The verdict in three sentences
For a mid-sized Amsterdam company receiving 8,000 customer emails a month split across 6 teams, AI triage costs 12,000 to 25,000 EUR excluding VAT for integration and 150 to 600 EUR a month in API usage (2026 order of magnitude). A reasonable goal is 85% correct classifications in the first months and first response time cut in half, because each request lands directly in the right queue with its priority. The non-negotiable condition is EU data hosting and documented GDPR scoping before go-live.
What the AI actually does with each email
A language model reads the email, its attachments and the customer history, then returns a structured decision. It does not answer the customer in place of your teams, it prepares the work.
| Task | Without AI | With AI triage |
|---|---|---|
| Identifying the reason for contact | Human reading, 1 to 2 minutes | Automatic, under 5 seconds |
| Routing to one of 6 teams | 20 to 25% errors, re-routing | 85% correct assignment targeted, the rest to a review queue |
| Priority (urgency, key account, churn risk) | Rarely assessed | Score assigned to every request |
| Data extraction (order number, contract, amount) | Manual lookup | Fields pre-filled in the ticketing tool |
| Draft reply suggestion | None | Optional, approved by the agent |
| Duplicate and follow-up detection | Manual | Threads grouped automatically |
The key is the review queue: when the confidence score is low, the email goes to a human. This mechanism is what makes an 85% rate acceptable from day one.
Detailed budget
Integration cost depends mainly on your ticketing tool and the number of contact reasons to distinguish.
| Item | 2026 range (EUR excl. VAT) | Comment |
|---|---|---|
| Scoping and contact-reason taxonomy | 2,000 to 4,000 | 25 to 40 reasons grouped into 6 queues |
| Connection to mailbox and ticketing | 3,000 to 6,000 | Zendesk, Freshdesk, Salesforce or in-house tool |
| Prompts, rules and test set of 1,000 real emails | 3,000 to 6,000 | Accuracy measured before go-live |
| Classification monitoring dashboard | 2,000 to 4,000 | Correct assignment rate, delays, volumes |
| GDPR compliance, pseudonymisation, logging | 2,000 to 5,000 | DPIA if sensitive data is involved |
| Total integration | 12,000 to 25,000 | 6 to 10 weeks |
| Monthly API usage | 150 to 600 per month | Depends on model, email length and attachments |
At 8,000 emails a month, API cost comes to 0.02 to 0.08 EUR per email. Choosing an EU-hosted model, or a provider committed to not using your data for training, shifts that cost by 20 to 40%.
GDPR and the AI Act: the rules of the game
Classifying customer emails is not a high-risk use under the AI Act, but GDPR fully applies, and the Dutch Data Protection Authority (Autoriteit Persoonsgegevens) is an active regulator. You need a data processing agreement with the AI provider, EU hosting or solid transfer safeguards, limited retention of content sent to the API (often zero days) and a mention in your privacy notice. If health or banking data passes through emails, a DPIA becomes necessary and pseudonymising before sending is recommended.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Daan, head of customer relations at a 450-employee services company in Amsterdam (hypothetical persona), receives 8,000 emails a month. Manual triage takes 1 minute per email, or 133 hours, and 20% of requests are misrouted, each re-routing costing 5 minutes, another 133 hours. With AI, only 15% of emails go to human review (20 hours) and the error rate drops to 6% (40 hours). The gain reaches 206 hours a month, about 6,600 EUR at 32 EUR per loaded hour. With 400 EUR of monthly API usage, the net gain is 6,200 EUR: a 20,000 EUR integration pays back in just over 3 months.
FAQ
How much does AI triage of customer emails cost in Amsterdam?
Budget 12,000 to 25,000 EUR excluding VAT for integration and 150 to 600 EUR a month in API usage for 8,000 monthly emails. Cost per email is 0.02 to 0.08 EUR.
What classification accuracy can we reach?
Targeting 85% at launch is realistic with a clear taxonomy and a test set of 1,000 real emails. After 3 months of tuning, 90 to 93% is common on well-defined reasons.
Does the data stay in Europe?
Yes, if you choose an EU-hosted model or a provider offering European data residency. This raises API cost by 20 to 40% depending on the offer.
Do we need to change our ticketing tool?
No, the AI plugs into your existing tool via API. An in-house tool without an API needs an extra connector at 3,000 to 6,000 EUR.
How do we measure first response time?
Measure for one month before launch, then weekly afterwards. Going from 18 to 9 business hours is a common order of magnitude when routing was the main cause of delay.
Let's scope your project. Send us your monthly volume, your ticketing tool and your list of teams, and we will return a taxonomy, an indicative budget and a test plan on your real emails. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
