RAG system on your knowledge
No generic LLM answers. Everything cited from your sources. With source links in the answer output.
03 · Service
Support that scales with you.
The Service Machine is a RAG system on your knowledge base. It answers 70 to 80 percent of inquiries autonomously, escalates the rest with full context to your humans. You grow without doubling the support team.
The problem
Growth means more support tickets. But not more headcount.
When you grow, tickets grow. Faster than you can hire humans. Generic chatbots are worse than no help. They send customers in loops, answer wrong, point at outdated FAQs. Plus: running a US cloud bot over your contracts is GDPR-risky.
- Support volume grows faster than the team
- Generic chatbots frustrate customers more than they help
- Knowledge sits in 5 tools: Confluence, help center, PDFs, emails, heads
- US AI bots are GDPR-tricky for contracts and data
- Nobody measures which answers actually help
Before / After
What the service inbox feels like on a Monday morning.
Drag the slider. On the left, an inbox before the Service Machine. On the right, the same inbox an hour later, after the machine has absorbed routine. Complex cases stay in the inbox, but sorted and prepped.
Inbox · 9:14 · AFTER 7 of 9 solved autonomously
- anna.lehmann@… 08:14
Delivery times to CH?
Service Machine: 5–7 business days, free above 5,000 €
- logistik@maerz.de 08:09
Warranty on pump XR-12
Service Machine: 24 months, source: T&C §9
- Frieder Stein 08:02
Change billing address
Service Machine: Self-service link sent
- buero@dehn-bau.de 07:51
Lost the manual
Service Machine: PDF link from customer area sent
- service@klingauer 07:46
Missing shipping confirmation
Service Machine: Tracking pulled from SAP, forwarded
- kontakt@nds-hh 07:40
Bulk order Q3
Escalated to sales (special pricing)
- Sandra Wenz 07:33
Filter F3 complaint
Escalated to QM (complex complaint)
- presse@nf-werk 07:21
Datasheet model 4
Service Machine: Datasheet link from docs sent
- h.bartels@… 07:18
Which cables fit module B?
Service Machine: Table from manual chapter 3
Inbox · 8:14 · BEFORE 9 unread, all on humans
- anna.lehmann@… 08:14
Delivery times to CH?
- logistik@maerz.de 08:09
Warranty on pump XR-12
- Frieder Stein 08:02
Change billing address
- buero@dehn-bau.de 07:51
Lost the manual
- service@klingauer 07:46
Missing shipping confirmation
- kontakt@nds-hh 07:40
Bulk order Q3
- Sandra Wenz 07:33
Filter F3 complaint
- presse@nf-werk 07:21
Datasheet model 4
- h.bartels@… 07:18
Which cables fit module B?
Tip: on a phone, drag the slider left and right with your finger.
How it works
From knowledge to running bot in 4 steps.
RAG, built clean. No hallucinations, no black box.
-
Step 01
Knowledge base ingest
We pull in your help articles, FAQs, contracts, PDFs, Confluence pages, Notion docs. Structured, deduplicated, versioned.
-
Step 02
Build the RAG pipeline
Embeddings, vector DB (Qdrant in the EU), retrieval layer, prompt engineering. Answers are cited from your sources only.
-
Step 03
Integrate the chat widget
Into your website, your helpdesk (Zendesk, Intercom, Freshdesk) or your customer portal. Brand-styled.
-
Step 04
Escalation layer
When the answer is uncertain or the customer wants a human: clean handover with conversation history and proposed solution.
What you get
A bot that gives answers. Not phrases.
On your knowledge base. In your tone. With measurable KPIs.
Multi-channel integration
Website widget, helpdesk plugin, Slack bot for internal teams. One knowledge base, multiple inputs.
EU LLMs
Mistral Large (France) or Anthropic Claude (EU tenant). No US cloud, DPA included.
Clean escalation
When the bot isn't sure: handover to human with context, conversation and a proposed solution.
Quality reports
Which inquiries came in, which were resolved, which escalated. Plus: knowledge gap report. What should land in the knowledge base.
Continuous tuning
AI expert checks answer quality monthly, calibrates thresholds, suggests knowledge base extensions.
Investment
Heavy setup, then passive.
Service Machine is the highest-margin of the three. High setup value, low ongoing effort.
All prices net, plus VAT. No minimum term, cancel monthly to month-end. LLM credits billed by volume above standard package.
Setup
€7,500
Knowledge base ingest, RAG pipeline, chat widget, escalation logic, EU hosting. One-time.
Monthly operation
from €2,400
EU LLM credits, tuning, knowledge updates, quality reports, AI expert.
FAQ
FAQ · Service Machine
-
Does the bot hallucinate?
We build the pipeline so it only cites from your knowledge base. If no relevant answer exists in your sources, the bot says so explicitly and escalates. Hallucinations are excluded by setup. No answers without source. -
Which helpdesk systems are supported?
Zendesk, Intercom, Freshdesk, HubSpot Service Hub, Helpscout. Natively. Others via API/webhooks. Plus: standalone chat widget for your own webspace. -
How fast are the answers?
1 bis 3 seconds on average. Depends on the LLM (Mistral Large usually faster than Claude), but under 5 seconds in 95% of cases. -
What happens to customer data?
Stays in the EU. Vector DB runs in Frankfurt, LLM calls via EU endpoints, no training on your data, DPA with you and with the LLM providers.
30 minutes, honest assessment 
Tell us what's burning right now.
We reply within one business day. No sales pitch, no drip campaign. Florian or a teammate calls, listens, frames it.
A personal callback
We call you back personally.
You write here, someone from our Hamburg team calls you. No sales team, no call centre, no email funnel. We listen, frame the problem, and tell you honestly whether our solution fits your situation.
The White Fox team · Hamburg