Difference between revisions of "Forecasting"

Revision as of 14:33, 23 March 2014

Forecasting Defined

Forecasting: The functional process that takes historical data and information for use in predicting what might happen in the future.

Forecasting is one of four cornerstones of Workforce Management: Forecasting, Scheduling, Real-Time Activity (RTA) and Business Intelligence.

Requirements: What Do We Need For Effective Forecasting?

Conceptually, forecasting means the same thing to most people: the idea of telling the future based on some existing knowledge. How we work within the WFM model to generate accurate forecasts requires us to base forecasting activity on solid methodology. It's important that we follow common guidelines so that when we aggregate forecasting data to make business decisions, we can do so as accurately as possible. For this reason, we've developed a model with 4 dimensions that defines "World Class Forecasting".

Key concepts behind forecasting include generating and utilizing:

• A set of established roles and responsibilities
• A solid process with clearly defined purpose
• A collaboration mentality, with all other departments within the organization (use of wiki!)
• Clean data, both historical & judgmental derived from collaboration
• Adoption of Data Governance M&Ps to maintain clean data
• Support from call center, regional, division and corporate management
• Technical Tools (WFM Software) and Knowledgeable Resources

Key Forecasting Definitions

Forecasting requires common definitions behind the terms used in call center environments. The most important of these terms associated with this activity are:

• Talk Time: Defined as everything that happens during the length of the call from the moment the call is answered, to when the call is disconnected after "goodbye". All activity between "Hello" and "Goodbye" is tallied in measuring talk time.
• After Call Work (ACW): Refers to the time spent immediately following the "goodbye" of a call, where agents are completing work associated with the previous call.
• Average Handling Time (AHT): The sum of the Average Talk Time plus the Average After Call Work
• Offered Calls: All attempts made by callers to reach a call center. Offered calls should take into consideration callers who receive busy signals, callers who disconnect before reaching an agent, and callers who speak with an agent. Forecasts should reflect offered calls discounted by 1) multiple attempts from individual callers, and 2) callers who were self-serviced in the IVR.
• Answered Calls: Calls handled by a live agent after reaching an agent group.
• Call Load: Volume x AHT for a specific time-frame. Also referred to as workload, this is the key component used to forecast resources required to handle the calls for any given time segment and call queue.

Forecasting Roles And Responsibilities

A Forecaster's core responsibility is to conduct functional and analytical processes which take historical data and information for use in predicting what might happen in the future. This role is key to Customer Care's ability to correctly schedule and staff to meet customer's expectations of their call being answered in a timely fashion. Across the Division, this translates into maintaining for all intervals that 80% of calls are answered in 30 seconds or less for all call types except High-speed internet repair calls, whose target is 70% of all calls that are answered in 30 seconds or less.

Additional responsibilities include:

• Forecast call volumes, AHT, shrinkage, OT requirements, Outsourcer requirements within the Customer Care Organization using Verint WFM forecasting modules, MS Excel, and other forecasting tools.

• Apply both quantitative and judgmental forecasting methodologies. Utilize times series analysis data provided by WFM and blend with both event-driven and driver-based approaches to adjust forecast.

• Support scheduling, real-time, and business intelligence functions through tight communication with outputs from forecasting exercises.

• Work closely with the Workforce Manager, Human Resources and Training to perform long and short-term capacity planning to determine staffing requirements for meeting department goals and objectives

• Develop and maintain resource utilization models and attrition analysis by quarter using historical data within CMS, WFM and other resources to support staffing recommendations.

• Facilitate regularly monthly meetings with Billing, Marketing/Programming, LMC, Training/HR to ensure all event-driven call drivers are considered in forecasting

• Leverage data streams to examine driver-based pattern irregularities in bill drops, service interruptions (collections), and techops W/O volume to ensure all event-driven call drivers are considered in forecasting

• Communicate with Real-Time and Business intelligence to capture event-based and driver-based intervals for future call volume analysis.

• Work with WFM team members, call center management and other department stakeholders to identify, implement and monitor continuous improvement opportunities in an effort to gain efficiencies and better support company objectives.

