[ Founder Case Study · AI Product ]
Yourly: the accountability partner that calls you back
01[ Context ]
Yourly is an AI accountability partner I am building that operates where users already are: WhatsApp and voice calls. Every morning, Yourly calls to set an intention. Every evening, it calls to reflect, without judgment, without pressure. In between, users manage tasks through natural language chat. The AI is designed not to feel like a productivity tool but like a partner who genuinely cares whether you had a good day.
02[ Business Problem ]
Standard productivity apps solve the wrong problem. They help you organise tasks. They do not help you start them. Procrastination is not a list management problem, it is an emotional one. The gap between intention and action is filled with avoidance, self-doubt, and the paralysis that comes from not knowing where to begin.
The tools that exist either add more friction or provide no emotional presence at all.
03[ Constraints ]
- Every download is a drop-off pointThe product competes with avoidance itself, so first engagement has to be near-frictionless. Anything that requires installing and learning a new app loses the very users it exists for.
- Voice costs real moneyCalls through Twilio with speech-to-text and text-to-speech carry per-minute cost, which forces monetisation to be designed into the experience from the start rather than added later.
- Failure states are the productA missed call, low-confidence speech-to-text, a user too tired to talk: for an emotional product these are the moments that decide trust, so every one needs a documented, graceful fallback.
- Consumer product, banking disciplineComing from banking delivery, the admin portal was scoped from day one with an immutable audit log: every create, update, and delete recorded with user ID, timestamp, and description.
04[ Stakeholders ]
The user is the centre of gravity: someone who wants to act on their intentions and needs presence, not pressure. Around them sit the operational roles the admin portal serves, user support and platform oversight with JWT and 2FA-secured access, and the platform dependencies: Twilio for voice and WhatsApp, OpenAI for language, speech, and synthesis, Stripe for billing, and Supabase for data.
05[ Research ]
The design centre came from diagnosing the emotional mechanics of procrastination rather than the mechanics of task lists. The most instructive edge case is the user who completed nothing. That scenario, not the happy path, shaped the evening reflection: celebrate what was handled, ask what got in the way with empathy, and move the incomplete task forward without shame. Designing for that case required explicit NLP intent handling and call script branching, not just tone guidelines.
06[ Strategy ]
Meet users where they already are, and make the phone call the product. WhatsApp carries the everyday task chat; the morning and evening voice calls carry the relationship. Text chat is table stakes, and the voice call is what makes Yourly feel like an accountability partner rather than a chatbot.
Monetisation is built into that same structure: the Free tier is call-limited specifically to create a clear incentive to upgrade, so the paywall sits exactly on the product's differentiator instead of fencing off arbitrary features.
07[ Options Considered ]
- option 01A dedicated app vs WhatsApp-firstAn app offers full UI control and no platform dependency. Rejected for the MVP: every additional download is a drop-off point, and meeting users on a platform they already trust removes the biggest barrier to first engagement.
- option 02Text-only chatbot vs voice as the coreText alone would be dramatically cheaper to run. Rejected as the centre of the product: a message is easy to ignore in exactly the way a ringing phone is not, and the call is what creates the felt sense of a partner.
- option 03Tone guidelines vs engineered empathyThe easy version of "non-judgmental" is a style note in the prompt. Rejected: the zero-completion day is handled through explicit intent handling and call script branching, making empathy a tested behaviour rather than an aspiration.
08[ Trade-offs ]
- Platform dependency over interface controlWhatsApp-first means living inside another company's rules and interaction limits. The reach and zero-install trust it buys is worth more to an accountability product than a custom interface.
- A limited free tier over frictionless growthCapping free calls slows viral adoption, and it is honest: the expensive, differentiating experience is the thing being sold, so the upgrade decision is made on the product's real value.
- Fallback engineering over feature velocityDocumenting a graceful path for every failure point, missed calls, low-confidence transcription, a user who cannot talk, cost design time that could have gone to features. For a product about showing up, never failing silently is the feature.
09[ Delivery So Far ]
Yourly is in development; the foundation is documented to build-ready depth:
- Full PRD (Product Requirements Document) with user stories, BDD acceptance criteria, and documented platform decisions
- 4-workflow n8n backend architecture: onboarding, task management via NLP, morning/evening scheduled calls, admin portal
- Defined success metrics: 30-day retention greater than 20%, call answer rate greater than 60%, NLP accuracy greater than 80%, NPS greater than 40
- Tech stack: OpenAI (GPT-4o-mini, Whisper STT, TTS), Twilio (voice and WhatsApp), Supabase/PostgreSQL, Stripe, n8n
- Admin portal spec: JWT and 2FA auth, user search and edit, ops dashboard, immutable audit log
10[ Technical Architecture ]
The backend runs as four n8n workflows: onboarding, task management via NLP, the morning and evening scheduled calls, and the admin portal. OpenAI provides the language layer, GPT-4o-mini for conversation, Whisper for speech-to-text, and TTS for the voice, while Twilio carries both the calls and WhatsApp. Supabase/PostgreSQL holds state, Stripe handles billing, and the admin portal is specified with JWT and 2FA authentication, user search and edit, an ops dashboard, and the immutable audit log.
Graceful degradation is architectural: when a call is missed or transcription confidence is low, the agent routes to WhatsApp, confirms understanding, and closes the interaction on a human note. It does not fail silently.
11[ Outcomes ]
At this stage, the documentation is the artefact. A full PRD with user stories and BDD acceptance criteria, a specified admin portal with operational discipline borrowed from banking, deliberate platform decisions with recorded rationale, empathetic failure-state design, and the n8n architecture to run it all: Yourly is proof of what product thinking looks like end to end, from problem diagnosis to backend architecture.
12[ Metrics ]
Pre-launch, the metrics are defined targets the MVP will be judged against, set in the PRD before a single user session.
13[ Lessons Learned ]
- Put the paywall on the differentiator. Monetisation designed into the product experience, the call itself, is cleaner than fencing off secondary features and hoping users hit the fence.
- Empathy is an engineering requirement. "Non-judgmental" only becomes real when the worst-case scenario has its own intent handling and script branch, testable like any other behaviour.
- Operational discipline transfers across domains. An immutable audit log feels like banking overkill for a consumer MVP right up until the first support escalation, billing dispute, or data question.