STRATEGY / OPERATING LEVERAGE

Clients & products per employee

Stop counting automation activity. Measure whether the institution can absorb more customers, products, transactions and revenue without expanding the operating structure at the same rate.

Bots, copilots, APIs and AI agents are inputs. Productivity is an outcome. The operating question is whether output grows faster than the structure required to produce, serve and control it.

José ÑáñezTechnology Advisor · Board Member

Published April 7, 2026Updated August 27, 202612 min read

THESIS

Clients per employee is a useful stress test — but it is not a complete productivity metric. Operating leverage appears only when scale, economic output, service capacity and quality move together.

THE MEASUREMENT ERROR

Automation activity is not operating leverage

A company can deploy dozens of automation initiatives and still carry the same operating burden. Time saved in one task may disappear into rework, exception handling, additional controls or simply more idle capacity.

The opposite can also occur: clients per employee may remain flat while monetization improves, cost to serve falls or a function absorbs materially more workload. That is why one ratio should never become a leaderboard.

The relevant test is not “how much technology did we deploy?” It is “how much more economic activity can the institution absorb with the same — or proportionally smaller — operating structure?”

MEASUREMENT STACK

Use a stack of evidence, not a single score

The metrics should move from raw scale to economic productivity and then to service quality. Compare the same institution over time before comparing different business models.

Scale density

Clients per employee = active clients / FTE

Tests whether the operating structure is absorbing more customer scale. Use active customers when possible and keep the employee definition consistent.

Workload density

Work units per FTE = transactions, cases or active products / operating FTE

A stronger functional measure when the institution can define the work unit consistently. “Products per employee” is useful only when active-product definitions are comparable.

Economic productivity

Revenue productivity = annual revenue / average FTE

Shows whether monetization grows faster than the operating structure. Profit or contribution per FTE can be even stronger when accounting definitions are comparable.

Cost outcome

Cost to serve = attributable operating cost / active client

Prevents a scale metric from hiding higher servicing, infrastructure, risk or control cost.

Quality guardrail

Quality = service + risk + control outcomes

NPS, complaints, losses, error rates, rework and control incidents determine whether higher capacity is actually sustainable.

CORE RELATIONSHIP

The most useful comparison is growth versus structure

Operating leverage gap = growth in economic output − growth in operating structure

A positive gap is evidence that output is scaling faster than the structure. It is not proof of causality: pricing, mix, market growth, acquisitions, risk appetite and accounting changes can also move the result.

PUBLIC EVIDENCE · FY2025

Two digital platforms show why one metric is not enough

The examples below use public company disclosures. They are not a ranking: Nu and Revolut operate different products, geographies and accounting frameworks. The objective is to show how the measurement stack changes the interpretation.

Scale stable · economics improved

Nu Holdings

Metric	2024	2025	Change
Customers	114.2M	131.0M	+15%
Year-end employees	8,716	10,027	+15%
Clients / employee	~13.1K	~13.1K	~0%
Total revenue	$11.5B	$15.8B	+37%
Monthly ARPAC	$11.3	$13.3	+17.7%
Monthly cost to serve / active customer	$0.9	$0.8	Improved

Customer scale per employee was essentially flat because customer growth and headcount grew at almost the same rate. Yet revenue, ARPAC and cost to serve improved. The conclusion is not “no productivity.” It is that the productivity signal appears more strongly in monetization and unit economics than in customer density.

Enterprise scale improved · support leverage became visible

Revolut

Metric	2024	2025	Change
Retail customers	52.5M	68.3M	+30%
Year-end employees	10,133	12,200	+20%
Clients / year-end employee	~5.18K	~5.60K	+8%
Revenue	£3.09B	£4.52B	+46%
Average customer operations FTE	4,974	4,700	−5.5%
GenAI-resolved support interactions	Not disclosed	>75% at year-end	New evidence

The enterprise client-density ratio improved, but the sharper signal appears inside customer operations: retail customers grew while average customer-operations headcount declined. Revolut also disclosed that more than 75% of support interactions were resolved by its GenAI-powered chatbot by year-end and that support NPS increased by 12 points. That is strong operating-leverage evidence in one function — but it still does not prove AI alone caused the outcome.

EVIDENCE DISCIPLINE

Separate observation from explanation

Observed

What the filings can support

— Customer, revenue and employee growth.
— Client-density changes when stock definitions are consistent.
— Function-specific headcount and workload movements when disclosed.
— Unit economics such as ARPAC or cost to serve when the company publishes them.

Inference

What management still has to test

— Whether automation caused the productivity change.
— Whether higher revenue per employee came from pricing, mix, credit growth or operating redesign.
— Whether released capacity was redeployed, removed or absorbed by new controls.
— Whether quality, risk and customer outcomes remained acceptable.

INTERACTIVE MODEL

Scale and productivity per employee

Enter base and current values. The gold point updates automatically versus FY2025 references.

Employees · baseEmployees · currentClients · baseClients · currentRevenue USD M · baseRevenue USD M · current

Clients / employee

13.1K-0.3%

Base 13.1K

Revenue / employee

$2M19.1%

Base $1M

Client leverage

-0.3%

clients − employees

Revenue leverage

21.9%

revenue − employees

FY2025 BENCHMARK

X = clients/employee · Y = USD/employee

ReferenceCli./emp.USD/emp.

Your company

Entered values

13.1K

$2M

Digital banking

13.1K

$2M

SoFi

Fintech / financial services

2.2K

$592K

Santander

Global banking

907.2

$356K

Interactive Brokers

Brokerage

1.4K

$2M

READING

High scale per employee and high economic productivity versus the reference set.

Directional comparison: customers, members and accounts are not identical denominators.

EUR/USD: Federal Reserve

COMPARABILITY RULES

Bad denominators create false productivity

Use year-end customers with year-end employees for stock-density ratios; use average FTE for annual revenue or annual cost where possible.

Prefer active customers to registered accounts when the disclosure allows it.

Do not compare “products per employee” unless both institutions define an active product the same way.

For payments, claims, support or collections, a transaction/case denominator can be more informative than customers.

Cross-company benchmarks are directional. The most powerful comparison is the same institution, same definition, over time.

SOURCES & METHOD

Public data, explicit limitations

2024 and 2025 values are drawn from Nu Holdings and Revolut public reporting. Ratios shown here are calculated from those disclosures and rounded. Different reporting currencies and business mixes mean cross-company economic ratios should not be treated as accounting-equivalent benchmarks.

Nu Holdings — 2025 Form 20-F↗Nu Holdings — FY2025 results↗Revolut — 2025 Annual Report↗Revolut — 2025 Annual Report PDF↗

THESIS

Automation becomes strategic when capacity changes

Count bots if you need an implementation inventory. Count licenses if you need procurement control. But if the question is whether the operating model improved, measure the relationship between economic output and the structure required to produce it. Clients per employee is a useful first ratio. The real answer appears only when scale, monetization, service cost and quality are read together.

Related research

Measuring Generative AI Productivity →AI agents, unpacked →Technology Obsolescence Management →