4B Systems
AutomationCase studiesBlogContactGet a quote
AI agents · automation

AI agents for business – implementation and business process automation

An AI agent doesn't just answer – it does the work. It reads emails and documents, checks data in your CRM and ERP, and prepares quotes, replies and reports. We design and deploy agents that work in your systems and leave the important decisions to people.

SalesCustomer serviceDocuments and invoicesHR
CRMSalesWizardEmailCalendarERPKSeFSpreadsheetsPDF documentsHelpdeskPaymentsTelephonyMessaging appsMCPn8nMakeREST APIClaudeGPTGeminiLlama

What is an AI agent and how does it work?

An AI agent is a program based on a language model that is given a goal, plans the next steps itself, uses tools – CRM, email, documents, databases – and checks whether the result is correct. When a case requires a decision, it stops and asks a human for approval.

1Goale.g. “prepare a quote”
2Plansteps to take
3ToolsCRM, email, ERP
4Actionthe agent carries out the steps
5Checkverifies the result
6Humanapproves what matters

Chatbot, automation or AI agent?

These terms often get mixed up. The difference is simple: a chatbot talks, automation runs a fixed scenario, and an AI agent pursues a goal and chooses the steps itself. The best implementations combine all three – we fit the solution to the process, not the other way around.

SolutionWhat it doesTriggerWorks in your systemsBest for
AI chatbotAnswers questions in a conversationConversationNo – it only answersFAQs, website help
Automation (n8n, Make)Runs a rigid “if A, then B” scenarioEventYes, but by fixed rulesRepetitive, predictable workflows
AI agentPlans the steps itself, uses tools and evaluates the resultGoal or eventYes – adapts its steps to the situationTasks with exceptions, documents and decisions

Where do AI agents work in a company?

An agent works well wherever your team spends the day reading, retyping and checking information across several systems.

Sales

Lead qualification, CRM updates, draft quotes from the price list, follow-up reminders, call summaries.

Customer service

Answers to tickets from the knowledge base, order status, complaints – harder cases go to an agent with a ready-made summary.

Finance and back office

Reading invoices and documents, matching payments, checking against purchase orders, preparing data for accounting.

HR and recruitment

CV screening against criteria, scheduling interviews, onboarding tasks, answering employee questions.

Operations and logistics

Monitoring orders and stock levels, detecting delays, raising issues with suppliers, daily reports.

Law firms and compliance

Analysis of contracts and letters, deadlines, comparing document versions, draft replies for a lawyer to review.

Marketing

Data-driven drafts of content and campaigns, customer review analysis, performance reports.

IT and helpdesk

Ticket classification, log analysis, suggested solutions, technical documentation updates.

Agent teams

Several specialized agents collaborate in one process – e.g. a sales agent hands a task over to a quoting agent.

What do you gain with AI agents?

An agent doesn't replace your team – it takes the tedious part of the work off their plate so people can focus on customers and decisions.

Faster service

An inquiry, ticket or document is handled right away – evenings and weekends included. People deal with what needs a decision.

Less manual work

The agent takes over retyping data, searching systems and preparing drafts. Your team handles more cases without adding headcount.

Fewer mistakes

The agent checks data in the source systems and follows the same rules for every case.

Full transparency

Every agent action is recorded: what it did, on what basis and who approved it. It's easy to check and correct.

Works in your systems

The agent works in the CRM, email and ERP you already use – your team doesn't have to switch tools.

Start with one process

We start with a pilot that quickly shows results. You add more processes once the first one has proven itself.

Implementing AI agents in your company – step by step

We start with one process and low risk. The agent first works on test data, then under human supervision – and only once it has proven itself does it get more autonomy.

1

Process audit

Together we choose the process where an agent will deliver results fastest.

What we doProcess map, list of steps, data and systems, risk assessment.
What you getA recommended first process and success metrics.
2

Agent design

We decide what the agent can do on its own and what a human always approves.

What we doPermissions, tools, approval points, exception scenarios.
What you getAgent specification and integration plan.
3

Prototype on test data

The agent works on a copy of your data – with no risk to the business.

What we doThe first version of the agent and integrations with your CRM, email and documents.
What you getA demo on your own examples in 1–2 weeks.
4

Integrations and security

We connect the agent to your systems via API and MCP.

