What 95 FTEs freed looks like: a benchmark of Dutch enterprise AI in 2025

The 2025 Freeday deployment cohort is the most detailed public dataset on enterprise AI automation in the Netherlands. Six organizations. 875,000 customer interactions automated. EUR 4.2 million in verified savings. 95 FTE equivalents freed.

This post unpacks what those numbers mean, where they come from, and what they tell organizations that are still in the planning phase.

The cohort at a glance

Six Dutch enterprises deployed Freeday AI digital employees across customer service, accounts payable, and KYC functions in 2025. The cohort spans financial services, consumer electronics, travel, and non-profit. Interaction volumes and automation rates varied by sector and use case.

Cohort total: 875,000 interactions, 80.9% average automation rate, EUR 4.2M savings, 95 FTE freed.

What 80.9% automation actually means

An 80.9% end-to-end automation rate means that out of every 100 inbound customer contacts, approximately 81 are resolved by the AI without any human involvement. The remaining 19 are escalated to a human agent, either because the query falls outside the AI's defined scope or because the customer requests human assistance.

This automation rate does not appear on day one. Teams achieve it after a full year of operation, refining the knowledge base through real customer conversations.

Early in a deployment, automation rates are typically lower. The AI encounters unknown edge cases and escalates them. The deployment team updates the knowledge base. Over time, the rate increases as coverage improves. Novum Bank achieved an 85% automation rate, the highest in the cohort, by handling mostly structured queries with clear right answers.

For CFOs building a business case, the appropriate assumption is 75% in year one, rising to 80-85% by year two as the knowledge base matures. Using 80.9% from day one will produce an optimistic business case. Using 75% will produce a credible one.

Where the 95 FTE equivalents came from

95 FTE equivalents across six deployments work out to approximately 16 FTE per organization. That range masks meaningful variation.

Bitvavo freed 26 FTE equivalents from 375,000 automated interactions. Novum Bank freed 15 FTE from a smaller but highly structured contact volume. The organizations at the lower end of the FTE range were earlier in their deployment cycle or had smaller initial contact volumes.

FTE equivalents are not the same as headcount reductions. The metric measures the equivalent human capacity required to handle the automated interactions. Some organizations redirected that capacity to higher-complexity work. Others used it to avoid hiring as volumes grew. A smaller number achieved genuine headcount reductions.

For any CFO modelling this: the FTE freed is real and verified, but how it translates into cost savings depends on your organizational decisions about what to do with the freed capacity.

The EUR 4.2 million figure explained

EUR 4.2 million in verified savings across six deployments is not a vendor-calculated estimate. It is based on the cost of human interactions replaced by automated ones, calculated using each organization's cost-per-interaction data.

The average saving per organization was approximately EUR 700,000, but the distribution was uneven. Bitvavo's EUR 1.4 million in savings reflects both its high interaction volume and the relatively high cost of customer support for financial services. Smaller or lower-volume deployments produced smaller absolute savings.

For organizations assessing their own potential return, the key variable is current cost per fully loaded interaction. If your cost per interaction is EUR 4, an 80% automation rate on 100,000 annual contacts produces EUR 320,000 in direct savings before implementation cost. If your cost is EUR 8, the same volume produces EUR 640,000. The economics of AI automation are highly sensitive to the current cost base.

What the CSAT data shows

Customer satisfaction scores across the cohort ranged from 2.67 to 4.24 out of 5.

Goede Doelen Loterij, with their digital employee Jennifer, achieved 4.24. This is the highest CSAT score in the cohort and notably higher than the lowest scores of 2.67 and 2.79 recorded at Hisense Gorenje and Novum Bank, respectively.

The pattern across the cohort is clear: higher CSAT correlates with better knowledge base maintenance and more carefully designed escalation paths. The organizations that invested in keeping their AI's knowledge current and that routed genuinely difficult situations to humans quickly consistently outperformed those that did not.

This has a practical implication. CSAT risk in an AI deployment is not primarily a technology risk. It is a risk in knowledge management and process design. Organizations that treat AI deployment as an ongoing operational capability rather than a one-time technology project achieve better results.

The ATAG implementation timeline: what 14 days means

ATAG, the Dutch home appliance manufacturer, went live with its AI digital employee within 14 days of signing the contract. This was for a consumer electronics customer service deployment handling fault code interpretation and spare parts queries.

14 days is not a marketing claim. It is the actual time from signed contract to live deployment in production. This timeline matters because it shifts how organizations approach AI deployment risk.

A project that goes live in two to four weeks is categorically different from a project that runs for six to nine months. The financial commitment required before you see results is smaller. The organizational change management is simpler. The window for cancelling or delaying the project is shorter.

For a COO assessing whether to proceed, the question is not whether you can afford to do this. The question is whether you can afford the four weeks it takes to find out.

How the 2025 results compare to the industry baseline

Traditional AI implementation projects typically take five to nine months to go live. First-generation chatbot deployments usually automate below 50% of contacts. Organizations that deployed rule-based automation between 2019 and 2022 now face expensive maintenance and systems that break when policies or products change.

The 2025 Freeday cohort sits well above those benchmarks. 80.9% average automation rate versus sub-50% for chatbots. Two to four weeks to go-live versus five to nine months for traditional projects. EUR 4.2 million in verified savings from a cohort that did not require 18-month IT programs to get there.

The Freeday platform page explains the architecture that enables those timelines.

What the data does not tell you

Any benchmark has limits. The 2025 cohort is six Dutch enterprises. They share certain characteristics: B2B and B2C contact centers in defined industries, the Dutch regulatory context, and specific CRM and contact center platforms. Their results may not translate directly to organizations in different sectors, with different contact mixes, or with fundamentally different operational structures.

The automation rate, CSAT scores, and FTE freed figures are real, but they reflect a specific deployment context. Before using them as direct inputs to a business case, compare your contact mix, knowledge base quality, and escalation design against the cohort average.

The Freeday security and compliance page covers the governance requirements that apply across all deployments in the cohort.

FAQ

What is a realistic automation rate for enterprise AI customer service?

Based on the 2025 Freeday cohort, the verified cohort average is 80.9%. Novum Bank achieved 85% on a structured contact mix. A conservative assumption for a new deployment is 75% in year one. The rate typically improves over time as the knowledge base develops.

How many FTEs can AI automation free in a customer service organization?

The 2025 Freeday cohort freed 95 FTE equivalents across six deployments, averaging approximately 16 per organization. The actual number depends on contact volume, automation rate, and current team structure.

How long does enterprise AI automation take to implement?

Freeday deployments go live in two to four weeks. ATAG's deployment went live in 14 days. This is significantly faster than traditional IT implementation projects.

What is the average ROI of enterprise AI automation?

The 2025 Freeday cohort delivered EUR 4.2 million in verified savings from 875,000 automated interactions. Individual returns varied from EUR 700,000 to EUR 1.4 million depending on volume and cost per interaction.

Does AI automation hurt customer satisfaction?

Not when deployed well. The highest CSAT score in the 2025 cohort was 4.24 out of 5. CSAT outcomes correlate strongly with knowledge base quality and escalation design, not with the decision to automate.

The 2025 benchmark data is the clearest evidence yet that enterprise AI automation in the Netherlands delivers at scale. The organizations in this cohort were not running pilots. They were running production deployments at meaningful volumes, and the results held up.

For organizations still at the assessment stage, the Freeday AI agents page explains how the deployment model works and what the integration requirements look like in practice.