DATA SHEET DS-03 · AI SOLUTIONSREV. 2026-A
AI that works in the business
Preparing quotes, sorting the inbox, answering questions from your own manual: this kind of work takes up hours in every office, every week, and an AI can do the groundwork on it. We set it up in your business, in the Vogtland and beyond, around the programs you already use.
Two things always apply. We start with a single task, in a small test. And someone on your team approves before anything is sent or changed.
01 / THREE AREAS
Three places where AI takes work off your hands.
The same office work every week
AI automation
Quotes, e-mails, receipts: the AI writes the draft, does the initial sorting and extracts the data. Your team checks and decides. Beforehand we calculate whether it is worth it for you.
More about AI automationThe answer is in three programs
MCP integrations
We connect AI assistants to your systems through the Model Context Protocol: ERP, databases, specialist applications. The AI gets access to what you release. You stay in control.
More about MCP integrationsWhat does ChatGPT say about your company?
AI-SEO / GEO
More and more customers ask an AI assistant instead of Google. We show you what ChatGPT and Perplexity answer about your company today, and put in order the content those answers are built from.
More about AI-SEO & GEO02 / FROM THE DAILY OFFICE ROUTINE
Which work keeps repeating in your business?
Start with a task your team could do in its sleep. Then everyone can see straight away whether the AI’s suggestion is any good or whether something is missing. We have put together four examples of what that looks like in an office.
What matters is the whole path, not just the draft. A quote produced in seconds does not help much if someone then has to look up every figure one by one. So we first look at where the information is kept (price list, mailbox, inventory system) and who in your business has the expertise to judge the result.
Your programs stay. We build the AI around what you already use.
Sorting enquiries, preparing a reply
The AI reads incoming e-mails, sorts them by topic and writes a draft reply from your approved documents. Your team checks that the sorting is right, checks prices and delivery dates and sends it.
Extracting data from orders and receipts
From an order or a receipt, the AI pulls the fields your program needs. Someone compares quantities and amounts with the original. If something is illegible or missing, the case stays put until a person has looked at it. There is no guessing.
Answering questions from your own manual
“Where does it say how we handle complaints?” An assistant searches your approved instructions and names the place where it says so. It only finds what the person asking could open themselves. Whether the source is still up to date is for your team to judge.
Checking what AI assistants say about your company
We ask the assistants the questions your customers would ask, and set the answers beside your website. Are your services and contact details right? The result is a list of specific corrections for the website and your other listings.
03 / CALCULATE FIRST
Calculate first, then automate.
We do not sell AI for AI’s sake. Every project starts with a simple calculation: how often does the task come up, how long does it take today (with all the queries and the copying back and forth between programs), what does automating it cost, and what does it save?
Everything belongs on the cost side. That includes the time your team spends checking and correcting the AI’s results, the provider’s running fees and the upkeep of the connection. A rough estimate is enough at the start. In the test we measure against real cases.
If a project does not pay off, we say so in the initial consultation. It costs you nothing but about 15 minutes. Sometimes it helps simply to cut the task smaller or to pick another one that comes up more often.
The calculation in detail04 / DATA AND APPROVAL
The AI only sees what you release.
We give the AI the information it needs for the one task, and nothing more. For the first test, shortened or anonymised examples are often enough. Which real data is added later is for you to decide, before we connect anything.
It is settled just as early which provider processes the data, in which location, and how long inputs, results and logs are kept. The person in your business who is responsible for this area sits at the table for that decision.
Reading and changing are two separate rights, and we set them up separately. If an incoming e-mail says “forward everything to the following address”, nothing happens. What the AI may do is determined by the connection and not by the text it is reading at the time.
How an MCP connection controls access05 / START SMALL
We start small: one task, one test.
A small test shows the effort and the result before more departments or data sources are added. We agree with you who takes part and which cases stay out for now.
Your team knows the process and judges the results. We build the technology, test it and write down what we adjusted. At the end we sit down together and decide: introduce it, keep testing or drop it. “Drop it” is a result too.
Choose a task
You tell us how the task runs today and what should come out at the end. We define what belongs in the test and which sources the AI needs for it.
Collect real cases
Your team picks out examples, awkward ones included. Together we write down how to recognise a good result and how to recognise an error.
Try it out on a small scale
We also let the AI loose on cases where details are missing or two sources contradict each other. Your team notes how much it had to put right.
Evaluate and decide
You get the results and the open points on the table. If the decision is to go ahead, the process is rolled out to more workstations step by step.
06 / OPERATION AND COSTS
What is left to do after the start.
