Which processes in your company are worth handing to an AI agent
Automation doesn't start with technology. It starts with picking the right process. Pick well and it pays for itself within months. Pick badly and you end up with a pilot nobody uses. Below are the criteria we score processes against during audits, and the three places where an AI agent earns its keep fastest.
- The best processes to start with are repetitive, rule-based, and have their data already in your systems.
- Don't start with the hardest process — start with one whose result is easy to measure.
- A well-chosen process can be costed with three numbers: cases, minutes, hourly rate.
How to spot a process that is ready for an agent
During an audit we check four things. If a process meets at least three of them, it can almost certainly be handed to an agent within a few weeks.
Repetition. The same steps run a dozen or several dozen times a day: read, check, copy over, reply.
Clear rules. The decision fits into a sentence: "if X, then Y". Exceptions exist, but they are rare and can be escalated to a person.
Data in systems. The information needed for the decision sits in the ERP, CRM, inbox, or documents — not only in your team's heads.
Measurable cost. You know, or can easily work out, how many minutes one case takes and how many cases come in per month.
The three processes with the fastest payback
In manufacturing, distribution, and service companies, the same three processes usually win the audit.
Handling enquiries and tickets
The agent reads the message, classifies the case, pulls data from your systems, and drafts the reply. A person approves only the unusual cases. This is normally the single biggest saving, because enquiries arrive every day and at every hour.
Copying data between systems
Orders from email go into the ERP, statuses from the ERP go to customers, invoices go to accounting. No migration, no tool swap — the agent works on the systems you already use.
Answers from company knowledge
A RAG system connected to price lists, procedures, and contracts answers your team and your customers in seconds, always with the source cited. The queue of questions for "the person who knows" disappears.
The difference between handling one quote request manually and handling it with an agent, in a model built for 900 cases per month.
What not to hand to an agent on day one
Negotiations, high-value complaints, and decisions that need context from outside your systems. Not because an agent can't help there, but because they shouldn't be the first implementation. Start with a process where a mistake is cheap and the result is easy to measure. Trust in an agent is built on the numbers from the first pilot, not on promises.
How to calculate whether it pays off
Three numbers are enough: how many minutes one case takes, how many cases you get per month, and what an hour of your team's time costs. For a process with 900 cases at 14 minutes each, at PLN 45/h, that is 210 hours and PLN 9,450 per month. If the agent takes over 77% of cases, you are left with 48 hours and PLN 2,173 — a difference of roughly PLN 87,000 a year.
We run this calculation during the free audit — on your numbers, not on market averages.
Frequently asked questions
Which process should I start my AI agent implementation with?
With a repetitive, rule-based process that has its data in your systems and a measurable cost. Most often that is handling enquiries, copying data between systems, or answering from a knowledge base.
How long does it take to implement the first AI agent?
A pilot on a single process is running 2–3 weeks after the audit. A full implementation with integrations usually takes 4–8 weeks.
How do I check whether automating a process will pay off?
Multiply the number of cases per month by the time per case and your hourly rate. At 900 cases of 14 minutes each and a rate of PLN 45/h, that is 210 hours and PLN 9,450 per month.