Service Experience Metric Suite

Overview

The Service Data and Strategy & Analytics teams are partnering to deliver the Service Experience Metric Suite (SEMS) which provides high-value insights to better understand and quantify EG Service experiences. Because these metrics are very specific, this suite will allow us to accurately and comprehensively capture the traveler and partner experience.

On this page, you'll find a list of metrics currently available or in planning/progress to improve service outcomes, inform product management direction, and so on. Below, you'll also find links to pages with in-depth explanations of these models.

Please note: If you have worked with previous models (e.g., CES v3), all have been renamed for clarity and this page is the authoritative source going forward.

Lastly, if you believe you have a use-case to leverage some of these metrics, want to propose metrics, or just have questions, please reach out to us in the dedicated Slack channel: #service-experience-metric-suite

Business Justification

Our industry uses many metrics (e.g., Net Promoter Score, Customer Effort Score) to quantify experiences. Expedia adopts and aligns with industry best practice metrics, and drives decisions and investments based upon them. These metrics, however, are sometimes abstract and subjective, hence their usefulness is limited.

For us to maintain our position as an industry leader, we need to expand our approach. We should define new metrics tied to specific outcomes driven by AI or statistical models powered by our rich datasets; the Service Experience Metric Suite will encompass all of these metrics. Our latest addition is the Repeat-Purchase Likelihood (RPL) model, which is an evolution addressing some limitations regarding the Detractor Likelihood (DL) and Repeat Contact Likelihood (RCL) models. Though industry uses metrics like Customer Effort Score (CES) to quantify effort which impacts outcomes, the metrics we are developing directly quantify outcomes which are far more powerful.

The SEMS will grow as we generate new ideas, acquire new datasets, and provide new experiences. What we build and provide will also be informed by you, so please reach out to us to discuss insights you want to unlock.

Metric Suite Details

Metric Data Grain Details Current Use Cases Status
Agent Quality Index (AQI) Agent A statistical model scoring an agent's ability to provide an excellent customer experience. This metric feeds into the Agent Quality Scorecard (AQS).

Input variables
  • Agent Follow Up %
  • Long Contact %
  • Short Contact %
  • Agent Disconnect %
Output Values
  • AQI = int[0 to 100]
  • Agent Effectiveness Coaching
Live in Production
Agent Quality Model Score (AQM) Agent A machine learning model that predicts agent quality as measured by repeat booking rate. This metric will feed into the Agent Quality Scorecard (AQS).

Input variables
  • Average Active Time
  • Average Assigned Agent Time
  • Average Talk Time%
  • Average Mute Time%
  • Average Hold Time%
  • Average Wrap-Up Time%
  • Average Response Time%
  • Average Follow Up Time%
  • Average Handle Time%
  • Percent Short Call%
  • Percent Long Call%
  • Average Received Message Count%
  • Average Sent Message Count%
  • Average Transfer Count%
  • Percent Initiated Transfer%
  • Percent Latest Handled Session%
  • Percent Agent Assisted Follow Up%
  • Percent Agent Disconnect

Output values
  • AQM = int[0 to 1]

  • Agent Effectiveness Coaching
Live in Production
Detractor Likelihood (DL) currently named CES, but to be updated in Q4 Conversation
Itinerary
Customer
An AI-based model trained to predict the likelihood of a customer being a detractor regarding Net Promoter Score (NPS).

Input variables
  • Repeat Contact Likelihood
  • Booking features
  • Loyalty features

Output values
  • DL = int[0 to 100]

Model Performance Metrics
  • Overall Performance (ROC-AUC): 0.76
  • Precision: 0.36
  • Recall: 0.55
  • TPSP Monthly Business Review
  • World Class metrics
  • Win-Back Initiative
  • Executive Escalation Capture
Live in Production
Predict Product (UPP) Conversation A machine learning model that predicts a traveler product based on transcript text, when latestvaprod= UNKNOWN.

Output values
  • 0 - Car
  • 1 - Destination Services
  • 2 - Flight
  • 3 - Insurance
  • 4 - Lodging
  • 5 - Other

  • Agent Effectiveness Coaching
Live in Production
Repeat Contact Likelihood (RCL) Conversation An AI-based model that measures the platform efficiency by predicting the likelihood of an agent-assisted follow-up from a customer after 72-hrs when conversation is completed.

Input variables
  • Customer effort drivers such as:
    • Queue Time
    • Agent Time
    • VA Time
    • Message Count
    • Conversation Duration

Output values
  • RCL = int[0 to 100]

Model Performance Metrics
  • Overall Performance (ROC-AUC): 0.73
  • Precision: 0.21
  • Recall: 0.81
  • Pending integration into Early Warning System
Live in Production
Repeat-Purchase Likelihood (RPL) Customer + Itinerary An AI-based model trained to predict the likelihood of a repeat-purchase, meaning a customer makes a repeat purchase within a 90-day window from an initial purchase.

Input variables
  • Customer effort drivers in a journey (across conversations) such as:
    • Queue Time
    • Agent Time
    • VA Time
    • Message Count
    • Conversation Duration

Output values
  • RPL = int[0 to 100]

Model Performance Metrics
  • Overall Performance (ROC-AUC): 0.78
  • Accuracy: 0.75
  • Precision: 0.81
  • Recall: 0.83
  • Customer Lifetime Value improvement
Live in Production
Customer Effort Score (CES) Conversation
Journey
Industry-standard metric that is a weighted average of all effort drivers to quantify the customer's effort during a service interaction.

Input variables
  • Customer effort drivers such as:
    • Queue Time
    • Agent Time
    • VA Time
    • Message Count
    • Conversation Duration

Output values
  • CES = int[0 to 100]
Requested but not yet available
Schedule Quality Index (SQI) Agent Forecast A metric that measures how well staffing levels were met when compared to established requirements for a given time frame. .

Input variables
  • Actual Billable Hours for Interval
  • Actual Required Hours for Interval

Output values
  • SQI= int[0 to 100]

  • Workforce Management Forecasting
In development
World Class Experience Compliance (WCEC) Expedia Group is focused on delivering a World Class Experience, and doing so requires us to adhere to SLAs across many dimensions. This model will quantify strict and comprehensive SLA compliance using variables representing our commitments to Travelers and Partners.

Input variables
  • SLA metrics such as:
    • Service Level
    • Resolution Time
Output values
  • WCEC = int[0 or 1]
Requested but not yet available