The South African used-car market is valued at approximately R160 billion annually, of which 55.71% flows through unorganised sellers— private individuals, micro-dealers, and informal traders outside organised dealership networks. The dominant friction is condition uncertainty. Hardware-based AI inspection (UVeye, $380.5M raised) solves this at organised dealerships at $100K-$250K per installation — economically impossible for the SA long tail. Visio Inspectis a smartphone-grade equivalent that produces a structured AI-graded condition report from twelve phone photos using Gemini 2.5 Vision in approximately five minutes. The system enforces an Honesty Protocol: if AI confidence on any panel falls below 0.5, that panel is excluded from the aggregate; if overall confidence falls below 0.7, the “Verified by Visio” badge is withheld. We argue this is the minimum viable trust infrastructure for the 55% of the SA used-car market that no incumbent serves.
1. The market gap
South Africa's used-car market is bimodal:
- Organised retail — Motus (R112.6bn revenue, 933 outlets), WeBuyCars (R26.4bn, 15,000 vehicles/month), AutoTrader.co.za, Cars.co.za. Offers roadworthy certification, dealer warranties, structured financing.
- Unorganised retail — private sellers on Gumtree, Facebook Marketplace, OLX; micro-dealers in townships and small towns; informal traders. 55.71% of transactions occur here (Mordor Intelligence, 2025). This is where buyers face the highest uncertainty and the lowest legal recourse.
Globally, the response to this problem has been hardware: drive-through scanning bays from UVeye ($380.5M raised, latest $191M January 2025), Ravin AI, ProovStation, DeGould, Spinframe, Monk AI. Typical capex: $100K-$250K per location + monthly software fees.
This is economically impossible in SA. There are roughly 2,000 independent dealers outside the major groups, plus an unknowable number of private sellers; the average SA used vehicle transacts at R416,082 (AutoTrader, January 2026). No UVeye ROI math works.
Visio Inspect attacks the same problem with the only piece of hardware every SA used-car seller already owns: a smartphone.
2. The system
Visio Inspect captures twelve photos in a guided wizard:
- Front 3/4 left, Front straight, Front 3/4 right
- Right side, Rear 3/4 right, Rear straight, Rear 3/4 left, Left side
- Interior (driver side), Dashboard, Odometer close-up, VIN plate close-up
Each photo is compressed client-side to under 500KB before upload (typical 4-6x compression via canvas JPEG re-encoding). On the server, each photo is sent to Gemini 2.5 Flash via the Vercel AI SDK 6 generateObject function with a Zod v4 schema constraining the response to four fields per panel:
{
damage: "none" | "minor" | "moderate" | "severe",
issues: string[],
conditionScore: 1-10,
confidence: 0-1,
}The prompt explicitly instructs the model to set confidence below 0.6 if photo quality is poor or the panel is not visible. This is the central honesty mechanism.
The aggregator then computes:
- Overall grade A/B/C/D, derived only from panels with confidence ≥ 0.5
- Recon cost band R0-5K / R5K-20K / R20K-50K / R50K+
- “Verified by Visio” badge — granted only if average condition ≥ 7.0 AND every panel has confidence ≥ 0.7
- VIN extraction via Gemini OCR with ISO 3779 check-digit validation
- Odometer reading via Gemini OCR, sanity-checked against vehicle year
3. The Honesty Protocol (three layers)
- Schema constraint. Every panel requires a
confidencefield. The model cannot return a finding without committing to its uncertainty. - Aggregation gate. Panels with confidence below 0.5 are excluded from the average. A vehicle with four blurry photos cannot accidentally receive a Grade A.
- Badge gate. The Verified by Visio badge requires every panel at confidence ≥ 0.7. A single uncertain panel withholds the badge for the entire vehicle. This pushes sellers to retake bad photos rather than gaming the system.
When confidence is insufficient, the report displays “Insufficient data to verify this panel — manual inspection recommended” rather than guessing. This is consistent with the broader VisioCorp Honesty Protocol which forbids fabricating data even when refusing to fabricate produces a worse user experience.
4. Pricing and market access
Visio Inspect is free for the first report. Subsequent reports cost R49 each for private sellers, dealerships can subscribe for unlimited reports at R299/month. At the SA used-car median transaction price of R416,082, R49 is 0.012% of the transaction value— well below any other source of friction in the deal.
5. Comparison to UVeye and global incumbents
| UVeye / hardware AI | Visio Inspect | |
|---|---|---|
| Capex per location | $100K-$250K | R0 (uses seller's phone) |
| Inspection time | ~30 seconds drive-through | ~5 minutes guided wizard |
| Confidence reporting | Vendor-defined | Per-panel, transparent |
| Suitable for private sellers | No | Yes |
| Suitable for micro-dealers | No | Yes |
| Recurring cost to seller | High (subscription + capex) | R49 per report or R299/mo unlimited |
Visio Inspect is not trying to compete with UVeye at premium organised dealerships. It is trying to bring the bottom 80% of the trust pyramid into existence at all, in a market where the alternative is no inspection and no trust whatsoever.
6. Integration with the Visio Auto Suite
Visio Inspect's primary value to VisioCorp is not the standalone R49 reports. It is the trust layerthat makes Visio Trust transactions possible. When a buyer clicks “Buy with Visio Trust” on a dealer listing, the first thing they see is the Visio Inspect grade and Verified by Visio badge for that specific vehicle. Without Inspect, Visio Trust is a payment processor; with Inspect, it is a Carvana-grade purchase experience.
The integration is loose-coupled: vt_transactions.inspect_report_id references vi_inspections.id. If Inspect is offline, Visio Trust degrades gracefully to “condition report unavailable” — which is itself an honesty signal to the buyer.
7. Limitations
- Smartphone constraint: undercarriage, frame straightness, and engine compartment internals are out of scope. Every report carries a disclaimer.
- Gemini Vision is not infallible; confidence scores are themselves estimates. We bias conservative and surface uncertainty rather than masking it.
- Network conditions in rural SA can make the upload slow. Client-side compression mitigates but does not eliminate this.
8. Conclusion
Visio Inspect is a deliberately small, deliberately honest tool. Its goal is not to replace a workshop inspection or to compete with hardware AI. Its goal is to give the 55% of the South African used-car market that has no trust infrastructure at all the cheapest possible upgrade from “no information” to “AI-graded structured report with explicit confidence”. At R0 marginal cost to deploy and R49 per report to use, it is the minimum viable trust layer for the SA used-car long tail — and the foundation on which Visio Trust's escrow + delivery experience is built.
References: AutoTrader 2025/2026 Industry Report; Mordor Intelligence SA Used Car Market 2025; UVeye funding (PRNewswire Jan 2025); ISO 3779 VIN standard; Vercel AI SDK 6 generateObject documentation. Repository: github.com/Iamhamptom/visio-inspect, commit f7edbe4. Live deploy: visio-inspect.vercel.app.
Cite as: Hampton, D. (2026). “Visio Inspect: Smartphone-Grade AI Vehicle Inspection for South Africa's Unorganised Used-Car Market”. Visio Research Labs, VRL-AUTO-005.