Power BI case study · Service operations

Contact centre performance in Power BI

I analysed 197,278 contact-centre interactions to investigate access to human support, abandoned transfers and customer feedback. The result was a four-page Power BI report with defined metrics, quality checks and operational priorities.

  • Power BI
  • Power Query
  • DAX
  • Data modelling
  • Operational analytics
Inbound decision mapTotals reconciled

Inbound call outcomes

Separating AI endpoints from calls routed to people establishes the right base for each rate.

Recorded inbound routing outcomes134,463 inbound calls: 34,492 end at an AI endpoint, 99,967 have a known human route and four remain unclassified. Of the known human routes, 31,602 abandon before human attention, 66,075 finish at an agent endpoint and 2,290 abandon after human attention. Inbound calls134,463 AI endpoint34,492 Routed to people99,967 Initial abandonment31,602 Human attention68,365 4 unclassified calls are outside the known-route denominator.
Inbound calls134,463
AI endpoint34,492
Routed to people99,967
Before human contact31,602Initial abandonments
Human attention68,365Including abandoned transfers

4 unclassified calls sit outside the human-route base.

Human attention includes calls abandoned after contact. An AI endpoint does not confirm resolution.

Inbound outcomes are separated before rates are calculated. The complete breakdown is provided in the outcome table.
Interactions analysed197,278

Inbound, outbound and chat records preserved.

Initial abandonment31.61%

Of known inbound human routes, before human attention.

Net Promoter Score · NPS34.28 points

7,708 responses; zeros included provisionally.

Satisfaction · scale 1–109.46/10

5,127 valid scores; 7.50% coverage.

The question

Where should the operation investigate first?

This practical analytics case examines inbound calls, outbound attempts and chat activity in a telecommunications contact centre.

My contribution

I prepared and classified the source records, built the data model and measures, designed four report pages, reconciled the results and translated the findings into an investigation plan.

Delivered: a Power BI report and an analytical decision memo.

The Power BI report

A view from overview to detail

The report separates executive monitoring, operational performance, customer experience and data quality. These aggregate previews retain the original Spanish labels.

Anonymised Power BI executive overview with channel totals, inbound outcomes and volume across six periods
Channel totals, call outcomes and grouped contact reasons establish the operational baseline. Period 6 is partial.Open larger view ↗
Anonymised Power BI operations view showing abandonment by initial team, hourly demand and waiting-time metrics
Initial team and hourly demand guide further investigation. Waiting-time units are assumed to be seconds, pending source confirmation.Open larger view ↗
Anonymised Power BI quality view documenting review cases, survey recording rules, denominators and NPS sensitivity
Quality checks, response bases and alternative NPS definitions stay visible alongside the results.Open larger view ↗
Access to support

95% of initial abandonments

Customer support accounts for 30,017 of 31,602. Investigate its queues and demand first.

Sales handoffs

12.2% abandoned after contact

1,081 of 8,864 calls with human attention in the initial Sales group. Review the transfer journey.

Feedback definitions

34.28 or 50.02 NPS points

The same data give different scores when zeros are excluded. Confirm recording rules before setting targets.

Operational evidence

Operational performance

Compare recorded volume, initial abandonment and the 90th percentile of waiting time on known inbound human routes. Period labels are generic in this public version; the final period is partial.

Compare periods

Choose a metric and period to inspect its values. Period 6 is partial. Waiting-time units are provisional; definitions and response bases are documented below.

View all performance values as a table
Aggregate period metrics. Initial abandonment uses known inbound human routes; NPS uses valid inbound responses.
PeriodAll interactionsInbound callsInitial abandonmentWait P90NPS
Period 126,15317,96537.94%146.1 s42.75
Period 230,86520,23435.16%152.0 s42.62
Period 344,43632,82746.09%164.3 s16.91
Period 441,99029,42624.92%164.8 s26.18
Period 530,25019,06611.58%140.6 s42.79
Period 6 · Partial period23,58414,94517.83%156.2 s42.93
All periods197,278134,46331.61%157.9 s34.28
Inspect the inbound outcome reconciliation
Terminal classification from the source, expressed with generic public labels. Shares use all 134,463 inbound calls.
Recorded outcomeCallsShare of inbound
Ends with AI34,49225.65%
Agent terminal outcome66,07549.14%
Initial abandonment31,60223.50%
Abandonment after human contact2,2901.70%
Unclassified40.00%

Customer feedback

Customer feedback and NPS

Net Promoter Score (NPS) measures willingness to recommend the company. I compared two zero-score policies because the brief describes a 1–10 scale but the records also contain zeros.

