Pantry · Product Design

The core product challenge

Every smart pantry product fails at the same point: data entry fatigue.
Users enthusiastically scan barcodes for two weeks, then stop. The system goes stale and trust collapses. Pantry solves this with a layered capture strategy where no single method carries the full burden. The system must work well even if the user does very little.

Capture methods — 6 signals

Priority 1

Receipt + Invoice Parsing

Photo of till slip, email receipt forwarding, or direct loyalty account integration with Checkers and Woolies. SKU-level detail, exact quantities, prices, store, date, and time.

Confidence 9/10 Low effort MVP

Priority 2

WhatsApp Micro-reports

Grace already WhatsApps what's running low. Bot parses the same message, adds to list, confirms back. This is the killer feature for the SA market — zero behaviour change required.

Confidence 9/10 Near-zero effort MVP

Priority 3

Voice Reporting (Nest Hub)

Household members tell the Nest Hub when something is running low or finished. Fuzzy language ("the crunchy cereal", "Sipho's yoghurt") resolved using purchase history and context.

Confidence 8/10 Minimal effort MVP

Priority 4

Predictive Consumption Model

System learns consumption rates from purchase history. If 2L milk is bought every 4 days, it predicts when it will run low. Requires 4–6 weeks of purchase data to establish cadence.

Confidence 7/10 Zero (passive) v1.1

Priority 5

Barcode Scanning

Camera-based item lookup and add. Best for one-time pantry bootstrap, unpacking after a big Makro shop, or scanning empty containers before discarding. Not the primary capture method.

Confidence 5/10 Active (optional) v1.1

Priority 6

Smart Device Integration (IoT)

Weight sensors, temperature sensors, Home Assistant via MQTT. Real-time inventory tracking without any human action. Phase 2 for households, core for Pantry Pro restaurants.

Confidence 10/10 Zero (hardware) v3+ / Pro

Layered capture strategy

#	Method	Signal type	Phase	User effort	Confidence
1	Receipt / invoice parsing	Items In	MVP	One-time setup	9/10
2	WhatsApp micro-reports	Items Out / Running Low	MVP	Near zero	9/10
3	Voice reporting (Nest Hub)	Items Out / Running Low	MVP	Minimal	8/10
4	Predictive consumption model	Estimated depletion	v1.1	Zero (passive)	7/10
5	Barcode scanning	Items In / Out	v1.1	Active (optional)	5/10
6	Smart device / IoT integration	Real-time inventory	v3+ / Pro	Zero (hardware)	10/10

Data confidence framework

High confidence (green) — In stock

"You bought 2L milk 2 days ago. Likely still have some."

Receipt-confirmed purchase within the predicted consumption window.

Medium confidence (amber) — Running low

"You might be running low on eggs."

Purchase confirmed but nearing predicted depletion date based on consumption pattern.

Low confidence (red) — Probably out

"You're probably out of bread."

Past predicted depletion date with no repurchase signal. System is honest about uncertainty — never says "you are out of" unless confirmed.

Unknown (grey) — No data

"This item hasn't been tracked yet."

No purchase history. Item purchased at an untracked store or not yet added to the system.

Voice design principles

Warm, not robotic

Pantry sounds like a helpful household member, not a call centre IVR. Uses first names when voice-matched. Friendly to kids ("Great idea, Sipho!"), professional to adults.

Brief by default

Responses are short. "Milk added." not "I have successfully added full cream milk, 2 litres, to your household shopping list." Detail is available on the screen or by asking.

Forgiving and fuzzy

"The crunchy cereal" and "Sipho's yoghurt" are resolved using purchase history and household context. Never ask for a formal product name when history can infer it.

Speaker-aware

Recognises household members by voice (Google Voice Match). Adjusts behaviour: kids' requests go to approval queue, financial data only shared with Lerato and Thabo.

Multilingual — English + Zulu

Responds in the language spoken to it. Zulu product names mapped to store catalogue. Gogo's "Ngidinga impuphu" is understood and confirmed back in Zulu.

Non-blocking

Never asks more than one clarifying question. If unsure, make the best guess and confirm. Offer WhatsApp as fallback if voice isn't working. One retry, then alternative offered.

Honest about uncertainty

"You're probably low on eggs" not "You are out of eggs." The system can't know for certain — it shouldn't pretend otherwise. Trust comes from calibrated honesty.

Voice initiates, screen confirms

Voice can initiate an order but never complete it. Final checkout always requires visual confirmation. Proactive alerts are screen-first (card), not voice-first — nobody wants random announcements.

Voice interaction scripts

Grace Batch item report — multiple items at once

Grace:"Hey Google, tell Pantry: we need Sunlight, eggs, and the blue milk."

