Aurora Banking Group · Banking

Workforce Productivity Intelligence: The Productivity of Cognition

Maturity
L4
Domain
Grow & Keep
Analytics
predictive
Income per FTE
≈$425k
Sponsor
Group Chief Human Resources Officer
Confidence
High

The situation

Where is cognitive-work productivity created and lost across Aurora, and which levers — AI augmentation, work redesign, spans, demand management — move it, measured honestly without crude output proxies?

Grow & KeepL4Sponsor · Daniel Osei

The recommendation on the table

Stand up honest productivity intelligence before scaling AI

Books AI value against real movement in income-per-FTE and value-adding task share, not against tool roll-out.

Decision ownerGroup Chief Human Resources Officer · Daniel Osei
MaturityL4
Priorityhigh

Trade-offRequires measuring cognitive-work productivity with stated limitations — harder than headcount ratios, but defensible.

The evidence

In a knowledge business where ~half the controllable cost is people, a few points of productivity is a cost-income story worth nine figures — yet Aurora had never had a defensible measure of cognitive-work productivity, the OEE-equivalent that manufacturing takes for granted. BNK-02 built honest productivity intelligence (income per FTE, value-adding task share, cost-income by function) and modelled the uplift available from AI augmentation, finding a large reworkable-task reserve in operations and an adoption gap that means uplift is being assumed before it is earned.

Aurora — Workforce Productivity Intelligence

Cognitive-work productivity made measurable: income per FTE, value-adding task share, and the augmentation uplift available — gated on adoption.

Income per FTE· operating income per FTE (group)
≈$425k+45vs target
On watch
Value-adding task share· value-adding share of task-time
≈55%
On watch
Modelled augmentation uplift· projected operations productivity uplift (±5pp)
≈18%
Neutral
Augmentation adoption· active augmented users vs. eligible licensed
≈45%+35vs target
Critical
Illustrative preview
Income per FTE by segment ($k)
02505007501,000CorporateWealthCommercialDigitalRetail ops

Key takeawayRetail operations lags peers while carrying the highest AI exposure — the first redesign target.

Value-adding vs. low-value task share — operations
100%TOTAL
  • Value-adding55%55%
  • Reworkable/low-value45%45%

Key takeaway~45% of operations task-time is the productivity reserve augmentation is meant to release.

Interactive view is best explored on desktop.

Key findings

About 45% of operations task-time goes to rework, coordination and low-value processing — the productivity reserve AI augmentation is meant to release. A modelled ~18% uplift (±5pp) is available, but only if work is redesigned around the freed capacity rather than the capacity simply being cut.

≈45%of operations task-time reworkable/low-value

What we can’t claim

Augmentation tool licences are well ahead of actual adoption — only ~45% of eligible users are active. The productivity case is being counted on roll-out, not on use, which is the banking echo of manufacturing's 'trained is not the same as capable.' Until adoption rises, the uplift is an assumption, not a result.

Recommendations

Stand up honest productivity intelligence before scaling AI

high priority

Books AI value against real movement in income-per-FTE and value-adding task share, not against tool roll-out.

Trade-off

Requires measuring cognitive-work productivity with stated limitations — harder than headcount ratios, but defensible.

Tie productivity-value booking to measured adoption, not licences

medium priority

Closes the gap between assumed and earned value by shifting focus from licence roll-out to adoption enablement.

Trade-off

Slows the booking of headline savings until adoption is real.

Analytical framework

How we reached this

Predictive — measure cognitive-work productivity honestly and model the uplift available from augmentation and work redesign, with confidence intervals.

ConfidenceMedium

Methods applied

Productivity analysisDriver analysisDemand-capacity analysisUplift modeling

Statistical techniques

Regression analysisVariance analysisCorrelationForecastingSegmentation

Algorithms

Linear/regression uplift modelsRandom forest (feature importance)

Data sources

Activity/time dataOutput & income dataRole/task taxonomyAI-tool telemetryFinance

Outputs generated

Productivity baselineUplift projection with bandsDriver rankingAdoption-vs-realization view

Why this confidence

Knowledge-work productivity is partly unmeasurable; outputs are ranges with a stated measurement caveat as the binding limit.

The reasoning

Business context

Builds directly on the BNK-01 baseline. Sponsored by the Group CHRO because the cost-income target cannot be met by technology alone — only by re-shaping what 96,500 people do, and measuring whether the re-shaping actually lifts value.

Expected value

Replaces crude headcount ratios with defensible productivity intelligence, sizes the augmentation uplift with confidence intervals, and ties value booking to measured adoption rather than tool roll-out. The largest closable part of the cost-income gap is productivity, not structural change.

Workforce landscape

Operations spends ~45% of task-time on reworkable/low-value work — the productivity reserve. Modelled augmentation uplift is ~18% (±5pp) in operations, contingent on adoption. Augmentation adoption (~45%) lags licences badly — the banking echo of 'trained ≠ capable.'

The analytics journey

Level 4, predictive. Models the productivity uplift from specific interventions with explicit confidence intervals, and is honest that knowledge-work productivity is partly unmeasurable — treating that as a stated limitation, not a footnote.

Under the hood

Activity and income data are joined to the task taxonomy to compute income per FTE and value-adding task share; an interpretable regression-based model projects uplift from augmentation/redesign with confidence bands; adoption telemetry gates whether modelled uplift is being captured.

Confidence & evidence

Why you can rely on this

82%
Analysis confidenceHigh

The inconvenient truth

Augmentation tool licences are well ahead of actual adoption — only ~45% of eligible users are active. The productivity case is being counted on roll-out, not on use, which is the banking echo of manufacturing's 'trained is not the same as capable.' Until adoption rises, the uplift is an assumption, not a result.

Method

Confidence is a deterministic read of KPI strength, target and benchmark coverage across this project — shown on an illustrative reference dataset, computed the same way it would be on live data.

Take this further

Where this project connects