Product evidence
Feedback about the work works. Feedback about the worker backfires. Every widget on your home screen is one or the other.
The research, distilled
The definitive meta-analysis — 131 papers, 607 effect sizes, 12,652 participants — found feedback helps on average (d = 0.41), but over 38% of measured effects were negative. The predictor: feedback pointing at the task helps; feedback pointing at the self hurts.
Kluger & DeNisi (1996), Psychological Bulletin 119(2)Velocity feedback — change vs last time — carries d = .55, the biggest positive row. Feedback containing the next action: .43. Praise: .09 (your confetti toast, statistically). Ranks and comparisons that threaten self-esteem: .08. Shame-framed metrics go negative.
Kluger & DeNisi (1996), moderator analysis, Table 2People given pedometers walked more and enjoyed walking less — and volunteers were hurt just the same. When a page counter was removed, measured readers kept reading at half the rate of never-measured ones (27.3% vs 48.5%). Quantify what users treat as work; be careful quantifying what they do for joy.
Etkin (2016), Journal of Consumer Research 42(6)In decision experiments, less frequent feedback beat constant feedback whenever the environment was noisy — people over-weight the latest data point. A weekly rollup with a smoothed trend is an evidence-based feature, not a compromise.
Lurie & Swaminathan (2009), OBHDP 108Everything in this deep-dive
Original data
We sampled 76 home screens and dashboards from top apps via Mobbin — 64 iOS consumer, 12 web SaaS — and tallied every one by hand. Explore the patterns, then see how each app compares to the overall takeaway.
One dual-value ring (Heart Points + steps) compresses the whole day into a felt number with goal progress inherent in the ring, today-default, and clear hierarchy beneath.
Textbook hierarchy: one large hero ring, three mini rings with "0 of 6" goal framing, everything today-scoped — readable even in a near-empty data state.
Course ring plus highlighted current node delivers progress and next action in one glance — three numbers above the fold, total.
The lesson path makes the next action structural — the current node is the CTA — while score and flame stay supporting cast instead of competing heroes.
Even at zero data it leads with one felt number ("0 Day Streak"), a milestone bar, and a task CTA: "Track a food to kick off your streak!"
The 846-day flame is the purest one-felt-number screen in the sample — a single hero streak with week checks and nothing else competing.
Hero streak framing and a clear JOIN CLASS task CTA align — but it spends the top row on a rare "Welcome back" greeting instead of an anchor metric.
Composite scores compress underlying metrics well — but three competing heroes (Strain 38 / Recovery 77 / Sleep 67) and 9+ numbers dilute the one-felt-number effect.
Leads with a single hero spend number — then stacks 9+ figures beneath it, and is one of only five screens in the sample using red negative framing.
Restrained density and clear hierarchy — but pure observation-over-time: historical default, no progress framing, no felt number, no core-task action.
The closest SaaS gets to a hero metric — one big stat over a large line chart — yet still historical, goal-free, and without a next action.
9+ competing numbers across three different mental models — cashflow, budget-remaining, category spend — with no single anchor metric to feel.
Every stat repeats a "0%" comparison chip (11+ visible) and uncollapsed $0.00 rows drown the real signal — equal-weight everything, the core anti-pattern.
A grid of chart cards with 9+ numbers, red -25% badges, and no anchor or task — data density standing in for hierarchy.
Eight equal stat cells on a historical month view with no hero, no goal framing, and no CTA — a SaaS-style grid inside a consumer app.
Six equal KPI cards with sparklines, neutral 0% chips, and an upsell as the only CTA — nine-plus numbers, zero felt ones.
How to read this data: Mobbin curates top, design-forward apps, so these percentages describe leading products — not the app-store average. Retrieval is relevance-ranked per query (streak prevalence is reported per-category and with the streak-focused batch excluded), sampling is screen-level rather than app-level, and the web subsample (n=12) is small, so iOS-vs-web contrasts are directional. Full per-screen observations are in the audit.
