Design a refrigerator for the blind

Q: Design a refrigerator for the blind.

A worked PM interview answer with the right framework, a strong and weak example, and why 2026 AI changes the correct answer entirely.

This question is a trap for candidates who reach for hardware modifications before understanding who the user actually is. The most common answer adds Braille labels and a door-open beep, then stops. That answer fails not because the features are wrong, but because the candidate skipped the work that makes features meaningful: a real user segment, a grounded problem, and a viable path to market.

What the interviewer is actually testing

The refrigerator question is a proxy for three skills: user segmentation (can you resist the fictional monolith “blind person”?), problem clarity (can you name the real unmet need before proposing solutions?), and the viable/lovable tension (does your solution work without being patronizing?).

In 2026, there is a fourth check: are your feature ideas technically obsolete? Samsung’s CES 2026 Bespoke Family Hub ships Gemini-powered food recognition and Bixby voice control as a production feature. If your answer is “add a voice assistant to the fridge,” you have proposed something that already exists on store shelves.

Structure a strong answer

strong

"One clarifying question before I start: are we designing a new fridge from scratch, or a software and hardware layer that works on existing fridges? I'd scope to the retrofit layer, because there are hundreds of millions of fridges already in homes and that's where the addressable market actually is. A purpose-built fridge needs OEM partnerships and a 5-to-10-year replacement cycle to reach users. A retrofit kit ships in weeks.

On users: blindness is not a single persona. The majority of blindness is acquired, not congenital. Macular degeneration, glaucoma, diabetic retinopathy, and cataracts skew heavily toward adults 60 and older. The CDC puts significant vision impairment in the US at about 12 million people 40 and over. The congenitally blind user, often younger, high Braille literacy, experienced spatial navigator, exists but is a minority. Braille literacy among blind Americans is roughly 10 percent. I'd prioritize the older acquired-blindness segment because it is larger, faster-growing, and more underserved by current tools.

The three real jobs in priority order: first, know what is in the fridge without opening it repeatedly; second, assess food freshness and safety without a visual spoilage check; third, maintain personal organization across a household where other members do not follow your system. Current tools like Be My Eyes and Microsoft Seeing AI require holding a phone up to each item one at a time. That is the actual gap, not the absence of Braille on the temperature dial.

The solution I'd build: an interior camera array with an on-device vision model that maintains a natural-language inventory, queryable by voice. 'What do I have on the middle shelf?' 'Anything expiring today?' A companion app uses spatial audio to guide retrieval without requiring the user to pre-memorize a layout. An audio confirmation loop prompts household members to confirm where they returned items, which addresses organization drift over time.

On viability: the universal design business case matters here. A voice-queryable fridge inventory is useful when your hands are full, when you are cooking and cannot touch your phone, when your eyes are occupied. The total addressable market is not 12 million, it is much larger, which is what makes this investable rather than a charity play. I'd build retrofit first for faster time to impact and lower cost barrier, knowing the older segment is unlikely to buy a new appliance specifically for this.

The one metric I'd prioritize: percentage of fridge interactions completed without requiring a separate phone camera assist, because that is the actual job being replaced."

weak

"I'd clarify who we're designing for, then add Braille labels to the temperature dial, a beep when the door is left open, and tactile bumps on shelves to mark food zones." This fails on four counts. It treats all blind users as a single persona, ignoring that most cannot read Braille. It jumps to hardware modifications without identifying the most painful unsolved problem. It skips viability entirely. And the features proposed are 1970s-era thinking: door-open beeps already ship on most fridges, and Braille knobs address a friction that does not rank in the top three for this user group. The interviewer hears "I bolt features onto existing form factors" rather than "I question whether the existing form factor is right for this user."

The 2026 reframe

The question has materially changed since 2019. Multimodal AI at consumer hardware cost makes real-time interior scanning and natural-language inventory retrieval feasible today, not as a speculative concept but as a shipping product. The correct framing is no longer “how do we bolt accessibility onto a sighted-person appliance?” It is: what does food storage and retrieval look like when voice is the primary modality, and what does AI make possible that was previously cost-prohibitive?

Lovable, in this context, means the system anticipates needs without becoming obnoxious. A morning audio briefing that surfaces items expiring today is useful. An alert every time you open the door is not. The distinction matters for the interview: a system that works but feels patronizing will not get used. Proactive without intrusive is the design constraint the strong answer names explicitly.

What interviewers penalize

Three specific failure modes appear consistently in Glassdoor reports from Amazon, Google, and Stripe interviews:

Spending several minutes on clarifying questions without committing to a direction. One clarifying question signals precision. Four signals avoidance. Ask the one that actually changes your answer (new fridge vs. retrofit), then commit.

Proposing a purpose-built fridge with no viability reasoning. Every company that asks this question has experience with hardware products and will probe the channel strategy and replacement cycle.

Treating all blind users as identical. Congenitally blind users and recently blind adults have different mental models, adaptation strategies, and friction points. Collapsing them into one persona signals the candidate has not thought about actual users.

The alarm-clock variant

This question belongs to a category: design for a constrained user who cannot use the primary modality the product assumes. The alarm-clock-for-the-blind variant tests the same framework. The structural move is identical in both cases: segment the user, name the real jobs in priority order, propose a coherent solution system rather than a feature list. For the underlying lens on why viable and lovable are now harder questions than feasibility, see feasibility is free and lovable, not just usable.