WFM guideData & analytics

Contact centre data quality

Q: What are the most common data quality problems in contact centre WFM?

The most common data quality problems affecting contact centre WFM forecasts are: (1) Abandoned calls included in handled contact counts — artificially inflates volume data; the forecast overestimates demand; (2) AHT spikes from outlier contacts — a single 90-minute call inflates the average AHT calculation for its interval; these should be excluded from AHT modelling; (3) Wrap code abuse — agents using a default or incorrect wrap code distorts the contact type distribution, making the forecast inaccurate by contact type; (4) System transfers counted as separate contacts — when a contact is transferred between queues or systems, each leg may be counted as a separate contact, inflating volume data; (5) Non-contact work coded as contacts — agents working on emails or admin tasks while logged into the voice queue can introduce phantom contact volume if the ACD counts their availability as contact time; (6) Historical data not adjusted for one-off events — a spike in historical contact volume from a past incident, system outage, or campaign should be excluded from the baseline training data used for forecasting.

A WFM forecast built on corrupted data produces confidently wrong staffing levels. Sophisticated forecasting algorithms applied to systematically incorrect contact data generate numbers that look precise but are wrong in a direction no one can easily explain. Data quality is the unsexy constraint on forecast accuracy.

How data errors propagate through the forecast

Data error propagation — example: abandoned calls in volume

Actual situation

True handled contacts (conversations): 800/hr
Abandoned contacts: 120/hr
True average AHT: 5.2 min
Required agents at 80% SL: 92

What the WFM model sees (abandoned calls included in volume)

Volume input to model: 920/hr (800 + 120 abandonments)
AHT input to model: 4.5 min (diluted by zero-AHT abandonments)
Model's output: required agents: 98 — 6 agents too many
Annual cost of overstaffing at £25k/agent: ~£150,000

This error compounds over the planning horizon — every capacity plan, recruitment plan, and budget built on this model is systematically biased. The contact centre appears to be continuously understaffed when it is not.

Six most common data quality errors in contact centre WFM

Abandoned calls counted as handled contacts

Impact on forecast

Volume data inflated — typically by 5–15% in a high-abandon-rate operation. Forecast overestimates demand. Centre is overstaffed relative to actual answerable demand.

Detection

Compare offer rate to handle rate in the ACD report. In a healthy operation, the gap is abandons. If offer rate = handle rate despite visible abandonment, check how abandoned calls are classified in the ACD.

Fix

Confirm ACD is reporting abandoned calls separately from handled contacts. Exclude abandons from the volume series used for forecasting. If the ACD conflates the two, correct at the reporting layer before the data enters the WFM model.

AHT inflated by outlier contacts

Impact on forecast

A small number of very long contacts (legal escalations, interpreter-mediated calls, system outages causing long hold periods) inflate average AHT. The forecast overpredicts handle time and overstaffs.

Detection

Review the AHT distribution, not just the mean. If the 95th percentile AHT is more than 3× the median AHT, the distribution is heavily right-skewed. Outliers are candidates for exclusion from the AHT modelling.

Fix

Set an AHT cap for forecasting purposes — typically the 95th or 97.5th percentile. Contacts above the cap are excluded from the AHT calculation used for staffing. This does not mean excluding them from operations — only from the AHT model input.

Wrap code misuse — agents using default or incorrect codes

Impact on forecast

Contact type distribution is distorted. The WFM team cannot produce accurate AHT-by-contact-type forecasts if wrap code data does not reflect actual contact types. Skills-based routing staffing calculations are wrong.

Detection

Check the wrap code distribution: if more than 10–15% of contacts are coded to a default, generic, or catch-all wrap code, data integrity is low. Run the distribution by agent — if some agents use the default code at 80%+ and others at 5%, the problem is compliance, not definition.

Fix

Short-term: remap the default code to the closest contact type based on call recording analysis. Long-term: enforce wrap code compliance through QA monitoring and coaching; tighten the wrap code list to remove ambiguous options; automate coding where the ACD or speech analytics can classify the contact type.

Transfers counted as separate contacts

Impact on forecast

When a contact is transferred between queues or between agents, each leg may be counted as a separate offered contact. Volume data is inflated; AHT for the receiving queue is underestimated (receiving agent only handles the transferred portion).

Detection

Compare transfer rate from ACD to the volume delta between originating and receiving queues. If the receiving queue has more contacts than incoming transfers plus direct contacts would suggest, transfers may be double-counted.

Fix

Configure the ACD to count each unique customer interaction as one contact regardless of transfers. If this is not possible in the current system, apply a correction factor to the receiving queue volume based on the known transfer rate.

Unanswered outbound contacts counted in volume

Impact on forecast

For blended or outbound operations, unanswered outbound calls (rings with no answer) may be counted in the contact volume metrics. These have zero AHT but are counted as contacts — deflating the average AHT used for staffing calculations.

Detection

In blended operations, compare the contact volume to the number of contacts that resulted in a conversation. If the ratio is significantly below 1.0, unanswered outbound calls may be in the count.

Fix

Define contact volume as conversations only, not dial attempts. Exclude ring-no-answer and answering machine contacts from the AHT calculation and, where possible, from the volume baseline.

Historical data not adjusted for anomalous periods

Impact on forecast

If the historical contact data used for forecasting includes periods affected by system outages, regulatory announcements, campaigns, or one-off incidents, the seasonal baseline is corrupted. The forecast reproduces the anomaly every year.

Detection

Review the contact volume history for spikes that do not have a clear seasonal explanation. Check event logs for the periods in question. A spike with no identifiable repeating cause should be excluded from the seasonal index calculation.

Fix

Replace anomalous periods in the historical data with an interpolated baseline value derived from adjacent periods. Document the replacement in the data quality log so the adjustment is visible and reviewable.

Data validation checks for the WFM planning process

Data validation should happen before data enters the WFM model — not after the forecast has been produced and the error is visible in the WAPE score. The following checks should be run on the raw contact data before each planning cycle.

Offer vs. handle vs. abandon reconciliation

Offered contacts must equal handled + abandoned + any other disposition. Any gap suggests a classification error.

AHT percentile review

If 95th percentile AHT exceeds 3× the median, review the top 5% of contacts for anomalies before including in the AHT model.

Wrap code distribution audit

If more than 10% of contacts are coded to a default or generic wrap code, flag for compliance action before using the data for contact-type-level forecasting.

Interval completeness check

Check for missing or zero-volume intervals that should have had contact activity. Missing data should be interpolated, not left as zeros that skew the intraday distribution.

Volume anomaly detection

Flag intervals where contact volume is more than 3 standard deviations from the seasonal average. Investigate before including in the baseline.

Year-over-year pattern validation

Compare the seasonal pattern from the current year to the prior year. Unexplained divergence in the seasonal index warrants investigation before the seasonal adjustment is applied.

Data quality questions

What are the most common data quality problems in contact centre WFM?

The six most common: (1) Abandoned calls counted in handled contact volume — inflates demand data by 5–15%; (2) AHT inflated by outlier contacts — a small number of very long calls skews the average used for staffing; (3) Wrap code misuse — agents using default codes distort the contact type distribution; (4) Transfers counted as separate contacts — inflates volume and underestimates receiving-queue AHT; (5) Unanswered outbound calls diluting AHT — zero-AHT ring-no-answer records reduce the average AHT used for staffing; (6) Historical data not adjusted for one-off anomalies — past incidents, outages, or campaigns corrupt the seasonal baseline if not excluded from the training data.