03 · AI & Automation
AI and automation where they take work off your team: preparation, documents, knowledge, repetitive flows. With a human deciding and a measurable result.
01 How you recognise it
Three starting points we see regularly. One of them is enough.
Research, meeting preparation and summaries happen by hand, if they happen at all.
Read, classify, type, forward. Each step harmless on its own, together half a working day.
The answers sit in documents, tickets and wikis, not at the moment the question is asked.
02 What you get
Small, bounded and measurable. Only when it runs in daily work does the next one follow.
Proposal, approval, execution. For commitments, payments and external communication a human always checks.
AI runs inside CRM, ERP, email and document storage, not in an extra tool nobody opens.
03 What it looks like
Schematic example with fictitious data
An agent is only as good as its boundaries. This one prepares CRM maintenance and may do nothing that reaches the outside.
| Element | Definition |
|---|---|
| Assignment | For one case, add the missing company details and summarise the status |
| Tools | Read access to the CRM and an approved company data source, write access only to the draft area |
| Boundaries | No emails, no meetings, no changes to stage, amount or date |
| Output | A proposal per field with a source, plus a list of the points without a reliable source |
| Log | Every run traceable with time, input, sources used and result |
| Approval | The responsible person accepts field by field, nothing is written automatically |
Before going live we test ten cases whose correct result we know. What is judged is agreement with the source, not the impression the wording makes.
When it fits
Research, documents or internal coordination tie up capacity that is missing elsewhere.
What a first step delivers
One selected process, recorded and assessed: what rules can take over, where AI helps and what deliberately stays human.
What we need from you
Examples of real cases, anonymised if you prefer, and a person who performs the process today.
We only need these accesses and documents once we start working together. For a first conversation, a short description of your situation is enough.
04 Principles
Three principles that shape every FIOXX project, from the first application to running operation.
We start with the process and the question of which work should disappear, not with the model. Whether AI, classic automation or a process change is the answer is decided by the benefit.
AI prepares, runs routine and documents. Wherever decisions have consequences, such as commitments, payments or external communication, the approval stays with a human.
AI features run inside the systems you already use: CRM, ERP, email, document storage. Not in an extra tool nobody opens.
05 Fields of use
Eight fields in which we implement AI and automation, always connected to your systems.
Agents that take on multi-step tasks: research, combine data, produce proposals and trigger defined actions in systems, within clear boundaries.
Assistants for your team that answer questions about products, processes and projects from approved sources, with a source reference and permissions.
Flows that involve several systems and people are orchestrated: triggers, rules, AI steps, approvals and escalations.
Account research, conversation preparation, summaries, follow-up drafts and CRM enrichment, the preparatory work in sales.
Read and classify incoming documents, extract and check data, generate outgoing documents from system data.
Make scattered knowledge accessible in a structured way: documents, wiki, tickets and CRM as one searchable base that respects permissions.
Enrich records, detect duplicates, derive status and tasks, transfer cases between CRM and ERP.
Research, content preparation and signal assessment for marketing workflows. Under the IYOS brand FIOXX works on exactly that.
06 Definition
An AI agent is software that pursues a goal, plans and carries out several steps for it, such as gathering information, querying systems and triggering actions. It works within defined rules and approvals.
The difference to a chatbot: an agent does not wait for the next question but works through an assignment. The difference to classic automation: it can handle unstructured input and derive intermediate steps itself. More on AI agents
An AI assistant answers questions and supports tasks based on approved company sources, with a source reference, permissions and the honesty to mark the unknown as unknown.
Typical uses are knowledge questions in sales and support, onboarding and preparing offers. More on AI assistants
Workflow automation connects triggers, rules and systems into a flow; within it, AI takes over the steps that require understanding, summarising or classifying.
Example: a document arrives, the AI recognises type and content, the system creates the case in the ERP, a human checks in case of a deviation. More on process automation
07 Human approval
We do not build approvals as a hurdle but as a design principle. Four patterns that have proven themselves.
The standard path for anything with an external effect: the AI produces a proposal, a human checks and approves, the system executes and logs.
Routine within defined limits runs automatically; deviations such as unusual amounts, unclear data or new cases go to a person.
For commitments, contracts, payments and external communication a human always checks, regardless of how confident the model appears.
Every AI step is logged: inputs, sources, result, approval. That way every decision can be explained later.
08 Security and governance
AI in a company needs rules before it needs speed. These principles apply in every FIOXX project.
AI components only see the data they need for their task. The permissions of the source systems are respected, not bypassed.
Whether a hosted model from a provider, an EU region or a model we run ourselves: the choice follows data protection, cost and quality, per use case.
Your data is not used to train third-party models. Contractual assurances from providers are part of the selection.
GDPR, data processing agreements and the requirements of the EU AI Act are checked per use case and taken into account in architecture and process.
Every automation has a metric: time saved, cycle time, error rate, response time. What cannot be measured is not sold as a success.
If an AI step fails or delivers uncertain results, the process continues manually under control, never unnoticed into the void.
09 AI × software
Many AI applications can be built with existing platforms. Where data volumes, interfaces or security requirements go beyond that, FIOXX develops the software to match: our own integrations, internal tools or a product. With IYOS we show that we do not only advise but build ourselves.
10 FAQ
No. It takes over the preparatory, repetitive and documenting work so people have more time for conversations, decisions and exceptions. The processes we build treat the human as the one who decides, not as the error correction at the end.
Less than people often assume. For assistants and automations the existing documents and CRM and ERP data are usually enough. What matters is not the volume but access, structure and quality. Preparing the data is often the first step.
A clearly bounded case, such as conversation summaries into the CRM or document classification, can be built and measured in a short time. We deliberately start small, with a process that happens daily, and extend from there.
We come from hands-on sales and we build software. That is why we start with the business process, not with the technology. And we can deliver both: the process design and the implementation including integration into CRM, ERP and existing systems.
Next step
We check together whether and how AI or automation can take on that work, and what should stay with people.