Interpretation: This graph shows a negative linear relationship between parallel search count and quote acceptance ratio (QAR) for autos. However, due to a few extremely high values in the search count, the confidence interval of the trend line becomes wide, especially at the higher end. These outliers reduce model precision. The overall trend supports that higher 3+ parallel searches correlates with lower QAR, but the presence of outliers makes this relationship noisier in the untransformed scale.
Interpretation: This graph normalizes the x-axis by applying a logarithmic transformation, which better distributes the data and compresses the influence of extreme values. A clear and strong negative correlation is observed between log(count) and QAR. The confidence band is much narrower than in the original scale, indicating more reliable model fit. This transformation highlights that increases in 3+ parallel searches — even modest ones — tend to reduce QAR in autos.
Interpretation: This plot excludes the highest outlier in parallel search count, which significantly improves the clarity and reliability of the linear fit. The downward trend becomes more apparent and better centered, with a tighter confidence interval. This confirms that a small number of extreme days were disproportionately affecting the regression, and the negative relationship between parallel searches and QAR is more robust without them.
Interpretation: Similar to autos, cabs show a downward trend in QAR with increasing parallel search 3+ count. The trend line is influenced by a few extreme outliers at the higher end of search volume. While a negative pattern is visible, the wide confidence band and sparse data in high-count areas make the relationship less reliable in the raw scale.
Interpretation: Using log-transformed count, this graph reveals a stronger and clearer inverse relationship between parallel search 3+ and QAR for cabs. The data is more evenly distributed, and the confidence interval narrows, showing greater confidence in the model. This suggests that QAR consistently drops with increasing search load, validating the negative pressure of parallel searches.
Interpretation: This plot removes the top outlier in the log-transformed data and further tightens the model fit. The downward trend remains consistent and the confidence band shrinks slightly, improving interpretability. It reinforces the conclusion that QAR is sensitive to search load, and extreme days slightly distort but don’t reverse the trend.
| Mode | Correlation | Slope | Intercept | R_Squared | P_Value |
|---|---|---|---|---|---|
| Auto | -0.739126 | -0.088091 | 1.503214 | 0.546307 | 0 |
| Cabs | -0.595446 | -0.075932 | 1.340780 | 0.354556 | 0 |
Interpretation: - Correlation values indicate a strong negative association between log(Parallel Search Count) and QAR, stronger in autos (-0.74) than in cabs (-0.60). - Slope is negative in both cases, showing that as search count increases (log scale), QAR decreases. - R² values show that 55% of the variation in Auto QAR and 36% in Cab QAR is explained by parallel search load — indicating a substantial but not exclusive influence. - P-values are 0, indicating high statistical significance in both models.
Overall, the analysis strongly supports that increased parallel search load reduces quote acceptance, with autos showing a slightly more sensitive response than cabs.
In Namma Yatri, a parallel search count of 3+ indicates multiple batches were initiated to fulfill a ride request. However, drivers in later batches do not see the offer, resulting in wasted search attempts. Current batch sizes are: - 15 drivers per batch during peak hours - 10 drivers during non-peak hours
This leads to inefficiencies, as high search counts correlate with lower quote acceptance ratios (QAR) — especially under demand-supply mismatch conditions.
By integrating batch resizing, smart delays, visibility rules, and driver scoring, Namma Yatri can reduce excessive parallel searches. These steps are backed by both theoretical and field-tested strategies from leading ride-hailing research. This not only boosts QAR but also ensures fairness, responsiveness, and better platform efficiency.