technologycircle
DEIntro call

AI improvement · Claude · MCP

Make existing processes better with AI. Step by step, with control.

Not the big AI leap, but the next concrete step: we find the places in your workflows where Claude already helps reliably today, build them in with approval and guardrails, and measure what it delivers.

Free intro call What we offer

The problem

The demo was impressive. None of it made it to production.

Many companies have tried AI and stopped there: the pilot depended on one person, nobody trusted the results, the data was not allowed out, the costs were unclear. We build AI in so it runs in daily business.

  • Pilots without production

    The prototype ran in a chat window. It was never connected to ERP, CRM and the inbox. We build into the real process from day one.

  • Full automation nobody trusts

    AI does not decide about money and customers. It prepares, proposes and explains. Approval takes seconds, not minutes.

  • Data that must not leave

    Contracts, HR data, account transactions. We send only what is needed, to providers with EU processing, and log every call.

  • Costs nobody sees

    Tokens, model choice, retries: without a budget and monitoring, AI gets expensive without anyone noticing. We build cost control in.

Services

From the question of where AI pays off to the agent that runs in daily business.

01

AI process audit

We walk through your workflows and name the places where Claude helps reliably today: classifying, extracting, drafting, summarising. And the places where it should not.

  • Audit
  • Prioritisation
  • 1 week
02

Claude integration with internal systems

Claude reads from ERP, CRM, ticket system and document storage and writes back. Via n8n workflows or directly via API, with permissions that match the user.

  • Claude
  • n8n
  • API
03

AI agents with guardrails

Check incoming invoices, answer requests, create tickets. With confidence thresholds, an approval step, a budget and a log that makes every decision traceable.

  • Agents
  • Human-in-the-loop
  • Guardrails
04

Knowledge assistant on company data

Claude answers questions about contracts, customers and processes, with access only to what each person is allowed to see. With sources instead of hallucinations.

  • RAG
  • Permissions
  • Sources
05

MCP & loopthink.ai

Expose internal tools, APIs and data to AI agents in a controlled way, with workspaces, permissions, SSO and audit log. Our own platform for this is loopthink.ai.

  • MCP
  • loopthink.ai
  • SSO
06

Enablement & governance

Which data may go where, who approves, how it is measured. Rules and training so AI grows in your company without losing control.

  • Governance
  • Workshops
  • GDPR

Process

One process. Measurably better. Then the next.

We do not start with an AI strategy, but with a process that hurts today. After two weeks you know what AI delivers there, in hours and in euros.

  1. 1

    Audit

    We walk through the workflows with the business unit and pick the process with the best ratio of benefit to risk.

    1 week · fixed price

  2. 2

    Pilot

    The process runs in production with Claude, an approval step and monitoring. Before and after are measured.

    2 weeks · fixed price

  3. 3

    Measure & expand

    Hit rate, time saved, cost per case. What works gets expanded. What does not gets stopped.

    Iterative · sprint rhythm

  4. 4

    Handover

    Your team operates and extends it themselves. With governance rules, training and support as needed.

    Enablement · optional retainer

Typical pilot

Incoming invoices through to approval, with Claude instead of typing

The invoice arrives by email, Claude reads it, checks it against purchase order and delivery note and presents it for approval with reasoning. Unclear cases land in a review list instead of the system. Accounting clicks instead of typing.

  • Claude
  • n8n
  • Approval
  • ERP
  • Cost control

From the blog

Hands-on knowledge about n8n, data automation and AI in the enterprise. A new article every two days.

The first articles are coming soon.

All articles

FAQ

Frequently asked questions

Which processes are suitable for AI?

Anything where people read, classify, type in or write drafts today: invoices, requests, contracts, reports, tickets. Less suitable are high-risk decisions without review. There AI prepares, but does not decide.

Does this replace our employees?

In our experience, no. It takes away the work nobody likes doing and leaves them the decision. Approval stays with the human, just faster.

What about GDPR and data protection?

We use providers with EU processing and a data processing agreement, send only the fields the step needs, and log every call. Sensitive processes run self-hosted on your infrastructure.

Why Claude?

Because in our projects it is the most reliable with documents, classification and structured output, and connects cleanly to internal systems via MCP. If another model fits better, we use that.

What if it does not deliver?

Then you know after two weeks and a fixed price, not after a year. We measure before and after. A pilot that does not deliver gets stopped, not talked up.

Which workflow should get better with AI first?

30 minutes, no pitch. We look together at where Claude pays off, what a pilot costs and how fast it can run.

Prefer writing?

Send an email

Book an intro call · 30 minutes

Calendar not showing? Open it directly: calendar.app.google