All showcases
Logistics · 99 Deliveries · May 2025

Lensiq predicted 99 last-mile courier deliveries in Lebanonwith 92.93% accuracy

Trained on 1,000 real delivery records. Tested on 99 it had never seen. The model got 92 right, and flagged its own uncertainty on almost every mistake.

92.93%
Accuracy
92 of 99 correct
94.2%
Avg Confidence
across all predictions
84
High-Confidence Calls
of 99 predictions
7
Mistakes
avg confidence 68.7%

Model insights

What drives delivery success

The model learned these five factors matter most, ranked by their influence on the outcome.

Attempt Number25.7%

First attempts succeed far more than retries. Third attempts fail at high rates.

Pre-Confirmation20.1%

Confirmed deliveries arrive at significantly higher rates than unconfirmed ones.

Driver Experience11.9%

More experienced drivers navigate challenges, traffic, and edge cases better.

Distance10.5%

Longer routes increase exposure to delays, traffic, and missed windows.

Weather9.4%

Storms and heavy rain meaningfully reduce delivery success rates.

Confidence

The model knows when it's sure

Confidence distribution: 99 predictions

84 High
84High confidence (85%+)
10Medium (60–84%)
5Low (below 60%)

When the model was wrong, its average confidence was 68.7% vs 96.1% on correct predictions. It knew when it was uncertain.

Set a confidence threshold of 85% and you would catch almost every mistake before dispatch.

Where it went wrong

All 7 mistakes and what they reveal

5 of 7 were Low or Medium confidence. The model was uncertain on nearly every mistake. That's the signal you act on.

#AreaPackageWeatherAttemptPredictedActualConfidenceLevel
14BeirutMedium ParcelClear1Not DeliveredDelivered60.1%Medium
30North LebanonDocumentRain1Not DeliveredDelivered90.2%High
41Mount LebanonLarge ParcelClear3Not DeliveredDelivered59.2%Low
44BeirutFragileRain2DeliveredNot Delivered75.3%Medium
55ZarqaSmall ParcelClear3Not DeliveredDelivered54%Low
69IrbidSmall ParcelStorm1Not DeliveredDelivered53.1%Low
83BeirutLarge ParcelClear1DeliveredNot Delivered89%High
5 of 7
mistakes were Low or Medium confidence

A simple confidence threshold would have flagged these for manual review before dispatch.

68.7%
average confidence on wrong predictions

Compared to 96.1% on the 92 correct ones. The gap makes the signal clear.

Sample results

Row-by-row predictions

A sample of 8 rows from the test run, including both correct predictions and the two highest-profile mistakes.

#AreaPackageWeatherAttemptPredictedActualConfidenceLevelCorrect
0Mount LebanonSmall ParcelClear1DeliveredDelivered100%High
5BeirutMedium ParcelStorm2Not DeliveredNot Delivered100%High
7AmmanLarge ParcelRain1DeliveredDelivered100%High
12AmmanFragileRain1DeliveredDelivered78.6%Medium
14BeirutMedium ParcelClear1Not DeliveredDelivered60.1%Medium
26BeirutLarge ParcelClear3Not DeliveredNot Delivered100%High
30North LebanonDocumentRain1Not DeliveredDelivered90.2%High
36Mount LebanonSmall ParcelClear1DeliveredDelivered99.99%High

Bottom line

The model learned real operational logic

Not rules someone wrote. Patterns it discovered from 1,000 real deliveries.

Confirm before you dispatch

Pre-confirmation is the second strongest signal in the data. Unconfirmed deliveries fail at significantly higher rates. A quick confirmation call before dispatch is one of the highest-ROI actions in your operation.

Match driver experience to route difficulty

Sending a new driver on a long route in bad weather is a compounding risk. The model learned that experience, distance, and weather interact. Each one manageable alone, dangerous in combination.

Treat repeat attempts as risk signals

Attempt number is the single strongest predictor in the dataset. A third delivery attempt has very different odds than a first. The model quantifies that risk so you can act before the driver leaves the warehouse.

92.93%
accuracy on data it never saw

These are not rules someone programmed. The model found them by studying your operations. Every business has its own version of this logic. Lensiq extracts it from your data and turns it into predictions you can act on in real time.

This is what your operations data can do.

Lensiq turns your historical delivery records into a live prediction engine. No ML team required.

Register Now