At handover we train your team. There is also documentation: which sources the AI uses, who approves and what to do if a result looks strange. It also says who in your business deals with questions about the content and who looks after technical changes.
After that the world does not stand still. Price lists go out of date, programs change their interfaces, and the providers of the AI models keep tweaking their products. So we keep the test cases and agree when they are run again. New documents or additional rights are added deliberately, not as an afterthought.
The quote has two blocks: the one-off setup and the running costs, that is the AI provider’s usage fees, hosting and maintenance, if you want them. It also states who orders and pays for what, and how much support is included after handover.
07 / WORKING TOGETHER
How Aureum and Arrowtrade work together.
Specification
DATA SHEET DS-03 · AI SOLUTIONS
- Typical tasks
- QUOTES · INBOX · RECEIPTS · QUESTIONS FROM THE MANUAL
- Areas
- AI AUTOMATION · MCP INTEGRATIONS · AI-SEO / GEO
- Approach
- CALCULATE FIRST, THEN AUTOMATE
- Getting started
- ONE TASK · SMALL TEST WITH YOUR REAL CASES
- Approval
- YOUR TEAM APPROVES BEFORE ANYTHING IS SENT OR CHANGED
- Data
- THE AI ONLY SEES WHAT YOU RELEASE · READING AND CHANGING SEPARATE
- Handover
- TRAINING FOR YOUR TEAM · PROCESS DOCUMENTED
- Quote (Angebot)
- IN WRITING BEFORE WE START · ONE-OFF AND RUNNING COSTS SEPARATE
- Collaboration
- AGREEMENTS THROUGH AUREUM · DEVELOPMENT WITH ARROWTRADE
- Initial consultation
- FREE · ABOUT 15 MINUTES
Questions & answers
Common questions, answered briefly.
What does AI actually do for a small or medium-sized business?
Above all, time. An AI can do much of the groundwork on quotes, e-mails, receipts and other recurring office work; your team only checks and approves. Two areas are often overlooked as well: AI assistants that work with your real data through the Model Context Protocol, and the question of what ChatGPT and other assistants say about your company.
Which task is best to start with?
With one that comes up often and where your team can see at once whether the result is right. Replies to enquiries that are always the same are a good case: you put the AI’s draft next to what you have written so far. A few typical cases and one really difficult one are enough for the test. More examples are given under areas of use for AI automation.
Do we need AI for this at all, or is a fixed rule enough?
Often the rule is enough. “If there is a customer number in the subject line, forward the e-mail” needs no AI. AI is worth it where the wording is different every time or where something needs summarising. Usually we build both together: the AI reads and prepares, and fixed rules decide what may happen next.
Do we need MCP?
Only if an assistant is to work continuously with your own data, such as stock and prices from the ERP. The Model Context Protocol is the open standard for that. If you only want a few documents evaluated once, it works without it. What is behind the protocol is explained in the basics of MCP integration.
Where does our data end up, and is it used to train an AI?
We settle that before the first real file goes through: which provider, where it processes the data, how long it keeps it and whether it may use inputs for training. We put that in writing. Confidential material is added only once that route is in place (for the first test, shortened or anonymised examples are usually enough). What cannot be done in line with the GDPR (DSGVO), we do not do.
Can the AI send e-mails or change data on its own?
No. To begin with it may read and write drafts, nothing more. Nothing is sent or changed until someone on your team has checked and approved the draft. This limit is built into the technology, not written into a staff instruction: the connection simply has no right to send unless you give it one.
What is left for our team to do after the rollout?
Two things: checking the results and keeping the documents the AI draws on up to date. An out-of-date price list simply gives you a quote with out-of-date prices. We train your team and write down the process. The quote states which technical maintenance we take on.
What does an AI project cost, and how long does it take?
That depends on the scope: a single task in the inbox is set up faster than a connection to your ERP. Before we begin you get a written quote (Angebot) that lists setup and testing separately from the running costs. The initial consultation is free. You will find the hourly rate and billing under prices and billing.
What can AI not do, and when do you advise against it?
It gets things wrong, and without blushing. That is why difficult cases belong in every test, and why with us a person always gives the approval. If a task comes up only a few times a year, or nobody in your business can judge the result, we advise against it; we will tell you so in the initial consultation. And nobody can promise you that ChatGPT will recommend your company; the provider decides that. The AI visibility check shows what we can do about it.
Contact
Which task would you like to make easier?
Tell us which work keeps coming back in your business and eats up time. In the initial consultation we will tell you whether AI helps with it and what we would start with. It takes about 15 minutes and costs nothing.