AI agents and integrations
Orders arrive as PDFs and have to be retyped. Delivery notes have to be checked against orders. The same summary is rebuilt every week. An AI agent does this work on the systems your company already has. It prepares the operation, a person confirms it, and every step is logged.
01 · copy the link to this sectionThe situation
The system has run the business for years. The problem is how much manual work it takes to feed it.
Orders arrive by email as PDFs, and someone retypes them line by line. Supplier delivery notes are checked against orders by hand. The summary for management is rebuilt every week by copying the same figures.
That is where an AI agent comes in. It is an assistant that goes beyond answering. It reads a document, checks the data in your systems and prepares an operation. There is no need to change your ERP.
Putting it off costs you every month. Orders go in late, overtime piles up at month end, and your most skilled people spend their time copying and pasting.
02 · copy the link to this sectionHow it works
Four steps for every operation, always in the same order.
An agent carries out operations in your company’s systems. That is why its access is designed like an external supplier’s, with written tasks, limited permissions and controls.
At first, a person confirms every operation. For low-risk operations, if the log shows the agent rarely makes mistakes, the company can decide to move to spot checks.
The path of an operation
03 · copy the link to this sectionThe connection
Usually not. We add a controlled connection and test it separately first.
The agent connects from outside, through the routes your systems already offer. Written rules set what it can read and what it can do. If an older system has no such route, we choose one with the people who maintain it.
The agent consults the databases without being able to change them or slow them down.
It creates draft orders but cannot delete them. It reads price lists but cannot change them.
Tests run in a separate environment on agreed cases. Work moves on only if the results add up.
A switch turns the agent off without losing anything. It is tested before launch.
04 · copy the link to this sectionReference
Two questions are enough. If the agent gets it wrong, can it be undone? And what does the mistake cost?
Updated 16 September 2026
| Operation | Can it be undone? | What a mistake costs | Recommended control |
|---|---|---|---|
| Routing an internal request to the right office | Yes, at once | A delay | Automatic, with spot checks |
| Reading a document and preparing a draft | Yes. A draft has no effect while it stays a draft | The time it takes to correct it | Reviewed by whoever uses it |
| Entering a customer order in the ERP | Yes, before shipping | Wrong goods, an unhappy customer | Confirmed by a person |
| Confirming a delivery date to a customer | No, once it is sent | A commitment made in the company’s name | Confirmed by a person |
| Updating the prices in a price list | Partly | Errors in quotes and invoices | Confirmation and a record of the change |
| Changing a supplier’s bank details | No | Payments to someone not entitled to them | Not one of the agent’s tasks |
An indicative example. The real classification is made for each company during the initial assessment, with the person accountable for the process. Our checks follow the list of the best-known risks of these systems, published by OWASP, a non-profit foundation dedicated to software security. These risks include an agent with more autonomy than it needs, and instructions hidden in the documents it reads.
05 · copy the link to this sectionHow we work
How long it takes, who looks after it, what the company keeps. The answers are below.
We look at the process, the systems involved and who runs them.
We record today’s volumes, exceptions and errors, so that the results can be compared.
Phases, timescales and costs, what is included and excluded, and the criteria for calling the work a success.
First in a test environment, then on real work, with a person confirming.
Strategic runs it, with support times and terms written into the contract. Or it passes to your in-house IT.
06 · copy the link to this sectionPrinciples
A good agent is not the one that does the most. It is the one whose every action, and the reason for it, you can see at any moment.
07 · copy the link to this sectionFrequently asked questions
08 · copy the link to this sectionObservatory
Analysis and guides by the Strategic editorial team.
First decide whether the process is ready to be automated, then decide what to use. Here are the criteria for choosing between an off-the-shelf product, custom software and a mixed solution, with multi-year costs and the clauses that protect your exit.
16 October 2026An organisation that is behind on artificial intelligence does not catch up by buying licences. It catches up through a few leadership decisions, taken in order over ninety days. Here is what they are, who takes them and what gets written down.
8 October 2026In a structured organisation, measuring an AI project is not left to whoever proposed it. Management accounting takes it on, with rules approved in advance, comparisons between sites, agreements on logging and indicators for the board.
2 October 2026The essential parts of this page print as a short summary of a few pages, without images or invitations to get in touch.
09 · copy the link to this sectionContact
Write a few lines on the work that keeps repeating, the systems it runs in and who handles it today. We reply by email, usually within two working days.
Please do not include confidential documents or data in the form.
Or write to us:WhatsAppstaff@strategic.sm
On WhatsApp, messages are answered by Strategic’s AI assistant.
