Arnab Arya Arnab Arya
All work
Research · Strategy

Smartphone Sensor App Opportunities

A ranked, evidence-backed playbook for a solo developer: where the money actually is across the phone's accelerometer, microphone, camera, and the under-used "small" sensors — and which ideas to avoid.

~12 min read 12 ideas ranked Top pick: Acoustic machine fault diagnosis

TL;DR

  • The single best risk-adjusted opportunity is B2B/B2G sensor-data licensing in the spirit of RideIndex — road/infrastructure quality, acoustic machine diagnostics, and indoor-context data — because these avoid FDA regulation, monetize highest in the USA, and build data moats. The clearest proof the model works is RoadBotics (smartphone-camera road assessment), acquired by Michelin in July 2022.
  • Avoid building medical-diagnosis apps as your headline product. Camera blood pressure (Binah.ai, Riva Health) still has no US FDA clearance as of 2025–2026, while the one clear medical success — ResApp's acoustic sleep-apnea screener — required clinical trials and a 510(k) before Pfizer acquired it at ~A$180M (~US$179M). Health is real money but the wrong first project for a solo dev.
  • For a strong solo developer with ML skills, the highest-leverage plays are acoustic "Shazam for machines" (HVAC/appliance/car fault detection), barometer+IMU vertical-context apps, and camera-based agricultural/material grading — all underexploited, feasible today, and monetizable via subscription or B2B licensing.

Key findings

1

The most defensible money is in selling data and inspections, not consumer gadgets. The RideIndex concept generalizes into a category that has already produced exits: RoadBotics — a 2016 Carnegie Mellon spinout that rated pavement from smartphone imagery, enabling 250+ governments across 14 countries and raising US$11.4M — was acquired by Michelin in July 2022, its 26 employees joining Michelin's Mobility Intelligence unit. The usage-based-insurance telematics market (running on the same accelerometer/gyroscope/GPS signals) was valued by Mordor Intelligence at US$30.31B in 2025, projected to reach US$69.18B by 2031 (14.76% CAGR). Municipal and infrastructure buyers pay for inspection data they currently collect manually.

2

Health-sensing research is spectacular but commercially gated by regulation. Two decades of UbiComp/CHI/MobiSys work proved phones can measure heart rate (PPG), blood pressure, anemia (HemaApp), jaundice (BiliCam/neoSCB), sleep apnea (ApneaApp sonar), cough/COVID, Parkinson's (voice), and fall risk (gait). Almost none became durable consumer businesses because medical claims trigger FDA regulation. The lesson: ship "wellness/screening"-framed versions, sell B2B to clinics/researchers, or pick non-medical adjacencies.

3

Several genuinely novel sensor capabilities remain commercially unexploited. Standouts with weak or no commercial presence: smartphone-microphone machine fault diagnosis for consumers/SMBs (only OEM tools like Škoda's Sound Analyser exist); magnetometer metal/rebar/stud sensing done well; barometer-based vertical context (floor detection, stairs-vs-elevator, fall confirmation); acoustic environmental/biodiversity mapping; and camera spectral grading of food/materials. These map directly to the "clever, non-obvious" criterion.

The ideas, by sensor

Five families — vibration & motion, acoustic, camera, the under-used "small" sensors, and sensor fusion. Each idea notes how it works, who pays, and an honest assessment.

