.
Atria 这个产品对外说的是「把广告投放数据,变成可执行的创意决定」。 但现在的产品里,这两件事是分开的:左边是个数据看板,右边是个 AI 出图工具,中间没有衔接。下面列出 10 个具体破裂的地方。
How we found these issues
Three sources: (1) two short videos the Atria team recorded showing how the product works — they explicitly named two problems they already know about; (2) screenshots of the six key screens from their design file; (3) walking through 17 made-up test scenarios to see where the product breaks.
We organized everything along Atria’s own three improvement axes — Clarity (Is the information easy to read?), Actionability (Can the user actually do something?), and Continuity(Does the user’s flow feel connected?) — and ranked issues by how often they hit and how much they hurt.
这些问题是怎么找到的
三个来源:(1) Atria 团队录的两段简短产品演示 视频 —— 他们自己点名了两个已知的问题;(2) 他们设计文件里的 6 张关键屏幕截图;(3) 用 17 个虚构的使用场景走一遍,看哪里会卡住。
所有问题按 Atria 自己提出的 3 个改进维度归类 ——Clarity(清晰度):信息好不好读?Actionability(可执行性):用户能不能真的做点什么?Continuity(连贯性):整个使用过程顺不顺? —— 然后按「触发频率 × 影响大小」排序。
MUST-HIT · Atria self-reported in Loom video #2Loom 视频 #2
必须命中 · Atria 团队在 Loom video #2Loom 视频 #2 自己提出来的
Four problems the team already knows about.
团队已经知道的 4 个问题。
They named these on camera and asked candidates to extend the list. Any redesign that misses them under-delivers.
他们在视频里点名了这几条,然后让候选人继续往下找。任何重设计如果漏掉这些就是不及格。
- Self-reported #1 — generation takes long, users always leave, but can’t find their work later (from Loom video #2Loom 视频 #2)团队自报 #1 —— 生成耗时太长,用户必然离开,但回来后找不到刚才的产物(来自 Loom video #2Loom 视频 #2)
- Self-reported #2 — Iterate is only available for image ads, but Radar doesn’t show which ads are video (from Loom video #2Loom 视频 #2)团队自报 #2 —— Iterate(迭代)按钮只对图片广告可用,但 Radar 表格里没标哪些是视频广告(来自 Loom video #2Loom 视频 #2)
- Implied — clicking a row opens Detail in a new browser tab, breaking the scan rhythm (from Loom video #1Loom 视频 #1)隐含问题 —— 点行直接在新浏览器标签里打开详情,打断扫表节奏(来自 Loom video #1Loom 视频 #1)
- Implied — generated images end up in Creation → Image Ads, not in the Radar workflow (from Loom video #2Loom 视频 #2)隐含问题 —— 生成的图最后跑到 Creation → Image Ads 下面,跟 Radar 工作流脱节(来自 Loom video #2Loom 视频 #2)
OVERVIEW · 10 ISSUES ACROSS 3 AXES全景 · 3 个维度下的 10 个问题
The issue map
问题地图
CLARITY · Information is hard to read
清晰度 · 信息不好读
ACTIONABILITY · User can't act on what they see
可执行性 · 用户看到了却动不了
Highlighted boxes (A-2, K-3) are the ones the Atria team self-reported.
高亮的两块(A-2、K-3)是 Atria 团队自己点出来的问题。
AXIS · CLARITY维度 · 清晰度
Clarity.清晰度。
Are insights and recommendations easy to scan and understand at a glance? Right now they read like decoration on a dashboard — pretty colors but no instructions.
洞察和建议是不是一眼就看得懂?现在它们像看板上的装饰品 —— 颜色挺好看,但没告诉你该干什么。
C-1
Recommendation tags look the same but do different things
推荐标签长得一样,但做的事完全不同
✓ Solved by Move 01 →
✓ 由 Move 01 解决 →
Each row shows a tag — Iterate, Pause, Scale, or Refresh. They look identical (same shape, same pill style), but they trigger four very different actions: Pause is one switch on Meta; Scale changes a budget number; Iterate triggers an AI image generation that takes a minute. Users can’t tell which is heavy and which is light at a glance.
