AI valuation explained
Inside FairPrice AI's pricing engine — what it does and what it doesn't.
Two layers: engine and AI
FairPrice AI uses a two-layer approach. The first layer is a deterministic engine that calculates price ranges from real data — comparable listings, MRP references, condition scores, and location adjustments. The second layer is Gemini AI, which explains those numbers in plain language.
The deterministic engine
The engine works like an appraiser, not a chatbot. It starts with a known MRP anchor (what the product costs new), applies a depreciation curve based on age and category, blends in real comparable listings from the marketplace, and adjusts for location (metro cities command slightly higher prices).
How comparables are selected
Not all listings are valid comparables. The engine filters aggressively: the product must match brand and model, the condition grade must be close, and the price must fall within the MRP band for that product. A POCO M7 should never be compared to an iPhone listing at a similar price.
Condition scoring
Condition grade alone isn't enough — the engine maps grades to numeric scores (Like New = 95, Excellent = 85, Good = 70, Fair = 55, Poor = 35) and uses those scores in the depreciation calculation. A 'Good' phone is not just a label — it shifts the anchor by a measurable percentage.
What Gemini AI adds
Gemini doesn't invent prices — it explains the ones the engine produced. Given the product label, MRP, fair range, condition, age, comparables count, and market trend, Gemini writes the explanation ('The original price was ₹X; after 14 months in Good condition it has depreciated ~42%…'), buyer verdict, seller tip, and negotiation talking points.
What it can't do
The engine will not produce a result when evidence is insufficient. If a product can't be identified, if there are no comparable listings, and no MRP reference is available, it returns INSUFFICIENT_DATA rather than inventing a number. This is intentional — a wrong number is worse than no number.
Confidence and what it means
Every result includes a confidence score. High confidence means strong identity match + multiple real comparables + a reliable MRP reference. Low confidence means sparse data. Use low-confidence estimates as a rough signal, not a final price.
FairPrice AI · July 2026