ServicesAI automation
AI automation services for the routine steps in your processes
We take routine steps out of your processes with AI and leave exceptions and decisions to your people. We map the process, measure what a case costs today, then give each step to the cheapest tool that handles it reliably, from a simple rule or an n8n workflow to a language model or an agent. A person approves where errors are expensive, and every change is measured on your own past cases. It suits order processing, finance and service teams still working from inboxes and PDFs.
How we automate a process
- 1
We map the process and measure what each case costs today.
- 2
Each step is assigned to a rule, a workflow, a model or a person.
- 3
The automation is tested on past cases with known outcomes.
- 4
It goes live with review steps, logs and a reported cost per case.
What changes for your business
AI automation pays off where people spend hours on the same steps: reading documents, typing data into the ERP system, sorting requests, checking one system against another. Taking those steps over means fewer hours per case, shorter waits for customers and suppliers, fewer typing errors, and a team that handles more cases without new hires. Computerwoche reported in 2023 that the food manufacturer Zentis, using a vendor's tool, was processing about 80% of incoming order documents automatically into SAP.
- Fewer hours per case, measured before and after on your own cases
- Faster turnaround for orders, invoices, claims and requests
- Fewer errors from retyping, with exceptions sent to a person together with the reason
- More cases handled by the same team, which matters when positions are hard to fill
- A known cost per handled case that you can set against today's process
Which processes to automate first
Start with the process that has the most routine hours, a clear correct outcome and data a system can reach. We score candidates on volume, how much the inputs vary, whether the systems have interfaces, how much review a person has to do, and whether past cases with known outcomes exist to test against.
More detail
Poor first candidates are work that happens only a few times a month, steps whose correct outcome nobody can state, and processes that are about to change anyway. Typical good candidates look like this:
- Incoming documents such as invoices, claims, orders and contracts, read and checked against your systems
- Customer and IT requests sorted, enriched and routed to the right team
- Back-office checks such as invoices against orders, or onboarding documents against requirements
- Reports and summaries assembled from several systems
- Transfers between systems that today depend on copy and paste
Rule, workflow tool, model or agent: who does which step
A workflow tool is often enough, and when it is, we say so. A model earns its cost where inputs vary, such as free-text emails or documents in many layouts. An agent earns its cost only where a case needs several steps across systems and some judgement.
More detail
Classic process automation follows fixed rules: if an invoice matches its order, approve it. It breaks when inputs vary, documents arrive in different formats or a step needs judgement. AI automation adds models that read documents, classify requests and draft text, and agents that carry a case through several systems. Many processes need a mix of these, plus people:
- Rules and workflow tools such as n8n or Power Automate for steps that are the same every time, such as collecting attachments or moving data between systems
- Language models for reading documents, extracting fields and classifying requests
- Agents for cases that pass through several systems and need some judgement: an agent is a program that carries out steps in your systems and presents the result for approval
- People for decisions that the law, your policy or the risk reserve for them
Would you like to talk this through for your own case? The first call is free and takes 30 minutes.
Book a 30-minute callWhere a person approves
Automation does not mean nobody looks. Any step that writes to a business system, sends something to a customer or commits money can require a person's approval, and keeps it until the test results justify removing it. Cases where the model is unsure, or where values do not match, go to a person with the reason shown.
More detail
Some decisions stay with people for good. Under the AI Act, uses such as screening and ranking job applicants or credit scoring are high-risk: the company deploying such a system must provide human oversight, keep logs and inform the people affected. We do not automate such decisions without people.
- Approval before any write to ERP, CRM or DMS until the tests justify removing it
- Thresholds per step, set by what an error costs
- A review queue that shows the reason, the source document and the proposed action
- The share of cases passed to people, reported alongside cost per case
A hypothetical example
Take an accounts payable team that receives supplier invoices as PDFs by email. Today each invoice is opened, typed into the ERP system and matched against its purchase order. Mapped out, the process has five steps. Collecting the attachments and posting matched invoices are rules. Reading the invoice fields goes to a model, tested on several hundred past invoices. Matching against the purchase order is a rule again, with tolerances agreed with finance.