Apply it to your app
This prompt distills everything above into instructions for an AI coding session (Claude Code, Cursor, or similar). It interviews you about your app first — so nothing changes until it understands your context — then audits every widget against the research and implements the fixes with your design system.
You are a senior product engineer applying feedback-science to my app's dashboard / home screen / stats surface. Grounding: Kluger & DeNisi's meta-analysis of 607 feedback effects (avg d = 0.41, but MORE THAN 38% OF EFFECTS WERE NEGATIVE): feedback about the task helps; feedback that points at the self backfires. The moderator table is a widget spec — velocity feedback (change vs last period) d=.55; feedback carrying the next action d=.43; praise d=.09; self-esteem threat (ranks/comparisons) d=.08; discouraging feedback negative. Also: measurement itself taxes enjoyment of intrinsically fun activities (Etkin 2016 — a visible counter cut voluntary continuation from 48.5% to 27.3%; safe when the activity is already framed as work), frequent feedback on noisy metrics causes noise-chasing (Lurie & Swaminathan 2009 — less frequent beat real-time), and progress framing should flip with commitment (to-date for new users, to-go for committed — Koo & Fishbach 2008).
BEFORE YOU CHANGE ANYTHING, ask me and wait for answers:
1. Screenshot or describe the current dashboard: every number/widget above the fold, top to bottom.
2. Who looks at this and how often? Is the underlying activity WORK for them (sales, invoices, training plans) or PLEASURE (reading, hobbies, casual play)?
3. Which single decision or action should this screen drive most often?
4. How noisy is the headline metric day-to-day, and what's the natural cadence (daily/weekly)?
5. Stack, design system, and charting library?
THEN audit each widget with one question — is this feedback about the work, or about the worker? — and show me a table (widget → classification → keep/fix/kill → evidence) before coding:
- One anchor: compress to a single hero number the user can act on, phrased as instruction-flavored status ("Safe to spend today: $42") rather than a grade. Consumer-grade screens: 1–3 numbers above the fold; everything else below.
- Add a velocity chip to the hero ("↓ $118 vs last week") — the highest-value widget in the table (d=.55).
- Convert every negative state into a next action: not "OVER BUDGET −$63 " but "Dining is $63 over — move $50 from Fun?" with a one-tap fix (d=.43). Kill shame banners.
- Replace praise/confetti with competence information: "3 days logged — your picture is now accurate" (praise ≈ d=.09; informational confirmation is what sustains engagement).
- Remove or bury leaderboards/percentile ranks for regular users (self-esteem threat row, d=.08) unless the product is explicitly competitive.
- Match cadence to noise: real-time tickers on noisy metrics get replaced by daily/weekly rollups with smoothed trends. A boring cadence is an evidence-based feature.
- Goal widgets: reachable targets that actually close; frame progress to-date for new/uncertain users, to-go for committed ones (implement the flip). If you show streaks, include repair/grace (broken streaks convert to churn without mercy — Silverman & Barasch 2023).
- Charts: encode anything users must compare as position (bars, dots, lines), not angle or area; today-default for action screens, trend-default for review screens. Use red only when you intend a risk-aversion nudge — color changes decisions (Bazley et al.).
- If the underlying activity is intrinsically enjoyable, quantify LESS: fewer counters, more qualitative confirmation (Etkin's boundary).
THEN implement incrementally with my design system. Measure: return frequency (DAU/WAU) and session depth over 4+ weeks — not just engagement with the new widgets; measurement effects show up late and in retention.
End with the widgets you deleted and the one-line evidence reason each.
See it, click it
A fictional budgeting app rebuilt widget by widget — eleven shouting numbers on the left, six work-focused widgets on the right. Toggle the annotations to see the research behind each change, tap "Move $50", and flip the goal ring between new-user and committed framing.
From Build With Kris
Subscribe to catch the teardown when it drops — the science, the audit, and the Penny rebuild, screen by screen.