View both NPS calculations and their response bases
Same captured survey data under two zero-response policies. The all-period NPS is recalculated from pooled responses, not averaged from period scores.
PeriodNPS with zerosResponse baseNPS without zerosResponse baseZero scores
All periods34.287,70850.026,899809
Period 142.7597358.5687697
Period 242.621,21357.991,095118
Period 316.911,44339.541,209234
Period 426.181,69641.821,509187
Period 542.791,42155.001,309112
Period 6 · Partial period42.9396252.6190161

The brief specifies 1–10; the records contain zeros. Their meaning remains unconfirmed, so the 0–10 calculation and its sensitivity scenario are both retained.

Method and validation

How I built the model

I preserved one row per interaction, separated channels and outcomes, and checked that each measure retained its definition as filters changed.

  1. 01Prepare

    Clean and classify

    Type dates and durations, normalise keys, join the destination dictionaries and retain records requiring review.

  2. 02Model

    Separate the relationships

    Connect calendar and destination dimensions to the interaction records. Keep initial team separate from final endpoint.

  3. 03Calculate

    Define each measure

    Use DAX for route-based rates, waiting percentiles and feedback measures, with numerators and denominators in the same filter context.

  4. 04Validate

    Reconcile and inspect

    Compare independent totals, inspect filter behaviour and verify the four report pages and exported results.

One fact table, two dimensions

CalendarOne row per date
1 → *
InteractionsOne row per interaction
* ← 1
DestinationsUnique endpoint mapping
Single-direction filters run from dimensions to the fact table.

Checks that informed the analysis

242 destination/time combinations, 4 unmapped inbound destinations and 1 extreme-duration record were retained for review. A flagged record is an investigation case, not automatically an error.

The saved report was validated in Power BI Desktop. The public evidence preserves aggregate results and documented definitions.

Metric logicInspect the initial-abandonment definition
Initial abandonment rate =
    Inbound calls abandoned before human attention
    ÷ Known inbound calls routed to humans

All-period calculation:
    31,602 ÷ 99,967 = 31.61%

Known human routes =
    66,075 agent terminal outcomes
    + 31,602 initial abandonments
    + 2,290 post-contact abandonments

AI terminal outcomes and 4 unclassified inbound calls
are outside this denominator.
Definitions and interpretation
Calls routed to people
Known inbound agent or queue destinations: 99,967 calls. Initial abandonment is 31,602 / 99,967. Human attention includes 2,290 calls abandoned after contact; these outcomes overlap with total abandonment.
Waiting time
Routing time for those known human routes, including zero waits and excluding AI endpoints. P90 is the value at or below which 90% of observations fall. Seconds are assumed, pending source confirmation; no SLA target is supplied.
Survey scales and coverage
The brief states 1–10, but zeros are present. NPS provisionally uses integer scores 0–10; the alternative excludes zeros. Satisfaction follows 1–10 among calls with human attention. NPS coverage is 5.73% of inbound calls; satisfaction coverage is 7.50% of human attention. Survey invitations are unknown, so these are coverage measures, not invitation response rates.
Period and outcome limits
The six public periods preserve chronological order; the final period is partial. AI endpoints do not establish resolution. Outbound records are attempts, without confirmed connection or commercial success. Initial team does not identify the final agent. These data alone do not establish causes, staffing requirements or first-contact resolution.
Public evidence
The page, report previews and downloads use aggregate data and generic labels. Individual records, agent identifiers, original dates, comments and source files are excluded.

Decision logic

Operational priorities

The findings support a targeted investigation plan. They identify where to look; capacity, routing and process evidence are still needed to explain the causes.

01
Concentration

Investigate support access

Action

Support contributes about 95% of initial abandonments. Start with its arrival pattern, queue coverage and waiting-time distribution; assess whether staffing, routing or callback changes warrant a controlled test.

02
Handoff signal

Inspect Sales transfers

Action

Sales records a 12.2% post-contact abandonment rate among calls that had human attention. Inspect transfer destinations, ownership and handoff steps before attributing the pattern to agent performance.

03
Measurement

Agree the feedback rules

Action

Confirm whether zero scores are submitted answers or defaults, retain the sensitivity comparison and track survey coverage alongside every score. Use comparable response bases before evaluating changes.

Compare broad teams with the denominator visible
Initial routing teams, rather than individual staff. Initial abandonment uses known human routes; post-contact abandonment uses calls with human attention, within each initial routing team.
TeamKnown human routesInitial abandonmentsInitial abandonment ratePost-contact abandonment rate
Customer support86,37430,01734.75%2.15%
Sales9,6287647.94%12.20%
Network operations3,74269218.49%0.00%
Other22312957.85%0.00%
Monitoring sequenceSignal → investigation → measured change
BaselineDefinitions, volume and quality
InvestigationSupport access and Sales transfers
EvaluationControlled changes and comparable periods

No intervention was implemented in this case study. The recommendations are proposed next steps, with impact to be measured after implementation.

Continue exploring

More data work

Explore the wider analytics portfolio or review the experience, tools and credentials behind this project.

Contact

Need a clearer answer from complex data?

I connect data preparation, analytical models and clear reporting to help people understand the evidence and decide what to investigate next.