A

Vibration & Motion — Accelerometer + Gyroscope

A1

Infrastructure & asset vibration monitoring ("RideIndex for structures")

HowMEMS accelerometer recovers modal frequencies; ML detects structural damage vs. a reference (validated on real post-tensioned concrete bridges). Extend from roads to bridges, buildings, scaffolding, escalators, rooftop HVAC.
Who paysMunicipalities, DOTs, facilities managers, insurers, construction firms — B2B/B2G licensing + per-inspection SaaS.
VerdictLarge market, moat from labeled vibration data + municipal relationships, low regulatory risk. Well-suited to a strong solo ML dev.
A2

Gait / fall-risk screening for older adults

HowAccel + gyro capture cadence, stability, symmetry, gait-complexity entropy; ML classifies fall risk (STURDY cohort and others — sometimes beats in-clinic tests).
Who paysMedicare Advantage plans, senior-living operators, physical therapists, caregivers (B2B2C).
VerdictBig USA market, but waist-phone accuracy lags wrist wearables and "assessment" flirts with medical-device territory. Best as wellness, sold B2B to senior care.
A3

Insurance / fleet telematics scoring

HowTrip detection + harsh-event scoring from IMU+GPS — the same engine the incumbents use.
Who paysSaturated for consumers (State Farm, Allstate/Arity, GEICO). Solo angle = a white-label telematics SDK for niche players: driving schools, micro-fleets, delivery co-ops, teen-driver apps.
VerdictDecent B2B revenue; low novelty.
B

Acoustic — Microphone (incl. ultrasound sonar)

Top novel pick B1

"Shazam for machines" — acoustic fault diagnosis

HowRecord a running engine/HVAC/pump/appliance; an MFCC/CNN spectrogram classifier flags bearing wear, belt slip, misfire, refrigerant issues. Phone mics can detect bearing/motor faults (limited at very low frequencies).
MarketMachine condition monitoring ~US$3.78B (2025) → US$6.58B (2033), 7.0% CAGR; acoustic/ultrasound the fastest-growing segment — yet dominated by fixed industrial sensors. Phone-based diagnosis is essentially uncommercialized (only OEM-tied Škoda Sound Analyser, ~90% accuracy).
Who paysDIY mechanics, independent + HVAC shops, appliance-warranty firms, used-car buyers. Freemium B2C + B2B licensing.
VerdictHigh novelty, strong moat (proprietary fault-sound corpus = data network effect), feasible now, low regulation. Best overall fit for a solo ML dev. Challenge: collecting labeled fault audio.
B2

Acoustic environmental & biodiversity sensing

HowClassify bird/insect/frog calls, urban noise, or mosquito wingbeats (HumBug shows even budget phones ID mosquito species by wingbeat).
Who paysPublic-health/vector surveillance (India/Africa/SE Asia), environmental consultancies, smart-city noise contracts, birders (cf. Merlin).
VerdictMixed monetization: birding proven; mosquito surveillance high-impact but grant/NGO-funded; noise-mapping a B2G niche.
Regulated / crowded B3

Contactless respiration / sleep sonar

HowEmit inaudible >18kHz FMCW chirps, measure chest-movement reflections → breathing rate, apnea events (UW ApneaApp: 90s sensitivity/specificity).
RealityThe passive-mic approach won: ResApp's SleepCheckRx got FDA 510(k) (2022) on a 220-patient trial (89.3% sensitivity, 77.6% specificity); Pfizer acquired ResApp (closed Sep 2022, ~A$180M). Apple Watch added FDA-cleared apnea detection in 2024.
VerdictMedical regulation + entrenched/acquired competitors make apnea a poor headline. Ship a non-diagnostic snore/breathing logger or baby/elderly monitor (with disclaimers) instead.
B4

Hearing test / audiometry & tinnitus tools

HowValidated apps exist (Mimi, uHear) with mixed accuracy; calibration and ambient noise are the hard problems.
VerdictThin for consumers; viable as a B2B screening tool for LMIC community health workers. Lower priority.
C

Camera — Computational imaging + ML

Novel · B2B-friendly C1

Agricultural / food grading via spectral-ish imaging

HowCamera (+flash, RGB/HSV features, optional UV chlorophyll-fluorescence) estimates ripeness, sugar/Brix, freshness, crop disease (ACM IMWUT "FruitPhone"; Nature/Sci Reports spectrometry).
Who paysGrocery chains, produce distributors, packhouses, farmers (esp. India/Europe/LatAm), restaurants — B2B licensing + per-seat SaaS.
VerdictHigh novelty, non-medical, moat via crop-specific datasets. Risk = accuracy without add-on hardware. Good solo-ML fit, strong in India/Europe.
Medical · regulated C2