More detail
Invoices that don't match, or where the model is unsure, go to a person with the reason shown. The figures that matter to finance are the cost and error rate per invoice, before and after.
Our use case pages show the same approach for document processing, finance and accounting, customer service and the IT service desk.
How we build and run it
We build in your environment and connect to your systems through their APIs, the integration tools you already use or MCP servers. Models run through providers your data owners have approved, or as open-weight models on your own infrastructure. Every automated step writes a log, and every write to a business system can require approval until the test results justify removing it.
More detail
- A process map showing which steps a rule, a model, an agent or a person handles
- A test set of past cases with known outcomes, run again before every change
- Monitoring of cost per case, error rate and the share of cases passed to people
- Runbooks, and an owner on your side for each automated process
AI automation agency or engineering partner?
Some AI automation agencies build workflows in a no-code tool and hand them over, which suits simple, low-risk automations. When a process touches finance, customer data or several core systems, it needs tests on past cases, access limited per step, logs an auditor can follow and a team that can change the automation when the process changes. That is the work we take on. Where a simple workflow tool is enough, we say so.
Cost per handled case, and how we price the work
Cost per handled case decides whether an automation pays off. It adds up model and infrastructure cost, the time people spend checking and correcting, and upkeep such as re-tests after a model changes, then divides the total by the cases handled. We measure today's figure before we build and report the new one continuously in production, so both are calculated the same way.
More detail
Each stage of our work has a fixed price, agreed in writing after the free first call. The price depends on how many steps and systems a process touches, how varied the incoming documents are, how strict the accuracy requirement is, and whether models run with a provider or on your own infrastructure. Once the automation is running, model and infrastructure costs are counted in the cost per case, so you see them next to everything else.
Questions
Which processes should we automate first?
The one with the most routine hours, a clear correct outcome and data a system can reach, often incoming documents or requests that are typed into another system. Rare tasks, steps nobody can define a right answer for, and processes about to change make poor first candidates.
Does automating a process with AI pay off?
It does when the cost per handled case, including review, correction and upkeep, ends up below today's cost per case. We measure today's cost before we build and test the automation on your past cases, so you see the comparison before production.
Will automation replace jobs in our team?
It changes the work first: fewer routine steps, more exceptions and decisions. Whether that means handling more cases with the same team, taking on new tasks or hiring less is your decision. We give you the hours per case so you can decide on figures. Where you have a works council, we document what the system logs and what it does not.
Is a workflow tool like n8n or Power Automate enough?
Often, for steps that are the same every time. It reaches its limits where inputs vary, where a step needs reading or judgement, or where you need tests on past cases and logs an auditor can follow. Then we add a model or an agent at those steps, inside the tool you already run where that works.
What is the difference between AI automation and AI agents?
AI automation is the whole job of taking routine work off people, using whatever tool fits each step. Agents are one of those tools, for cases that need several steps and some judgement. Our page on AI agents covers that part in detail.
Can you build on the workflow or RPA tools we already use?
Yes, where they suit the process. We start from the tools you run today and add models or agents at the steps where those tools reach their limits.
How accurate must an automation be before it goes live?
You set the threshold with us, step by step, based on what an error costs. Until the test results meet it, the step keeps a human review. After launch we report how often the automation was wrong, not only how it scored before.
Can the automation run on our own servers?
Yes. Workflows, agents and open-weight models can run on your infrastructure. Commercial models are used only through data flows your data owners have approved.
How does an automation project start?
With a free 30-minute call about one process, for which we need no access to your systems. If the process is a good fit, we propose the first stage in writing at a fixed price: mapping the process and measuring what a case costs today. Building the automation comes after that, priced separately once the figures are in.
What do we have at the end of an automation project?