Forecast Approach

Forecasting at the call center level is a significant and critically important portion of the World-Class Forecasting model. To maximize our effectiveness of staffing, each call center must be armed with the tools and knowledge to generate tactical forecasts. In setting the framework for generating accurate call center forecasts, we take three distinct forecasting approaches:

Historical Approach

Historical forecasting refers to the prediction of future volumes, AHTs and arrival patterns based solely on past historical data. This function will be generated through the WFM software application, and is not dependent on any outside inputs. Historical forecasts account for seasonal, day of week and interval variances. To generate forward looking forecasts, start working with the output of WFM software as the basis, and then adjust it with driver and event activities (below).

Driver Based Approach

Driver based forecasts account for call volumes tied to specific recurring organizational activity. Driver based data from activities that are recurring within the organization may be already accounted for in the forecast generated from historical data. An example would be customer calls generated in conjunction with a billing cycle. Discounting customer churn, a customer obtaining a bill has a predictable correlation to call volume that would already be captured in historical based forecasts. The key to integrating driver based data into the forecasting process is to identify driver-based data that falls outside historical patterns.

As part of our forecasting model, we look to identify and publish drivers that have a strong correlation to call center contacts, so that patterns identified as outside past events can be accounted for in our adjusted forecast. ^[1] These will initially include: Bill drops, collection-based service interruptions, and daily W/O volumes for repair and new installs.

Event Driven Approach

Event driven forecasts account for call volumes tied to events that need to be addressed as individual occurrences. "Events" may or may not be recurring. An example would include a Pay-Per-View event(PPV); while PPVs are recurring, each event has a unique independent time and volume impact to the organization. Event Driven call loads may be predicted from past events, but need to be extracted from historical forecasts and analyzed independent of historical data.

"Events" have two primary attributes to consider in our forecasting model:

• Predictable Events:
  • Planned Maintenance Outages
  • Channel Changes
  • Marketing Events
  • Programming Events (PPV)
  • Severe Weather w/ Lead Time Predictability (i.e., Hurricanes)
• Non-Predictable Events:
  • Fiber Cut
  • Severe Weather w/ Little or No Lead Time (i.e., Tornadoes)

Call Center Forecast: 3-Step Approach

Blending the three approaches of generating a forecast into a logical three-step process addresses performance measurement, a key attribute of our "World Class Forecasting" model. In measuring our forecast accuracy, it is not as important to measure the specific accuracy of our forecast in the past, as it is to measure how the inaccuracy is addressed in an ever-improving process.

The three step process is diagrammed to the right. The Forecast Build is broken into the three areas described below, and detailed in this section.

3-Step Approach

Baseline Forecast

• Acquire time-series day of week historical data
• Adjust for any out-of-normal patterns

Driver Based Adjustments

• Examine Driver Based data for any out-of-normal patterns
• Adjust time-series acquired in Step 1
  • Collections Cycles
  • Bill Drop Cycles
  • W/O volumes
• Contact Rate Adjustments
• Adherence and shrinkage adjustments
• Outsource Volumes
• Other

Collaboration Adjusted

• Examine upcoming calendar / event driven data
  • Marketing
  • Engineering
  • TechOps
  • Programming
  • HR/Training
  • Billing & Collections
  • National Care
• Adjust (time series + driver-adjusted) to account for events

As a best practice, call centers should maintain a Forecasting Collision Calendar

Forecast Review

Upon completion of the three steps, a process should be conducted for forecast review. The review process should be conducted as a weekly meeting at a regional level, which involves internal call center management, as well as other departments as necessary. The collaboration portion of the forecast build may facilitate some of the review necessary. These processes also integrate the scheduling activity, and the determination of call handling surplus or deficit. Upon completion of the Call Center Forecast Review, the forecast should be rolled to the Regional resource for surplus/deficit aggregation.

Forecasting Accuracy

As described in #Forecasting Roles And Responsibilities above, the forecasting role is critical to ensure that the schedules we generate will best meet the arrival call patterns; and thus meet our customer's expectation of their call being answered in a timely fashion. A common goal for all call centers is to minimize variance in forecasting and scheduling so that customers receive consistent levels of service (80% of calls handled in 30 seconds or less). Forecasting Accuracy methodology should be adopted as part of a comprehensive resource planning strategy.

Notes

1. ↑ Project: Integrated in with a CMS portal, we look to capture pulse-based data feeds from billing, collections and repair w/o's. This data stream would be integrated with the Event-Driven portal, so that the forecasting exercise has one core resource outside the WFM application to study "Drivers" and "Events", and their impact to the historical forecast.