AI Customer Support Agent for Europe

Share

In short

There are two routes and they cost two orders of magnitude apart, so picking the wrong one is expensive. SimplyBoost is our own product: an AI support agent that answers from your website, PDFs, FAQ and catalogue on website chat, WhatsApp, Instagram and Facebook Messenger, published from 39 USD a month. It is live in days and it fits most businesses. A custom agent, which Crux Digits builds, starts at a 2,500 EUR audit and makes sense when the answer has to come out of your own systems rather than your own content: an order in an ERP, a policy in a claims system, a patient record. This page is written so you can tell which one you are.

What it looks like in practice

Four moments, not four features

The same agent, in the places your customers already are.

After hours
The question that arrives when you are closed
Most support questions land outside the hours you staff. That is the gap the agent covers first, and the one nobody is paid to sit through.
Every channel
One agent, trained once
Website chat, WhatsApp, Instagram DMs and Facebook Messenger get the same answers, because it learns your content once rather than per channel.
The handover
What it does when it does not know
It says so and passes the conversation to a person, with the transcript attached, into Slack, email or your helpdesk.
Across Europe
In the language the customer wrote in
It answers in the language it was asked in. Deciding which languages you staff for the handover is the part that needs a decision.

The choice

Buy the product, or have one built

	SimplyBoost, from 39 USD a month	A custom agent, from a 2,500 EUR audit
Where the answers come from	Your own content: website, PDFs, FAQ, catalogue	Your own systems: ERP, WMS, CRM, claims or patient records, through their interfaces
Live in	Days. One line on the site, three steps on WhatsApp	Weeks. The integration is the work, not the model
Channels	Website chat, WhatsApp, Instagram DMs, Facebook Messenger	Whatever you already run, including the ones a product cannot reach
Who runs it after	You do, in the dashboard	Your team or your ICT partner, with the code and the evaluation set handed over
Best when	The twenty repeat questions are answerable from what you have already published	An answer is wrong unless it reads live data, or the process is regulated
Worst when	Every answer needs a live lookup in a system	The questions were never that complicated and a product would have done

What it does

What the agent actually does, and what it refuses to do

The useful part of a support agent is not the answering. It is the not-answering. An agent that guesses confidently is worse than no agent, because a wrong answer in writing is a commitment your customer will hold you to.

It answers from what you gave it. Website, PDFs, FAQ, catalogue. It does not know anything else, which is the point.
It says when it does not know, and hands over. To Slack, to email, to your helpdesk, with the conversation attached so nobody starts from the beginning.
It asks what you would ask. Name, email, phone, company, captured in the conversation rather than in a form nobody fills in, scored hot, warm or cold, exported as CSV.
It books. Into Google Calendar, at the end of the conversation, while the person is still interested.

On Instagram it does one more thing worth naming, because it is the feature people ask for by name: someone comments your keyword and the link lands in their DMs in seconds, on one post or on all of them, with up to six follow-ups. If you already run ManyChat, the keywords you use today come with you.

Europe

What changes when your customers are across Europe

Three things, and none of them is the model.

The AI Act tells them they are talking to a machine. Article 50 transparency applies from 2 August 2026: a system that interacts with people has to make that obvious. In practice this is one line in the opening message, and it costs nothing. Pretending otherwise is the part that costs.
Language is a support decision, not a translation one. An agent that answers Dutch in Dutch and German in German is straightforward; deciding which languages you will then handle when it hands over to a human is the real commitment.
GDPR applies to the transcript, not just the answer. A support conversation collects a name, an email and often an order. Decide the retention period before you switch it on, because the honest answer to a data request is the one you can produce.

Where the questions touch a regulated decision rather than an order status, an agent should route to a human by design. See the EU AI Act risk checker for where a given use case lands.

The numbers

What each route costs

SimplyBoost publishes three plans: Starter at 39 USD a month (website widget across 15 pages, 500 AI replies a month, calendar booking), Growth at 89 (30 pages, 2,000 replies, lead capture and workflows, plus WhatsApp, Instagram and Messenger), and Pro at 169 (50 pages, 6,000 replies, analytics with export, custom branding). Current plans and what is in them are on simplyboost.io/pricing.

A custom agent follows our three fixed steps: an AI audit and strategy at 2,500 EUR that decides whether you need one at all, a Production-ready MVP at 20,000 EUR working against your own systems, and production from 50,000 EUR, integrated, monitored and handed over with the evaluation set and the code. Prices exclude VAT and the audit is credited in full against the MVP if you continue within 60 days.

The gap between 39 a month and 20,000 is the whole reason the first section of this page exists. Most businesses asking for a custom support agent want the product and have not been told it exists.

What you can verify

KvK 87456346Registered in the Dutch Handelsregister
Since 2022Crux Digits B.V., Utrecht region
13 casesPublished, each with its own number

One real engagement

AI Discharge-Summary Automation (Clinical NLP)

What we found. Producing and interpreting discharge summaries by hand was slow and error-prone, holding up the discharge process and pulling clinicians away from care. The group needed faster, clearer documentation without compromising the clinician’s final review.

−60%Discharge-process turnaround time

ClearerDischarge summaries, fewer errors

Real result from a Crux Digits engagement; client name withheld pending permission. Decision support only: the clinician retains sign-off on every summary.

Read the full case study →

FAQ

Frequently asked questions

Should we buy SimplyBoost or have an agent built?

Buy the product when your repeat questions are answerable from content you have already published: opening hours, delivery, returns, what is in stock, what a service includes. Have one built when the answer is wrong unless it reads live data from a system, for example an order status in your ERP, a claim in a policy system, or an appointment in an EPD. If you are unsure, the 2,500 EUR audit answers exactly that question and regularly concludes the product is enough.

What does the agent do when it does not know the answer?

It says so and hands the conversation to a person, into Slack, email or your helpdesk, with the transcript attached. That behaviour is the point rather than a limitation: an agent that guesses puts a wrong commitment in writing, and in support a wrong answer costs more than a slow one.

Which channels does it answer on?

Website chat, WhatsApp, Instagram DMs and Facebook Messenger, with the same answers on all of them because it is trained once. On Instagram it also handles keyword comments: someone comments your keyword and the link arrives in their DMs in seconds, with up to six follow-ups.

Does it work in more than one language?

Yes. The agent answers in the language the customer writes in. The decision worth making before you switch it on is which languages you will staff for when it hands a conversation to a human, because that is where a multilingual promise actually costs something.

Do we have to tell customers they are talking to AI?

Yes. The EU AI Act's Article 50 transparency duties apply from 2 August 2026 to systems that interact with people, so the agent has to make clear it is not a person. It is one line in the opening message. Where a conversation touches a decision with legal or similarly significant effect, route it to a human by design rather than by exception.

Who owns the conversations and the data?

You do. For a custom build the code, the models, the prompts and the evaluation set are handed over and you can give the whole thing to another supplier. For SimplyBoost the terms, sub-processors and retention are published on simplyboost.io; set the retention period before you switch it on, because a data request is easier to answer than to explain.

AI customer support agent