Recommendation algorithms, one-click checkouts, and exploitative targeted social feeds lead to frequent impulse purchases that users regret within days. Simple resistance doesn't work, and we know that from both research and experience (yikes).

05. Existing vs. Cartbreak

Here's 3 existing solutions + what I think is wrong with them that I'd like to change:

Finance apps:

Reach the problem after the money is gone. Plus a shame tactic that makes users feel bad about money spent.

Finance apps:

Reach the problem after the money is gone. Plus a shame tactic that makes users feel bad about money spent.

Cartbreak intercepts before checkout with neutral reflection, focusing on clarity rather than shame.

Checklist with questions and points: Saved online, but nobody reads a 30 half-relevant questions and adds up their points before clicking "buy".

Medium-Long term success: Just an educated guess for now.

Interactive, changes questions based on responses for relevancy, produces a real verdict and scores automatically in 2 mins.

Just talk to AI, right?: AI has repeated shown to be highly influenceable by the tone and direction of the user.

Medium-Long term success: Just an educated guess for now.

While AI can be added later, Cartbreak should currently have a structured method that exposes the user to all points of reflection.

That led to Cartbreak.

05. Questions and logic

The diagnostic engine is built around 3 psychological lenses:

Origin & Pull (Questions 2–4): Identifies where the urge came from (external exposure vs. internal need) and what need it satisfies.
Weight (Questions 5–6): Tests whether the practical utility holds up when emotion is removed, and evaluates its true priority relative to current goals.
Reversibility (Questions 7–8): Weighs the cost of undoing a wrong purchase against the actual loss of walking away.

Then, the take. Users see their situation outlined, and decide.

Decision (Questions 9–10): Simple yes/no on whether they will make the purchase, and why. For those who say no, an optional question about what the product roughly costs.

06. Shareable cards

Cartbreak also generates a personalized outcome card that turns a quiet decision to skip a purchase into a visual badge of intentionality. This is easy to save, share, and discuss, and brings growth to the tool :)

Scoring Architecture: Answers are dynamically aggregated using a weighted system starting at 50 points. Points are added or deducted based on intent strength, urgency, and reversibility.

The outcome maps to six distinct behavioral archetypes (such as The Comparison Chaser or The Clear Eyed Buyer) paired with personalized diagnostic reads.

07. Metrics and Insights

I want to separate this section into 2: metrics for success as a benefit to the user, and metrics for success of the tool itself.

For the user:

% of Purchases declined: Tracks how often Cartbreak's reflection pause successfully stops impulse. Currently at ~45%, showing that almost half of all users who are able to interrupt buying patterns to use Cartbreak will not buy it.
Avg clarity score: Measures decision confidence and alignment before making a call. Currently at an avg ~60/100, not a great indication of thought process before a buying decision, but this awareness is what the tool is for.

For the tool:

Card share rate: How often people will share the final card graphic. Currently at ~40%, very high (possibly because its early stage) but still reflects that the card is intriguing to share, especially in conversation about a purchase.

Haven't been built for constraints, but would contain important signals for the tool's future:

Completion rate: Percentage of users who finish all 10 questions without dropping off.
Re-use rate: Frequency of returning users bringing new high-intent purchases back to Cartbreak.

08. Next Steps

I think there are 3 main issues with Cartbreak as it stands, and if I were working on this further, here's how I would solve them:

Intentionality:

Stopping to decide, something predatory e-commerce practices work against.

Intentionality:

Stopping to decide, something predatory e-commerce practices work against.

Browser Extension Integration that automatically detects cart pages or checkout URLs (like Amazon, Shopify, or brand sites) to pop up at the exact moment of purchase.

Hyper-relevance:
Not tailored to the exact product being deliberated over.

AI-Powered Product & Intent Extraction: An LLM agent in the extension automatically parses the cart page to dynamically tailor questions to specific products.

Medium-Long term success:

Just an educated guess for now.

Medium-Long term success:

Just an educated guess for now.

Medium-Long term success: Just an educated guess for now.

30-day follow-up loop over email/messages asking users if they bought the item anyway or if they still feel good about skipping, feeding this data back into the AI engine.

09. What I learnt

Micro-tools are a different game, and teach a form of clarity ecosystems don't.

Small scale wins on speed and low-friction, but won't be re-accessed without real utility. So a key part of this project was balancing asking enough questions to deliver a good take, without asking so many the drop-off rate shoots up.

They also need shareability and artifacts (like the summary cards) that turn personal reflection into social currency to create a repeat of the brief interaction.

Vibecoding and its use in validation with real people.

This project was vibecoded in a day. My intent to explore vibecoding for this problem statement was about handling faster prototyping speed with users.

Here, a realization: Most UX designers I know (and me) prefer live watching users use their figma prototype. Helpful to be there, but adds something I hadn't fully realized… until I sent this project's link to 7000+ people while it wasn't "done", and heard interesting comments about what I thought was a very simple build. Lots of changes, we're on v7 :)

Apurva Bhure

Resume

Apurva Bhure

Resume

Apurva Bhure

Resume