An automation running in your process, together with the process map it is built on and the test set of your past cases it was checked against. Monitoring reports cost per case and how often the automation was wrong, runbooks tell your team how to handle it day to day, and a named owner on your side is responsible for it after launch.
Related services
AI agent evals
Find out how often an agent gets the job done before it acts in your systems, by checking outcomes, tool calls and how reliably runs repeat.
AI chatbots
Customers and staff answered faster from your data, with approved actions and a handover to people when it matters.
AI strategy and rollout
AI strategy and rollout that starts with one running, measured system on your own processes and builds the roadmap around its figures.
Data integration for AI
Answers from file shares, SharePoint, ERP and tickets in seconds, for assistants and agents, with permissions carried through and tested.
AI engineers for your team
Extra AI engineering capacity when your team lacks AI skills: experienced engineers who build with you and leave the know-how.
AI evals engineering
Evals built from your own traces show whether each change made your AI product better or worse, with a release gate that stays in your repository.
AI system testing
Before go-live, acceptance tests on your own cases tell you how often your chatbot, assistant or agent is right, in terms your QA team already uses.
Free AI usage review
One page for management on where your AI money goes, whether it pays off and what to change, from exports you choose to share. Free, no system access.
Enterprise AI
One approved AI platform for the whole company: an internal assistant instead of shadow AI, approved models, data that stays in-house and spend per team.
LiteLLM gateway
Budgets and access rules for every team's AI use, through a LiteLLM gateway that is hardened, connected to your identity provider and kept up to date.
AI platform for all teams
One approved route to AI for every team, with budgets, logs and a free choice of model, run on your servers or in your cloud.
System access for agents (MCP)
AI agents that look up and update data in your systems, with narrow permissions, human approval for changes and tests.
On-premise AI
AI on offers, contracts and customer files, with the data staying in your company: own servers or own cloud account, tested before go-live.
Azure OpenAI and Bedrock
GPT and Claude on company data, under the cloud contracts you already have: Azure OpenAI and Amazon Bedrock in your own account, with EU processing and one gateway.
Knowledge assistants (RAG)
Staff find answers in internal documents in seconds, with sources cited, permissions kept and tests on your own questions.
RAG evaluation
An assistant that answers from your documents, tested on retrieval and on answers in each language your users write in, so you know how often it is right and cites the right source.
vLLM on Kubernetes
Open-weight models on your own GPUs, so data stays in house and costs follow capacity you plan, sized from measured load.
AI engineering and architecture
Assistants, voice interfaces and agents in the products you sell, with the serving and retrieval underneath.
Custom model training
A small open-weight model trained for one task and compared with your current model on cost and accuracy.
From the blog
AI business case
How to work out whether AI pays off for one process: measure today's cost per case, count every cost of the AI system including the people who check it, and agree in advance when to stop. With a worksheet for your own figures.
Choosing the first AI workflow
How to choose the first process to automate with AI by scoring every team's proposal on volume, variation, system access, review effort, cost of a mistake and known outcomes.
Change management for AI
How to introduce AI without losing your staff: an honest answer on jobs, the works council from the start, shadow mode with human approval and training on real work.
Human approval in agent workflows
Which agent actions on business records need a person's approval, what the approver must see and how to keep approvals from stalling every case.
Talk to an engineer
Bring one process that still runs by hand and costs your team hours. In a free 30-minute call an engineer tells you whether it makes a good first candidate and what the first stage would involve. We need no access to your systems or data.
Booking runs on Google Calendar.
When you book a call, you speak to Mickey Graf
He founded Cloudsail and has put machine learning into production since 2017, including AI engineering at trivago and SAP. The call can be in English, German or Polish.
Prefer to write first? Leave your email and a few lines. Mickey reads every message himself and replies by email with times for a call.
Look for the draft in your email program
Send it from there, and Mickey replies in person with times for a 30-minute call. If no draft has opened, write to us at miki@cloudsail.com.