Introduce AI-Powered Activity Summaries
What is the goal?
Help users understand their design patterns and decision-making habits over time by generating weekly AI-powered summaries of their diary entries. The summaries should surface recurring themes, blind spots, and confidence trends without requiring any manual effort from the user.
Solution brainstorming
A weekly digest email and an in-app Insights panel that uses LLM analysis to review the past 7 days of diary entries. It identifies: 1. Recurring problem themes across entries 2. Categories where confidence is consistently low 3. Questions that are frequently skipped 4. Shifts in solution stage patterns The summary is written in a conversational, coaching tone -- not robotic bullet points. Users can thumbs-up/down each insight to improve future relevance.
Research
Problem Definition
Define the problem clearly
What is the problem you're solving?
Users create diary entries consistently but rarely look back at them to identify patterns. They treat each entry as isolated rather than part of a larger body of work. Without synthesis, the diary becomes a write-only tool -- valuable in the moment but not compounding over time. Users are missing their own blind spots because nobody is reading across their entries.
Who experiences this problem?
Active users who have created 5 or more diary entries over at least 2 weeks. These are engaged users who have bought into the reflection process but are not getting the full value because they lack the time or habit to review their past entries and spot trends.
When does it happen?
It becomes apparent after about 3-4 weeks of use. Users have accumulated enough entries that patterns exist, but they are invisible. The user opens the app, creates a new entry, and closes it -- never scrolling through past entries or noticing that they have rated Problem Clarity as low in 8 out of 10 entries.
Problem Clarity
Understand what the user is trying to accomplish
What is the user trying to accomplish?
Users want to become better designers and decision-makers. They started using this tool to be more intentional, but they need help connecting the dots between individual reflections. They want someone (or something) to hold up a mirror and say: here is what your entries reveal about your process.
If your design could only do ONE thing well, what would it be?
Surface one genuinely surprising insight per week that makes the user stop and think. Not a dashboard of charts -- a single, well-articulated observation like: "You have rated your confidence in Problem Definition as 3 or below in your last 4 entries. You might be jumping to solutions before fully understanding the problem."
What exactly is the user struggling with right now?
Users do not have the time or mental bandwidth to read through their own entries and synthesise patterns. Even if they wanted to, they would need to open each entry, compare confidence ratings, look for recurring themes in their answers, and draw conclusions -- a task that could take 30+ minutes and most people will never do.
User & Context
Features don't exist in isolation
Who is the primary user for this feature (not the product)?
Mid-level product designers (3-7 years experience) who use the tool at least twice a week. They are past the onboarding phase and have developed a reflection habit. They are motivated by professional growth and would value data-driven feedback on their own process.
What's the job they're trying to finish when this feature appears?
They are trying to improve their design practice over time. They use the diary as a tool for accountability and growth, but they need the tool to give back -- to reward their investment of time with insights they could not generate on their own.
Product Intent & Leverage
Why this feature exists for this product
How does this feature reinforce the product's core value?
The product is about thoughtful design reflection. AI summaries close the feedback loop -- they transform individual reflections into compounding wisdom. This is the feature that turns a journaling tool into a coaching tool, which is a significant leap in perceived value.
If we removed this feature later, would users revolt or barely notice?
Users who have experienced the weekly insights would strongly miss them. In user interviews, the most common request from power users is exactly this: help me see what I am not seeing. However, users who have not yet hit 5 entries would not notice its absence.
Success Definition
If you can't measure it, you're guessing
What user behaviour should change if this works?
Users should start revisiting past entries after reading their weekly summary. We should see an increase in edits to existing entries (users updating their confidence ratings or adding notes after reflection). We should also see users creating entries more consistently to get better summaries.
What does "success" look like in one week? One month?
One week: 60% of eligible users open the summary email or view the Insights panel. One month: 25% of users who read summaries revisit at least one past entry. Retention among users who engage with summaries is 20% higher than those who do not.
What metric moves because of this feature?
Weekly active user retention (target: +15-20% for summary readers), entries per user per week (expecting a slight lift from 2.1 to 2.5 as users are motivated to feed the system), and NPS score among power users (currently 42, targeting 55+).
Constraints & Trade-offs
Designing is choosing what not to do
What constraints are you working within?
LLM API costs are a concern at scale -- we need to keep per-user weekly cost under $0.05. The summary generation must run asynchronously and not block the UI. We cannot store raw diary text in third-party LLM logs for privacy reasons, so we need to use a provider with a zero-retention data policy or self-host. Email delivery infrastructure does not exist yet.
What design debt might this create?
The Insights panel introduces a new navigation paradigm -- it is neither a diary entry nor a settings page. We need to find the right home for it without cluttering the existing simple navigation. The AI-generated text also sets a quality expectation; if early summaries are generic or wrong, trust will be hard to rebuild.
Discoverability & Learning
A feature unused is a bug
How will users discover this at the right moment?
The weekly email is the primary discovery mechanism -- it arrives in the user inbox on Monday morning with a subject line referencing their specific entries. In-app, a subtle notification dot appears on the Insights tab when a new summary is ready. First-time users see a brief explanation of what the feature does.
How will they understand it without reading?
The summary is written in plain English with a clear structure: one headline insight, 2-3 supporting observations, and a single suggested action. No charts, no jargon, no configuration needed. The thumbs-up/down buttons are self-explanatory for providing feedback.
Pre-Mortem
Imagine your design failed
What went wrong?
The AI summaries launch and early feedback is positive, but within a month the novelty wears off. The insights become repetitive because users tend to work on similar types of problems. The summary starts saying the same thing every week: your confidence in constraints is low, you skip the alternatives section. Users stop opening the emails and the feature quietly dies. We built a feature that impresses on first use but does not sustain engagement because it lacks depth and personalisation over time.
Want to create your own design diary entries?