ZrobotIT ExpertsPLContact
Case studies

Case studies: examples of AI and automation projects

No client names and none of their data. We show how the work looked before, what we did, how we protected the data and what changed for the team.

case-studies.log
cat ./case-studies/*.case

Three examples from recent projects

Each description shows the situation before, what we did, the safeguards and the result.

dane-dla-agenta-ai.caseintegration · AI agent

Accounting data for an AI agent and reports

A services company with its own data analytics team.

BeforeThe team was building an AI agent and reports on data from an accounting system running on a local SQL server. The data had to be sent regularly to an external server, without exposing the database to the internet.

What we did

  1. We created separate views with only the data needed: accounting entries and reference lists, no employee data.
  2. We set up a read-only account that can see only those views.
  3. We launched an automatic transfer: a full load every night and on demand every few minutes, notifying the agent's system after every run.

Security

Connection encrypted with an SSH tunnel and TLS with server certificate verification. A separate service account on the server, a key with no shell access, the database port not reachable from the internet, and the machine covered by backups.

ResultThe AI agent and the reports work on current data, the team refreshes it on demand, and the production database is not exposed.
raport-wyniku-na-dzialy.casereporting

Profit by department report from the sales and accounting system

A trading company, its board and department heads.

BeforeProfit by department was built in spreadsheets put together by hand from exports out of the sales and accounting system. The summary came late and every time someone had to check whether the numbers matched.

What we did

  1. Data from the system's database is sent automatically, one way, to an external reporting server.
  2. We built a profit by department report that refreshes itself.
  3. We agreed the department and metric definitions with the board, so everyone looks at the same numbers.

Security

The database is read through a read-only account, and the transfer is encrypted and one way only. The source system receives no changes from outside.

ResultThe board sees current departmental results without building spreadsheets by hand.
faktury-z-poczty.caseautomation

Invoices from email straight into the accounting system

Companies whose finance team receives cost invoices by email.

BeforeInvoices arrived in the mailbox as PDFs, and the data from each one had to be retyped into the finance and accounting system.

What we did

  1. Invoices go to a dedicated mailbox and are read automatically.
  2. We save the data as an XML file that the finance and accounting system imports on its own.
  3. The document waits in the system for the finance team to check and post it.

Security

A separate mailbox for invoices only; the automation can access only that mailbox and the document import. Posting and payments stay with people.

ResultThe finance team no longer retypes invoice data, it checks and approves ready documents.

These are examples. We have built many similar integrations and automations on accounting, sales and ERP systems: reading data for reports, importing documents, reminders and summaries.

→ how the AI audit works   → agents in Copilot Studio

faq --list

Frequently asked questions

Short and to the point.

Why don't you name your clients?

Because we work with companies' financial and HR data. We describe what we did and how, and we can go into detail in a meeting.

Is every project an AI agent?

No. Some are ordinary automations and integrations without AI. We propose an agent where content has to be understood, and a simpler flow where fixed rules are enough.

Can you build something similar for us?

Yes. We start with an audit of the process, the data and the permissions, and then propose the scope and how to secure it.

Don't see your process?

Show us a task that eats up your team's time. We will look at the workflow, the data and the systems. We will propose an agent, or tell you plainly that a simpler automation is enough.

Discuss your process →   How the AI audit works
contact --now

Let's talk about your process

zrobot · IT Experts
Phone+48 506 506 350
Emailoffice@itexperts.com.pl

zrobot.pl is a brand of IT Experts sp. z o.o. · ul. Wolfkego 11/13, 01-494 Warsaw, Poland · in Warsaw and remotely across Poland