Take the wheel
How much human work does a minor car accident really cost? Help us find out — and test whether AI can reduce that work without losing the quality of the accident record.
The work is real.
But the data can’t see it.
Administrative records capture dates, statuses and payments — but not the active human effort behind each accident report. Explaining fields, clarifying circumstances, re-entering data: that work stays invisible.
64.16% on average for Bulgaria (2021–2025)
Estimated residual share of motor TPL property claims not attributable to police-registered accidents. Model-based estimate, not a direct count of bilateral reports. Sources: FSC and Ministry of Interior; own calculations.
How the experiment works
Pairs of participants document the same simulated accident — one as driver A, one as driver B. We compare standard reporting with AI-assisted reporting.
Watch
A short visual simulation of a minor accident. No list of facts — just what you see.
Take a role
You become driver A or driver B, working in a pair.
Report
Describe what happened — with standard reporting or with an AI assistant.
Done
We measure the effort it took and how complete and faithful the record is.
Effort saved, quality kept
Less effort counts only if the quality of the information holds.
Human effort
- Active human time H1 ↓
- Number of operations H2 ↓
Information quality
- Factual completeness H3 ↑
- Contradiction detection
- Faithfulness to statements H3 ≥
- Correct reporting eligibility
What the AI does — and doesn’t
Listens to each driver separately
Each driver gives an independent statement. The AI extracts factual assertions and compares them.
Asks neutral questions
When something is missing, it asks. Silence is never treated as agreement.
No decision on fault
Original statements stay unchanged. If accounts materially contradict, bilateral reporting stops.
Where we are
Framework
The experimental framework is presented at ITEMA 2026.
Experiment
Participant pairs, four scenarios, standard vs. AI-assisted reporting.
Results
A reusable method for researchers. A decision basis for insurers.
Take the wheel
Want to be driver A or B — or help make the experiment happen? Get in touch.
Nadezhda Minkova
University of National and World Economy
Business Faculty
ORCID 0009-0001-2209-1033