Low-cost diagnostics: anemia / jaundice / wounds

HowHemaApp (hemoglobin from finger video) and BiliCam/neoSCB (newborn jaundice from sclera/skin, validated on 300+ babies in Ghana) are scientifically strong — but squarely medical.
VerdictOnly viable as research tooling / B2B to LMIC hospitals, or via an NGO. High impact, uncertain monetization, high regulatory risk.
C3

Camera measurement / room scanning

HowSaturated on iOS (Apple Measure, AR Ruler apps, RoomPlan/LiDAR).
VerdictDifferentiate only via a vertical niche: insurance claim docs, moving-quote estimation, ADA-compliance audits, material-volume estimation. Low novelty unless verticalized.
No FDA clearance C4

PPG vitals (heart rate / HRV / respiration)

HowFingertip-camera heart rate is mature and commoditized (Azumio). Blood pressure is the prize — but no smartphone-camera BP app has US FDA clearance as of 2025–2026 (Binah.ai = wellness-only; Riva Health raised US$15.5M in 2021 but remains in the FDA process).
VerdictAvoid BP claims; HRV/stress wellness only, low margin.
D

The under-exploited "small" sensors — Magnetometer, Barometer, Light, NFC

D1

Magnetometer metal/rebar/stud/EMF sensing

HowPhones reliably detect ferrous metal via magnetic flux; existing apps are crude novelties.
Who paysA pro tool for trades — rebar locator, stud finder, buried-pipe hint, EMF survey for home inspectors — sold one-time or as a pro subscription.
VerdictModest market, low regulation, medium novelty. Good small solo win.
Skip as a startup D2

Indoor positioning via magnetic fingerprinting

HowMagnetic-fingerprint positioning to ~1–2m (MaLoc).
CautionIndoorAtlas (Finland, 2012) raised ~US$17.9M incl. a US$10M Baidu round, never scaled; magnetic lost to BLE/UWB/Wi-Fi. Market is large (~US$12B 2024 → ~US$31B 2029) but venue-by-venue mapping is a solo killer.
VerdictSkip as a startup; possibly useful as a feature.
Underexploited D3

Barometer + IMU vertical-context apps

HowBarometer detects ~1m altitude changes — floor detection, stairs-vs-elevator, "which floor am I on", fall confirmation (pressure + impact), elevation-gain accuracy — more energy-efficiently than accelerometers.
Who paysIndoor-navigation SDKs, emergency-services/E911 vertical location (FCC mandates = regulatory tailwind), senior-safety, hiking/fitness.
VerdictNovel, low regulation, feasible. An E911-grade floor-detection SDK is an interesting B2B wedge. Good solo-ML fit.
D4

Ambient light → circadian / light-hygiene tracking

HowLux logging for circadian health, SAD, light-therapy compliance, plant-care surveys — but the sensor is coarse and screen-occluded.
VerdictNiche wellness subscription; low priority.
E

Sensor Fusion — Context / activity

Domain-specific fusion, not a generic context engine

WhyActivity recognition, transport-mode detection, and context awareness are mostly commoditized into OS APIs.
VerdictThe fusion worth pursuing is domain-specific — combining IMU+baro+mic for the machine-diagnosis or fall-confirmation use cases above, not a generic engine.

Ranked composite scoring

Each idea scored 1–5 (5 best) on Revenue, Solo-Build feasibility, Technical feasibility, Moat, Low-Regulatory-risk, Time-to-MVP, and Novelty. "Composite" rank weights revenue, moat, and novelty most heavily. Click any column header to sort.

