Convergence Radar

← Feed

F

On-Device Scam-Call Detector for Elders (iOS)

20/100

An iOS app that live-transcribes calls/voicemail on-device and flags bank/government impersonation scripts, sold to adult children protecting aging parents.

Kill. · created 2026-07-14 12:45 UTC

aiandroidtoo complexrevisit later

Scorecard

newness 6/10
convergence 6/10
demand evidence 4/10
existing spend 4/10
solo feasibility 3/10
speed to mvp 4/10
speed to revenue 3/10
distribution 2/10
competitive gap 3/10
expansion 3/10
founder fit 2/10

Penalty flags
marketplace approval risk long trust cycle too complex platform policy risk adequate free path pii risk (−18 from raw 38)

Opportunity brief

What changed
Apple's SpeechAnalyzer API reportedly delivers Whisper-class on-device speech-to-text at zero marginal cost, making continuous, private transcription economically feasible without cloud costs or shipping voice off-device (source: get-inscribe benchmark — vendor blog, treat as HYPOTHESIS not verified fact).
Why now
FTC reports $3.5B in imposter-scam losses in 2025 (~3x since 2020), a large and growing pain, arriving at the same moment on-device transcription becomes free — the classic capability×complaint convergence.
Converging signals
Two signals: (1) a quantified, growing pain (FTC $3.5B imposter losses) and (2) a new cheap on-device capability (Apple SpeechAnalyzer). Both are cited; the convergence is real conceptually.
Customer pain
FACT (FTC): imposter scams cost consumers $3.5B in 2025. HYPOTHESIS: adult children of at-risk elders feel acute anxiety and would pay for protection — no demand_evidence array was provided (empty), so willingness-to-pay is unproven here.
Who pays
Proposed: adult children (consumer subscription) and credit unions/banks (white-label). Both are discretionary consumer/relationship-sales channels the founder normally avoids; banks are a long enterprise trust cycle.
Solved today
Carrier spam labeling (AT&T/T-Mobile ActiveArmor, Verizon Call Filter), Apple's built-in Live Voicemail transcription and Silence Unknown Callers, Google's on-device call-scam detection (Pixel), and after-the-fact fraud reimbursement by banks.
Why current solutions are bad
Existing tools label unknown NUMBERS, not the CONTENT of a call in progress; they don't parse the impersonation script itself. But the gap is narrowing fast as OS/carrier owners fold detection in natively.
Proposed product
An iOS app that transcribes voicemail and (attempts) live-call audio on-device, runs a classifier tuned to bank/government/'fraud-department' urgency+payment scripts, and surfaces a live warning card plus alerts to a linked adult child.
MVP version
Voicemail-only scam scorer (NOT live-call): ingest voicemail transcripts via SpeechAnalyzer, classify against a scam-script corpus, and text the adult child a risk alert. Sidesteps the live-call-audio blocker.
30-day build
Build a labeled corpus of real scam-call phrasings (FTC transcripts, YouTube scam-baiter recordings, public complaint text); prototype the classifier; validate precision/recall on held-out real scam + legit bank calls (the KILL TEST).
60-day build
Ship voicemail-only TestFlight to a small group of adult children; measure false-alarm rate and whether users keep it enabled.
90-day revenue plan
Convert to paid subscription ($5-10/mo) via App Store; attempt a credit-union pilot conversation for white-label — but revenue in 90d is uncertain given consumer CAC.
Distribution path
Weak spot: reaching elders/adult children means paid consumer acquisition, ASO, or partnerships — an ad-spend-heavy channel the founder explicitly avoids. No demonstrated-value/self-serve wedge exists for a phone-safety app.
Pricing hypothesis
$5-10/mo consumer sub, or per-seat white-label to a credit union.
Technical difficulty
HIGH and possibly fatal: iOS does NOT give third-party apps access to live in-call audio (CallKit exposes metadata, not the audio stream). Real-time transcription of an active phone call is effectively impossible on iOS today; the epiphany's core 'listen to a call' premise is not buildable. Only voicemail (post-call audio the user imports/shares) is feasible.
Legal / regulatory risk
Two-party call-recording consent laws in ~11 states; recording/transcribing calls raises wiretap exposure. PII risk from processing elders' call audio, even on-device.
Platform dependency
Total dependency on Apple: App Store approval, and Apple/Google/carriers are actively shipping equivalent native scam detection — high risk of being obsoleted or deplatformed.
Founder fit
LOW. This is a consumer social-adjacent, ad-spend-driven, distribution-hard app with no government-portal/forced-buyer shape and no filing/claim monetization — the opposite of the founder's proven FMCSA-style edge. It plays to none of his strengths.
Breakout potential
If live-call audio were accessible it could be big, but that door is closed on iOS. Voicemail-only is a modest niche utility, easily copied and likely absorbed by the OS.
Final recommendation
PASS / KILL. The premise depends on live-call audio access iOS does not grant, the residual voicemail version is a thin utility racing free OS features, and it fits none of the founder's forced-filer/public-money/claimable-money edges. Redirect the FTC-imposter-fraud interest toward a B2B or public-records angle (e.g., a data/report product on scam-script patterns for credit-union fraud teams) rather than a consumer phone app.
Next action
Do not build. If interested in the fraud vertical, validate a credit-union-facing fraud-pattern data product against a real buyer before writing code.

Kill arguments (adversarial)

  • iOS gives third-party apps NO access to live in-call audio — the central 'real-time call detection' product cannot be built on iOS; only post-hoc voicemail works, gutting the value prop.
  • Platform owners (Apple, Google, carriers) are already shipping native on-device scam/spam detection for free; a solo consumer app gets absorbed within a release cycle.
  • Consumer distribution to elders/adult children requires ad spend and trust-building the founder avoids; no self-serve demonstrated-value wedge and empty demand_evidence (WTP unproven).
  • Call-recording two-party-consent and wiretap law plus elder-PII sensitivity create real legal exposure for a solo operator.

Competitors

Apple (built-in) (link) — Live Voicemail transcription + Silence Unknown Callers already on-device and free.
Google Pixel Scam Detection (link) — On-device AI that detects scam-call patterns in real time — the exact feature, already shipped by the platform owner.
Carrier apps (ActiveArmor / Call Filter) (link) — Free carrier-level spam/scam labeling bundled with service.

Source citations (facts)

FTC Data Show People Reported Losing $3.5 Billion to Imposter Scams in 2025 — Imposter-scam losses reached $3.5B in 2025, validating the scale of the pain.
Apple's SpeechAnalyzer API benchmarked against Whisper — Vendor blog claims Apple's on-device SpeechAnalyzer approaches Whisper-class accuracy at zero marginal cost — unverified, treat as hypothesis.

Actions