From research to interface
Part 1 covered the problem and the evidence. This part covers what was built from it: the structure, the flow, the screens, the rules that keep the interface consistent, how it will be tested, and how it could earn money. Read part 1 first if you have not.
An app that holds your data and an extension that meets you where you shop.
The first IA put a Shop tab inside the app. Part 1 showed why that fails: most regretted purchases start in a feed, not a store the user opens on purpose. The product split into two surfaces that share one backend.
The spending moment, step by step.
Everything the product does converges on one interaction. It triggers when an item is added to the cart, takes about twenty seconds, and every outcome is logged as a win. Dashed steps are optional.
Low fidelity, built to argue about structure.
These are the revised wireframes. They are deliberately grey: the question at this stage is hierarchy and flow, not polish. The interactive demo shows the same screens at high fidelity.
Rules that make "no judgement" visible.
Principles
Witness, never judge
No warning colours or verdicts on past purchases.
Every outcome is a win
Buying, pausing, and removing all raise the score.
Quiet by default
Private notifications. Nothing visible to a bystander.
On device first
Anything involving the camera is processed locally.
Colour
Red and amber are banned from the spending experience. They read as danger and would teach the user to associate buying with wrongdoing. They are reserved for system errors such as a failed payment.
Voice
Nudge is the named coach. It asks more than it tells, opens with something specific and observed, and never uses clinical or moralising language.
Each mechanic has a reason, and a risk.
| Mechanic | Principle | Risk to watch |
|---|---|---|
| Question order varies | Fresh prompts resist habituation (Fogg) | Harder to build consistent pattern data |
| 10 second pause on "I need it now" | Friction scales with urgency | Users abandon the flow instead of waiting |
| "Buy anyway" is the quietest option | Choice architecture, defaults steer | Steering is still influence. Must be disclosed |
| Optional line in the user's own words | Commitment and consistency (Cialdini) | May read as friction. Untested |
| Short delay before the score appears | Anticipation, variable reinforcement | Borrows the mechanic it aims to counter |
Most of this is a hypothesis until it is tested.
Concept validation, about 15 people
Click through the prototype while thinking aloud, using a product they were genuinely considering. Anything that fails is cut or redesigned.
Closed beta, 150 to 300 users
A working build over at least four weeks, with weekly check in surveys and screen level analytics. Retention needs eight to twelve weeks before any decision rests on it.
Merchant pilot, 3 to 5 stores
Eight weeks against a held out control period. This is the only stage that can test the return rate claim with real transactions.
Staged rollout
10, 25, 50, then 100 percent, with rollback per cohort. The A/B infrastructure stays on permanently.
A/B hypotheses
Varying order keeps the check from becoming autopilot.
Metric: completion rate by week four, plus tap speed over time as an autopilot signal.
Writing a reason may raise follow through. It may also just add friction.
Metric: does writing a reason predict waiting out the pause. If over 80 percent skip it, cut it.
A pause on high urgency should raise pause and remove rates without raising abandonment.
Metric: outcome mix among "I need it now" users. Watch for abandonment spikes.
A short delay may feel rewarding or manipulative.
Metric: next day open rate. In the survey, reported discomfort overrides a positive number.
Success measures (targets, not results)
| Measure | Target |
|---|---|
| Check in completion | over 60% by week 4 |
| Pause then remove | over 35% |
| 30 day retention | over 25% |
| Score trending up | 70% of active users by week 6 |
| Merchant return rate | significant drop against control |
Known risks: novelty inflating beta engagement, merchants who volunteer already having healthier return rates, and users performing "pause" to watch the score climb. The third is the likeliest and the hardest to see in analytics.
Merchants fund it. Shoppers keep it free.
Fashion, beauty, lifestyle
Emotional, identity driven purchases with high regret returns. This is where the mechanism has something to act on.
Hardware and staples
Functional, pre planned purchases. A pause prompt has little to change.
| Stream | Model (hypothesis) |
|---|---|
| Shoppers | Core experience free. A premium tier later for deeper trigger analysis and bank connection. |
| Merchants, install | Free |
| Merchants, analytics | Flat monthly fee, to be tested against willingness to pay |
| Merchants, performance | Small share of verified return savings, charged only after a baseline comparison |
Monetisation rules
- No sign up wall before the user has felt value.
- No countdown timer discounts. A product built to resist urgency cannot sell with it.
- One paywall screen, never a cascade.
- Merchant fees tied to proof, so the risk of an unproven claim sits with SpendWise.
Unresolved
- Native in-app mobile purchases are out of reach for a browser extension.
- Operating systems or payment platforms could ship a native pause. The merchant relationship and longitudinal data are the defensible parts, which is why the model leans toward B2B.
- Procurement teams often prefer predictable flat fees over revenue share.
The instruments for real fieldwork.
These are the questionnaire and interview guide to run with real people. Aim for 20 to 30 survey responses and 5 interviews. Questions ask about a specific recent event instead of general habits, because recall of a real moment is more accurate than self description.
Survey (about 4 minutes)
- How old are you?Multiple choice: 18 to 24, 25 to 34, 35 to 44, 45 and over
- In the last 30 days, how many online purchases did you later regret or return?0, 1 to 2, 3 to 5, more than 5
- Think of your most recent regretted purchase. What were you doing just before?Select all: scrolling social media, stressed, bored, late at night, seeing a friend's purchase, celebrating, other
- At that moment, how aware were you that your mood might be affecting the decision?Not at all, slightly, aware but bought anyway, fully aware
- What did you do with it afterwards?Kept and used, kept unused, returned, resold
- Have you used a budgeting app? If you stopped, why?Never tried, felt judgemental, too much manual work, did not help, still use one
- Would a single question before checkout be helpful?Helpful, depends on the question, I would ignore it, annoying
- How would you feel about camera based mood detection if all processing stayed on your device?Comfortable, depends on the app, uncomfortable, never
- What would make you trust an app like this?Open text
- Optional: would you take part in a 20 minute follow up interview?Yes with email, no
Interview guide (20 to 30 minutes)
- Tell me about the last thing you bought online that you were not sure about.
- Walk me through the hour before you bought it. Where were you, what were you doing?
- How were you feeling? When did you first want it?
- What happened after the order was placed?
- Have you ever tried to stop yourself from buying something? What did you try? Did it work?
- Show the prototype. Think aloud as you go through it with a real item you are considering.Observe where they hesitate, skip, or smile. Do not explain.
- Which part felt useful? Which part felt annoying or intrusive?
- What would make you delete an app like this?