Rank ▲ Idea Rev Build Tech Moat LowReg MVP Novelty Verdict
Score: 1 2 3 4 5

Recommendations

A staged path: a fast solo win to build distribution and a data pipeline, then commit to the flagship, then run a parallel B2B data play.

Stage 1 · 0–3 months

Ship a fast solo win to build audience and a data pipeline

Build D1 (magnetometer pro trades tool) or a focused acoustic logger as a paid utility — solo-only (no backend), low-regulation, fast to MVP. Use it to start collecting the labeled sensor data the bigger plays need. Threshold to proceed: ~US$1–2k/mo or 5k+ engaged users validates distribution.

Stage 2 · 3–9 months

Commit to the flagship — B1 Acoustic Machine Fault Diagnosis

The RideIndex-caliber bet. (1) Pick ONE vertical with abundant, labelable failure sounds and motivated payers — recommended order: HVAC/refrigerant → car engines → washing machines/pumps; (2) partner with 2–3 repair shops to collect labeled fault audio (the moat); (3) train a spectrogram CNN; (4) launch a freemium consumer app + pitch B2B licensing to a warranty company or repair chain. Benchmarks: if held-out accuracy can't clear ~85%, narrow the fault set or add calibration; if consumer retention is weak, pivot fully to B2B.

Stage 3 · 9–18 months

Parallel B2B data play — A1 Infrastructure vibration OR C1 Ag grading

Choose by network: US municipalities/facilities → A1; ag/produce buyers or India/Europe → C1. Both monetize via B2B/B2G licensing with data moats. Validate with one paid pilot before scaling.

Cross-cutting monetization guidance

  • USA: prioritize B2B/B2G licensing and prosumer subscriptions (machine diagnosis, infrastructure, telematics SDK, E911 floor-detection).
  • India / SE Asia / LatAm: ag grading, mosquito/vector surveillance (grant-funded), and low-cost diagnostics have impact but monetize via NGOs/government, not consumers.
  • Europe: ag grading, infrastructure (mind GDPR for crowd-sensed location data), and condition monitoring.
  • Best models by idea: machine diagnosis = freemium + B2B; infrastructure/ag = B2B/data licensing; trades tools = one-time/pro subscription; anything health = avoid direct B2C medical claims.

Hard rules to avoid value-destroying mistakes

  1. Never make diagnostic medical claims without a regulatory plan. Frame health features as "wellness," "screening," or "logging," or sell as research tooling. The ResApp path (clinical trial → 510(k) → acquisition) shows the real cost of going medical.
  2. Treat labeled data as the moat. Generic sensor reading is copyable in a weekend; a proprietary corpus of fault sounds / vibration signatures / crop spectra is not.
  3. Privacy: crowd-sensed location/audio triggers GDPR/CCPA. Do on-device inference where possible; license aggregated, anonymized data.

Caveats

  • Market-size figures vary 2–10× across publishers. Cited mid-range, named estimates (Mordor for UBI, Grand View for condition monitoring, MarketsandMarkets for indoor location) are directional, not precise. The IndoorAtlas funding figure differs by source (~US$17.9M per Tracxn vs. higher elsewhere).
  • Sensor accuracy is device-heterogeneous. Mics, barometers, and magnetometers differ by phone model — cross-device generalization and calibration is the recurring hard problem and the biggest technical risk.
  • Several "research successes" are lab-validated, not field-proven at scale. Apnea sonar, camera BP, and camera diagnostics show strong controlled-setting numbers that degrade in real-world noise, lighting, and motion. Plan for a robustness gap.
  • Regulatory status changes fast. FDA clearances for camera-BP and similar may arrive after this writing; re-check before committing to any health vertical.
  • A few sub-topics (soil/water colorimetric analysis, precise magnetometer pipe-detection accuracy) could not be fully searched; the magnetometer and ag-camera assessments rest on adjacent sources and should be validated with a focused prototype before heavy investment.