每一行末尾都有一个标签:Iterate(迭代)、Pause(暂停)、Scale(扩量)、Refresh(刷新)。 它们看起来一模一样(同样形状、同样的小圆角标签),但其实对应 4 件完全不一样的事:Pause 只是在 Meta 后台关一个开关、Scale 只是改一个预算数字,而 Iterate 会触发要等一分钟的 AI 出图。用户一眼看不出哪个轻、哪个重。
Hovering the Recommendation tag shows a short popup. The «AI Recommendation» button on the same row opens a full detail page. Both are AI, both are on one line — users have to guess which one gives more.
鼠标悬停 Recommendation(AI 建议类型)标签会弹出一个简短气泡。 同一行还有一个独立的「AI Recommendation(AI 建议)」按钮,点 进去是完整的详情页。两个都标着 AI、都在同一行 —— 用户得猜哪个 能给到更多信息。
C-3
«Customize columns» is built for the wrong user
「自定义列」是给错了用户准备的
✓ Solved by Move 09 →
✓ 由 Move 09 解决 →
The default table already shows 11 columns of metrics. Adding a «Customize columns» button implies the user is a data analyst who wants more. But the real user — creative strategist — wants to decide which ads to fix, not pile on extra numbers.
默认表格已经有 11 列数据指标,再加一个「Customize columns(自定义列)」按钮, 暗示用户是个想要更多指标的数据分析师。但 Atria 真正的用户是 creative strategist(广告创意策略师),他们想的是「该改哪条广告」, 而不是「我还要加几列数字」。
AXIS · ACTIONABILITY维度 · 可执行性
Actionability.可执行性。
Can the user actually move from data to a creative action? Today the path makes the AI guess for the user, not for the platform.
用户能不能从「看到数据」走到「真的做出新创意」?现在的路径让 AI 把决定权又推回给用户,而不是 AI 自己做主。
A-1
AI lays out 5 options flat — pushes the decision back to the user
AI 把 5 个选项平铺,等于把决定权又推回给用户
✓ Solved by Move 02 →
✓ 由 Move 02 解决 →
Click into Detail Drawer → the AI Recommendation tab shows a 5-row table with checkboxes for every recommendation. The user has to read all five and tick the ones they want. But the product’s tagline is «think strategically» — Atria should pick a priority, not hand the user a menu.
点进 Detail Drawer(详情抽屉) → AI Recommendation 标签页是个 5 行表格, 每条建议旁边都有一个勾选框。用户得把 5 条都读完、自己挑该勾哪些。 但 Atria 的卖点是「帮你做战略决策」 —— 那 AI 应该自己有个优先级, 而不是甩一份菜单给你。
A-2
Generation takes minutes, but there's no queue or history
生成要几分钟,但没有任务队列、也没有历史记录
✓ Solved by global Tasks panel + Move 06 →
✓ 由全局 Tasks 面板 + Move 06 解决 →
AI image generation takes one or two minutes. Most users naturally switch tabs or do something else while waiting. When they come back later, there’s no global Tasks list, no notification, no easy path to find where their work went — they have to dig into Creation → Image Ads and hunt.
AI 出图要 1-2 分钟。用户自然会切到别的标签页做别的事。等他们回来, Atria 没有一个全局的「我的任务」列表、没有完成通知、也没有简单的 入口找回去 —— 用户只能去 Creation(创意)→ Image Ads(图片广告) 里面翻找。
A-3
Variants page gives 16 images but no tools to pick
Variants(变体)页给了 16 张图,但没有任何挑选工具
✓ Solved by Move 04 + 05 →
✓ 由 Move 04 + 05 解决 →
After generation, the user sees 4 directions × 4 images = 16 variants. There’s no predicted score, no compare mode, no bulk select, no «push to ad platform» action. The page hands the user a wall of images and walks away.
生成完之后,用户看到 4 个方向 × 4 张图 = 16 个变体。 但页面上没有预测分数、没有对比模式、没有批量选择、也没有「推回投放 平台」的按钮。等于把一墙图扔给用户就走了。
A thumbs-up/down sits at the bottom of every detail page. Click «not helpful» and... nothing changes. The next time the user opens Atria, the same kind of recommendation comes back unchanged.
