ML based NCNS Score

Introduction

HyperTrack has tracked over 100 million shifts, giving us an unparalleled dataset to understand workforce behavior. With this scale, we’re able to uncover predictive features that determine whether a worker will arrive on time. This going beyond the naive ETA approaches that only extrapolate from the last known location.

The features we use include:

• Dynamic travel patterns: remaining travel time and distance, plus how these evolve in real time
• Tracking quality: reliability and consistency of location signals
• Worker engagement: app permissions, session activity, and overall usage behavior
• Historical performance: patterns in past arrivals and attendance

For every shift that uses HyperTrack pre_shift tracking, our platform automatically generates a delayed_from_scheduled_at risk score. This risk score is powered by our machine learning model, which predicts the likelihood of on-time arrival.

Labeling ground truth data

The baseline on-time arrival model works out of the box. When you provide additional labels with your ground truth data, we use them to train a custom NCNS (No-Call, No-Show) model that adapts to the specific meaning of work in your context.

To label orders, update the metadata field using the PATCH /orders/{order_handle} endpoint (see API reference).

{
  "metadata": {
    "ht_on_time_arrival": true | false,   // Worker arrived within your definition of "on-time"
    "ht_worked_shift": true | false       // Worker showed up and completed the shift (show/no-show)
  }
}

Canceled Orders

When an order is canceled, add additional metadata to record no-shows:

{
  "metadata": {
    "ht_no_show_cancel": true | false
  }
}

Control Group / Holdout Set [OPTIONAL]

You can also mark orders for experimentation (e.g., control/holdout group):

{
    "metadata": {
         "ht_no_show_holdout": true|false,
    }
}