AI agents · workflow automation · n8n, Make, code
An AI agent is a workflow that reads, decides within clear rules and acts in your tools: it sorts the inbox, updates the CRM, prepares a report or a document. We build them in n8n, Make or custom code, with a person approving every step that matters.
Price Custom quote, after scoping the process
01 · What we automate
Incoming inquiries and emails classified, summarised, routed to the right person, with a reply draft.
New leads, call notes and email threads written into HubSpot, Pipedrive or Salesforce without retyping.
Weekly sales, support or ops summaries pulled from several systems and written in plain language.
Invoices, orders and RFQs extracted and checked. See AI document processing.
Answers to common questions from your knowledge base. See AI customer service agents.
Leave, purchase and IT requests collected in chat, validated and passed to the right system.
02 · Tools
We pick the tool for the process, not the other way round.
Most projects combine them: n8n or Make for orchestration, an AI model (OpenAI, Anthropic or Azure OpenAI) for reading and writing, and your systems' APIs.
03 · How we build an agent
One process, written down step by step: inputs, decisions, outputs, exceptions and who approves what.
A working version on real (or anonymised) examples, tested against how your team handles the same cases today.
Approval steps where a mistake is costly: sending to clients, changing records, money. The agent proposes, a person confirms.
Every run logged, errors alerted, quality spot-checked. Rules are adjusted as edge cases appear.
04 · Example processes
Illustrative example
B2B company, shared sales inbox
Illustrative example
Service company, several systems
Scenarios show typical designs, not results of a specific client.
05 · Cost of ownership
Running an agent has three parts: the automation platform (a plan or a server for n8n), AI model usage billed per amount of text processed, and maintenance when your systems or processes change.
AI usage scales with volume: a few hundred emails a month costs little, thousands of long documents a day is a budget line. We estimate this per process before building, and log usage so there are no surprises.
FAQ
A classic automation follows fixed rules. An AI agent adds steps that read unstructured input, such as emails or PDFs, and choose between options. In practice we combine both, with fixed rules wherever possible.
Yes. That is why we add approval steps for anything that goes to clients or changes money and records, log every run and test on real examples before launch.
Both, plus Zapier and custom code. n8n is our default when data is sensitive or flows are complex, because it can be self-hosted in the EU.
A focused workflow usually takes 2–4 weeks from scope to pilot. Agents that touch several systems or need approvals across teams take longer.
Either your team, after a handover and training, or us under a support agreement. Either way you get documentation and access to every workflow.
A free call to find the workflow with the best payoff and the lowest risk.
Book a discovery callPrice: Custom quote, after scoping the process
Keep reading
Next step
A short form. We reply within one business day: whether we can help, and what the first step would be.
Price Custom quote, after scoping the process
Enquiry sent. Reference .
We reply within one business day. Anything to add? Write to hello@kreal51.com.
Enquiry prepared. Reference .
Press send in your e-mail app and we reply within one business day. If nothing opened, write to hello@kreal51.com.