每个详情页底部都有一个赞/踩按钮。点了「没用」之后...什么都不会变。 用户下次再打开 Atria,同类型的建议照样原样推荐。
AXIS · CONTINUITY维度 · 连贯性
Continuity.连贯性。
Does the user's path feel like one connected flow? Right now every step opens a new tab or jumps to a different part of the sidebar.
用户从头走到尾,体验是不是连成一条线?现在每一步要么开新标签页,要么跳到侧边栏完全不同的位置。
K-1
Ad Detail opens in a new browser tab
广告详情页是在新的浏览器标签页里打开的
✓ Solved by Detail drawer →
✓ 由 Detail 抽屉解决 →
Click any ad row → it opens a new tab. The user’s Radar list is now in tab #1, the detail in tab #2. If they want to check the next ad, they have to either close this tab and go back, or end up with five tabs open in 10 minutes.
点任何一条广告 → 在新的浏览器标签页里打开。原本的 Radar 列表在 第 1 个标签页,详情在第 2 个标签页。用户想看下一条广告,要么关掉 当前标签返回,要么 10 分钟开 5 个标签 —— 浏览器很快爆掉。
K-2
Generated images land in the wrong place in the sidebar
生成的图最终落在了侧边栏错误的位置
✓ Solved by Move 10 →
✓ 由 Move 10 解决 →
The user discovers the ad in Analytics → Radar, but the output of iteration ends up in Creation → Image Ads — a different sidebar section, mixed with unrelated generation tools like Clone-S and IT-RID. The treatment loop never closes visually.
用户是在 Analytics(分析)→ Radar 里发现广告的,但迭代后的产物却跑到了 Creation(创意)→ Image Ads 里 —— 完全不同的侧边栏分组,还跟 Clone-S、IT-RID 这些不相关的 生成工具混在一起。一个完整的「发现-治疗」闭环在视觉上从来没合上过。
Radar doesn’t mark which ads are image and which are video. The user clicks into a video ad → walks all the way to the Iterate action → only then discovers the button is disabled because Atria only supports image iteration. Wasted attention with no early warning.
Radar 表格里没标哪些是图片广告、哪些是视频广告。用户点进一条视频 广告 → 一路走到 Iterate(迭代)按钮那一步 → 才发现按钮是灰的 (因为 Atria 只支持给图片广告做迭代)。注意力白白浪费,前面没有 任何提示。
COVERAGE · all 10 issues addressed覆盖率 · 10 个问题全部解决
Every issue above is solved by one or more moves in the prototype.
上面每一个问题,都被原型里的一个或多个 move 解决。
Round one (Moves 01–07) reframed the product’s posture from «analytics + image generator» to «AI copilot». Round two (Moves 08–11) closes the four leftover issues — collapsing redundant AI entries, replacing column customization with role presets, surfacing iteration output in the right sidebar subtree, and making feedback actually change something.
第一轮(Move 01-07)把产品的姿态从「分析工具 + 出图工具」 改成了「AI 副驾」。第二轮(Move 08-11)收掉剩下的 4 个问题 —— 收掉重复的 AI 入口、用角色预设代替自定义列、把迭代产出放回 到正确的侧边栏分组、让反馈真的能改变系统行为。
Mapping issues to moves
The mapping below lets reviewers verify every issue is answered. Each move is documented in the Solution chapter and demonstrated in the interactive prototype.
Round one (Moves 01–07) tackled the highest-impact six issues first. Round two (Moves 08–11) closed the remaining four with smaller, focused changes — making the prototype cover the full diagnosis, not just the headline subset.
问题与 move 的对应关系
下表方便评估者逐条核对「每个问题对应哪个 move」。每个 move 在 方案章节 里有详细说明,在 可交互原型 里可以实际体验。
第一轮(Move 01-07)先解决了影响最大的 6 个。第二轮 (Move 08-11)用 4 个更小、更聚焦的改动把剩下的 4 个收掉 —— 让原型覆盖整个诊断,而不只是其中最显眼的那一部分。