BACK HOME



CASE STUDY
Milk.me
Designing for the hardest weeks of motherhood. milk.me is a pumping companion app for new mothers — intelligent tracking, personalised guidance, and the encouragement to keep going when it gets hard.

ROLEProduct Creator,  Designer

PROJECTProduct Design · Ideation Phase · AI Assisted · 2026



THE PROBLEM




Most mothers
quit not from
inability
but from
discouragement.

Pumping is one of the least-supported parts of new motherhood. Existing tracker apps treat it like a fitness log — raw numbers with no context, no encouragement, and no intelligence. A mother staring at 2oz after 30 minutes has no way of knowing whether that's normal, concerning, or actually a sign of progress.

The emotional stakes are enormous. Supply anxiety is a leading cause of early breastfeeding cessation. Mothers stop not because they can't do it — but because nothing is telling them they're doing fine.

milk.me is designed to fill that gap: intelligent tracking that give
s numbers meaning, and a tone that feels like a knowledgeable friend, not a medical dashboard.



This project is speculative — it grew from personal experience, observation and secondary research into maternal health apps, not primary user interviews. That's an honest limitation, and one that shapes what comes next. The design decisions are grounded in publicly available research on breastfeeding cessation and competitive analysis of existing apps — Huckleberry, Baby Tracker, and LactApp — not direct user testing.  


MY PROCESS

01
Problem definition & feature scoping

The PRD was a thinking tool, not a deliverable. Used Claude to challenge the problem statement early — generating scenarios, surfacing blind spots, asking the uncomfortable questions before any design decisions were made. The key reframe: milk.me isn't a tracking app, it's a confidence app. That insight drove every feature decision that followed. Features were mapped against user needs and cut hard — the smallest set with the biggest emotional impact, nothing more.



Claude · ChatGPT



02
Exploration & concept generation

Ran parallel explorations across Stitch, Claude Design, and Figma Make — not to find a visual style, but to stress-test product assumptions through making. Generating screens quickly forced real decisions: what information does a exhausted mother actually need at a glance, what's noise, and where does the interface need to get out of the way. AI output served as raw material to react against, not solutions to ship. The aesthetic direction — warm, editorial, minimal — emerged from those product decisions, not before them.



Stitch · Claude Design · Figma Make


03
User flows & rapid prototyping

Mapped the three flows that matter most — logging a session, reading your supply trend, and receiving a nudge at the right moment. Sketched by hand first to commit to the logic and sequence, then moved straight into Google Stitch with a defined design direction to prototype interactions at pace. Testing with visual fidelity from the start meant feedback was immediate and real — does logging feel fast enough at 3am, does the encouragement land or feel intrusive? Committing to a direction early wasn't a shortcut, it was the constraint that made validation possible.



Stitch · Figma



04
Design system foundations
The ideation process converged into a set of intentional design decisions — colour, type, spacing, component behaviour, tone of voice. Rather than leaving these locked in Figma, documented the system using Design.MD, Google Labs' open-standard format for expressing design systems in plain-text Markdown. The goal was practical: give AI coding agents like Cursor a reliable, version-controllable brand guide so generated UI stays consistent without constant correction.




Stitch Design.MD · Cursor

GOOGLE STIT



AI  AS A DESIGN PARTNER


More ideas,
faster decisions,
less attachment.
That's the leverage.


Speed
Visual directions that would have taken days to moodboard were testable in hours. That compression meant more iterations, earlier decisions, and less attachment to any single idea.


Thinking Tools
Every AI tool was used to force a decision, not produce an output. Claude to interrogate the problem. Stitch and Figma Make to make hypotheses concrete enough to evaluate and discard quickly.


Where AI doesn’t go

The reframe — confidence app, not tracking app — didn't come from a prompt. The cuts, the hierarchy calls, the moments of restraint: those required judgment. AI accelerated exploration. Design thinking determined what mattered.



WHAT’S NEXT

Experimenting, testing,
& iterating.




Talk to mothers who stopped

Pain, exhaustion, low supply, work pressure — these are all real quit factors. The bet is that what tips mothers over the edge is facing those moments without context: nothing telling them whether what they're experiencing is normal, temporary, or a sign to stop. Five interviews with mothers who quit in the first six weeks would test one question: would knowing have changed anything?


Test the concept itself — not the UI

Before showing any screens, present two product concepts: one data-forward ("here's your output, trends, and averages") and one coaching-forward ("here's what your numbers mean and whether you're on track"). Do mothers actually prefer the confidence framing — or is that a projection of what they should want?


One-handed interaction audit

Logging a session at 3am, one-handed, with a baby attached, is the primary use case — and it hasn't been observed. Even a rough paper prototype session with a mother would expose friction the current flow almost certainly has. Every extra tap at that moment is a reason to abandon the app entirely.


Get real experts to weigh in

The contextual guidance needs a lactation consultant's eye before anything ships. Confidence is only useful if it's accurate.

BACK HOME