Most of the time in service isn't spent on answers, but on everything around them: reading the request, sorting it, finding the right person, looking up the order status in the ERP, piecing together a reply. That's exactly the chain we automate, regardless of channel. In the end, your team reads a finished draft and approves it, instead of starting from zero.
Requests come in via email, phone, form, and WhatsApp, and all land in the same overcrowded inbox. By the time someone knows what it's about, who's responsible, and what the customer's status is, ten minutes are gone — and the actual answer takes two.
We're not replacing the answer, but the work before and after it. Each of these steps can be introduced individually — you don't have to switch everything at once.
Every incoming request automatically gets a category: complaint, status question, quote request, invoice matter. It then goes to the responsible person or group, with priority and deadline. Manual sorting is no longer needed.
Savings 2–4 h per week in the inboxFor recurring cases, an AI agent writes the draft reply from your approved knowledge base and the case data. The draft lands in the responsible person's inbox, not with the customer. A human reads it, corrects if needed, and sends it.
Impact Check the reply instead of inventing itWhen a customer asks about a delivery's status, the workflow pulls the current status directly from the ERP or inventory system and attaches it to the case. Nobody has to jump between inbox and system anymore. For clear-cut cases, the information can be sent back directly.
Savings No manual lookup per requestA complaint isn't just text — it's a process: capture, review, decision, credit note or replacement, feedback to the customer. We map this workflow with clear approval points, so nothing gets stuck and every step is logged.
Impact No case gets stuck unnoticedBecause every request gets a category and a timestamp, you see for the first time which cases really tie up your service. You get this analysis automatically every week. That's the basis for deciding what to automate next or fix at the root.
Impact Decisions based on numbers, not gut feelingThe process is the foundation; the channels build on top of it. Which path we choose is decided by your system landscape, not a partner program.
Process EngineThe service process runs as a workflow: request comes in, gets classified, data is pulled from the system, assigned, a draft is generated, approval is awaited. Self-hosted in Germany on request, so customer data never leaves the building. Every step stays visible and individually traceable.
For drafts to be accurate, the agent needs a vetted knowledge base: product documentation, service rules, your team's typical answers. You maintain this base, not us. It's also the foundation for when a website chat or a voice agent is added later.
If most of your requests come in by phone, the best email workflow won't help much. Then a voice agent takes calls outside business hours and hands them over, structured, into the same service process. Channel and process stay separate by design.
We don't automate the entire service at once — just the one workflow carrying the most load, live in 30 days and at a fixed price.
We look at your request types, the channels, and the systems behind them. In the end, the question is which workflow ties up the most time and whether it can be automated cleanly.
You get in writing which request type will be automated, which systems will be connected, and what it costs. A fixed sum, no open-ended day rate.
We build the workflow and first run it in shadow mode: the agent generates drafts, your team compares. Only once the results hold up does the workflow go live.
Your team gets access, documentation, and a walkthrough; you keep control over the rules and text templates. The analysis then shows which workflow is worth tackling next.
This page describes the process behind the service, regardless of which channel the request comes through. If your bottleneck is clearly on the phone, because nobody picks up, AI telephony is the faster entry point.
In the free initial analysis, we spend 60 minutes going through your requests, channels, and systems, and name the one workflow that pays off first. No obligation — and if your service doesn't have an automation problem, we'll tell you that too.
No, not without your decision. In the standard setup, the agent generates a draft reply that a human reviews and approves before it reaches the customer. Only once you have enough confidence after a test phase, for a clearly defined case such as pure status information, can you switch that one case to automatic sending. You keep control over what goes out directly.
That depends on how many channels and systems are connected. NordFlux works at a fixed price: after the free initial analysis, you get an offer with a fixed sum for the first workflow, no open-ended day rate. We state ongoing costs for hosting and model usage separately in advance.
In the businesses we see, that's rarely the goal: the service is overloaded, not overstaffed. What falls away is the sorting, looking up, and retyping, not the conversation with the customer. Your team handles the same volume of requests with less friction and finally gets to the cases that really need a human.
In most cases yes, the interface is what matters. We connect systems with an open API directly; for older systems, we check export, database access, or, if in doubt, robotic operation of the interface. In the initial analysis, we look concretely at your system landscape and tell you honestly if a connection would cost more than it's worth.
It stays wherever you decide. On request, we run the service workflow self-hosted in Germany, conclude the necessary data processing agreements, and make sure requests aren't used to train third-party models. You decide which data the agent is even allowed to see and how long it's stored.