Pantry:"Adding three items: Sunlight dish soap, eggs eighteen pack, and full cream milk two litres. All on the list."

Batch input is critical for Grace's workflow. "Blue milk" resolved to full cream from purchase history. No follow-up questions when history is clear.

Sipho (7) Fuzzy product request — kids' queue

Sipho:"Hey Google, tell Pantry I want more of the crunchy cereal!"

Pantry:"Hey Sipho! I think you mean Kellogg's Crunchy Nut. I'll ask Mom to add it to the list."

Kids' requests route to Lerato's approval queue, not added directly. "Crunchy cereal" resolved from household purchase history. Warm, encouraging tone for kids.

Gogo Zulu language input

Gogo:"Hey Google, tshela iPantry ngidinga impuphu ne-sugar beans."

Pantry:"Kulungile, Gogo. Ngifake impuphu ne-sugar beans ohlwini."

Responds in the language spoken. Zulu product names mapped to store product catalogue. Cultural inclusion is non-negotiable, not optional.

Lerato Recipe check before cooking

Lerato:"Hey Google, ask Pantry: do I have enough for my chocolate cupcakes?"

Pantry:"Checking your famous chocolate cupcakes recipe. You have flour, sugar, and eggs. You're missing: cocoa powder and buttermilk. Want me to add them to the list?"

Lerato:"Yes."

Pantry:"Added. Cocoa powder R38.99 at Woolworths, buttermilk R24.99 at Checkers."

Matches "my chocolate cupcakes" to saved recipes. Includes store routing with ZAR pricing in the response.

Lerato Morning status check

Lerato:"Hey Google, ask Pantry what we need."

Pantry:"You've got 7 items on the list. Top priority: milk and eggs are probably out. Grace reported you're low on dish soap and rice. Sipho wants yoghurt. Want me to read the full list?"

Lerato:"No, that's fine. Send it to my phone."

Pantry:"Sent. You can review it in the app."

Voice gives the summary. Detail is always on the phone. Surfaces Grace's reports and kids' requests as attributed items — not just an anonymous list.

Thabo Quick stock check

Thabo:"Hey Google, ask Pantry: do we have milk?"

Pantry:"Probably not. You bought milk 4 days ago and usually run out after 3. It's already on the list."

Predictive status based on consumption model. Honest about uncertainty — "probably not", not "no". Confirms it's on the list so no duplicate action needed.

Voice personality — do and don't

✓ Do

"Milk added." / "Got it, Grace."

"You're probably low on eggs" (honest)

Use first names when voice-matched: "Got it, Grace"

Friendly to kids: "Great idea, Sipho!"

Respond in the language spoken (English or Zulu)

Make the best guess, then confirm

✗ Don't

"I have successfully added full cream milk, 2 litres, to your household shopping list."

"You are out of eggs" (can't know for certain)

Generic: "Item has been added to the list"

Same formal tone for kids as for adults

Always respond in English regardless of input

Ask two clarifying questions in a row

The four interaction surfaces

🔊

Google Nest Hub

Grace · Kids · Gogo · Lerato

Kitchen, hands busy, quick interactions

Add items, report stock, quick status checks. Voice-first, screen-supported. Proactive alerts as screen cards (silent).

💬

WhatsApp Bot

Grace · Thabo · Lerato

On the go, at the store, asynchronous

Task delegation, shopping lists, confirmations, receipt capture via photo. Zero app install required.

📱

Mobile App (PWA)

Lerato · Thabo

Commute, couch, quick review moments

List management, approvals, ordering, notifications. Full inventory view. In-store checklist mode for Thabo.

🖥

Web Dashboard

Lerato

Evening planning, laptop

Analytics, budget tracking, monthly planning, household settings. Deep analysis and trend view not available on mobile.

Visual design direction

Recommended ✓

Kitchen Table

Warm, homey, tactile. Terracotta primary (#C4653A) + sage green on linen. Plus Jakarta Sans. Ingredient-style chips.

Best for: premium household feel, strong differentiation from competitor apps.

Alternative

Fresh Market

Clean, modern, fresh. Deep green (#2D6A4F) + warm coral on white. DM Sans. Store-branded accents.

Best for: broadest audience appeal, grocery-focused users.

Alternative

Family Hub

Friendly, accessible, colourful. Warm indigo + amber on soft cream. Nunito. Family avatars, colour-coded categories.

Best for: multi-member households and accessibility across all ages.

Note: Kitchen Table was recommended as the primary direction. The live app currently uses Plus Jakarta Sans as the typeface with a cyan accent (#00ACC1) — aligned with Kitchen Table's warmth principle but with a lighter colour system for cross-surface readability.

Capture Methods + Voice Design