# AI and process automation

> AI and automation inside the system your team already uses. Reading images and documents, reports that assemble themselves and timely alerts.

Source: https://boosting.tech/en/services/ai-and-automation/

## Where does AI actually help in a business system?

It helps with tasks where a person spends time looking at, reading or retyping something. Classifying a photo, extracting data from a document, summarizing a long record and suggesting the next step are good examples. It helps little when the rule is simple and fixed, because ordinary code does that better and costs less.

In the [health app](/en/case-studies/health-app/) we built, a vision model classifies the photo the user sends on a clinical scale and returns a short tip. The model is instructed to decline when the image cannot be assessed, and it never gives a diagnosis.

## What can be automated without AI?

Quite a lot, and that is usually where we start. In our systems, closing a job already triggers the assembly of the PDF reports, the billing spreadsheet and the email to the client with the attachments. A change in the status of a scholarship application notifies the applicant. A technician dispatched to an emergency gets the notification on their phone.

See how this works in the [industrial tank cleaning](/en/case-studies/industrial-tank-cleaning/) case.

## How do you keep AI from making mistakes in production?

With clear limits. We define what the model may answer, in what format and when it must decline. We store responses for review, measure the cost per use and keep a person in the loop where an error has consequences. If the model fails or is unavailable, the system keeps working through the manual path.

## Frequently asked questions about AI and automation

### Do I need to switch systems to use AI?

No. In most cases AI comes in as one more function inside the system the team already uses, through an integration. If that system needs work first, see [support and modernization](/en/services/support-and-modernization/).

### Is the company's data used to train models?

We choose API services whose terms do not use submitted data to train models. Personal data is removed or masked before sending when the task does not need it.

### Which models do you use?

We choose per task. In production we have used OpenAI vision models through OpenRouter, a router that lets us switch models without rewriting the system.

### How much does it cost to run AI in production?

The cost depends on volume and on the size of each call. We measure it in a pilot with real data before opening it to all users, and the system logs consumption so the bill is not a surprise.