What we doMinimal permissions, logs of every action, cost limits, encryption.
What you getAn agent ready to work with real data.
5

Testing and evals

We test the agent on dozens of real cases.

What we doTest scenarios, accuracy, resistance to prompt injection.
What you getA quality report and a list of fixes.
6

Supervised pilot

The agent works alongside your team – a human approves its decisions.

What we doTeam training, monitoring, collecting feedback.
What you getMeasurable results: time, number of cases, errors.
7

Scaling and development

We expand autonomy and add more processes.

What we doModel cost optimization, new tools, more agents.
What you getMaintenance, SLA and a development roadmap.

What we build AI agents with

We fit the technology to the process, not the other way around. We design each agent so the AI model can be swapped later without rebuilding the whole solution. Increasingly, we connect agents to your systems via MCP (Model Context Protocol).

AI models

Claude, GPT, Gemini – or open-source models (Llama, Mistral) on your server when data must not leave the company.

Agent orchestration

Custom code and proven libraries (e.g. LangGraph) for complex agents; n8n or Make for simpler workflows.

MCP and APIs

Model Context Protocol and APIs connect the agent to your CRM, email, documents and databases – with permission control.

Memory and knowledge

Vector databases (pgvector, Qdrant) and RAG: the agent draws on your procedures, price lists and case history.

Queues and schedules

Agents triggered by an event (new email, lead, invoice) or on a schedule – with automatic retries on errors.

Monitoring and costs

Logs of every action, a supervision dashboard, model spending limits and alerts when the agent needs a human.

Control and security

The agent works, a human decides

An AI agent has access to your company's systems, so we design security from day one, not at the end. Every agent follows these rules:

✓A human approves decisions with consequences: sending a quote, making a payment, changing a contract.
✓The agent has only the permissions it needs – and nothing more.
✓Every action is recorded in a log: what, when and on what basis.
✓Protection against prompt injection: content from emails and documents is never treated as instructions by the agent.
✓Data handled in line with GDPR; on request, models in the EU or on your infrastructure; designed with the EU AI Act in mind.
✓Cost limits and a kill switch – the agent can be stopped with a single click.
Contact us

Which process will you hand over to an AI agent? Let's talk

Describe the process that takes up most of your team's time – handling inquiries, documents, tickets, reports. We'll call you back, ask about the details and suggest what the first agent in your company could look like.

✓Free call: which process is worth handing over to an agent
✓A prototype on your own examples in 1–2 weeks
✓We reply within one business day
✓NDA before we discuss the details – on request

Prefer another way? Contact details and office addresses →

Let's talk about your AI agent

Fields marked with * are required.

Tick the consent box to send your inquiry.

Related services

AI agents for business – frequently asked questions

An AI agent is a program based on a language model that is given a goal (e.g. “prepare a quote for this inquiry”), plans the steps itself, uses tools – CRM, email, documents – and checks the result. Unlike a chatbot, it doesn't just answer: it does the work in your company's systems.

Classic automation runs a rigid “if A, then B” scenario and stops at exceptions. An AI agent understands the content of an email or document and adapts its steps to the situation. In practice we often combine the two: predictable steps in the automation, decisions and documents in the agent.

It can carry out low-risk steps on its own, such as looking up data or preparing a draft. Decisions with consequences – sending a quote, making a payment, changing a contract – are always designed with human approval. We increase autonomy gradually, once the agent has proven itself in a pilot.

With one process that is repetitive, takes up a lot of your team's time and has a clearly measurable outcome – e.g. handling quote requests or documents. We run a short audit, a prototype on test data and a pilot with human oversight.

We usually show a prototype working on your examples within 1–2 weeks. A full implementation with integrations, testing and a pilot typically takes a few weeks, depending on the number of systems and the complexity of the process.

Any system that has an API or can be connected to: CRM (including our own SalesWizard), email, calendars, ERP, accounting systems, helpdesk, documents and spreadsheets. Increasingly, we use the MCP (Model Context Protocol) standard for this.

The agent has minimal permissions, and every action it takes is logged. We use models on business plans, with no training on client data, and for sensitive data – models in the EU or on your server. We design in line with GDPR and with the EU AI Act requirements in mind.

The cost is made up of integration maintenance and model usage fees, which depend on the number and complexity of cases. We prepare an estimate as early as the audit stage, and in a live agent we set spending limits and a cost report.