AI in your existing systems: one bounded job at a time
Torn Studio builds websites, products and content with AI tooling, from Sweden. Projects run in weeks.
Short answer
Yes. A language model connects to your existing systems over their APIs and does one bounded job there: reading invoices, classifying tickets, answering from your own documents. Swedish supplier Swivrr publishes the bands — a simple API integration at 50,000–80,000 kr, ERP systems like SAP or Dynamics at 150,000–250,000 kr, plus 15,000–40,000 kr per extra system.
Behind the question usually sits a worry: that AI demands new systems, data migration and a project that grows. For most businesses the opposite holds — a language model connects to the ERP, the accounting system and the document store you already run, over the APIs those systems already have. This article covers what that work looks like, what can be connected, and the bands the published Swedish market puts on it.
How a language model works inside an existing system
The shape is a bounded job: one defined task with a quality check around it. The model reads an incoming invoice and hands structured data to the accounting system. It classifies a ticket and proposes a queue. It drafts a reply from your own documents. In every case the task is narrow enough to check, and whatever gets written back to the system passes the check first — an error stops in review, well before the ledger.
What can be connected?
Most things with an API, and the bands for the work are published. Swivrr prices a simple API integration — a language model against a form or an internal chat — at 50,000–80,000 kr, LLM integration with existing systems at 30,000–100,000 kr, document automation for invoices and contracts at 80,000–130,000 kr, RAG search over your own documents with a vector database at 100,000–180,000 kr, and AI integration with ERP systems like SAP or Dynamics at 150,000–250,000 kr.
Published Swedish bands show integration cost following the weight of the system — from 30,000 kr for an LLM integration to 250,000 kr against an ERP system.
- LLM integration with existing systems30,000–100,000 kr
- Simple API integration (form, chat)50,000–80,000 kr
- Document automation (invoices, contracts)80,000–130,000 kr
- RAG search with a vector database100,000–180,000 kr
- ERP systems (SAP, Dynamics)150,000–250,000 kr
Every extra system is its own job
Cost grows per connection more than per project. Swivrr prices every additional connected system at 15,000–40,000 kr, and Denmark’s AI Enterprise charges 20,000 Danish kroner for its integration add-on against CRM or ERP. That is good news for planning: integrations can be ordered, measured and evaluated one at a time before the next is added, and the sum is known up front.
How Torn Studio delivers
The delivery shape is the same as the shape above: one bounded job per integration, a quality check around each, writes behind the check. Flows are built in your own accounts and documented for handover. The order is set by the review — two weeks, €2,200 — which measures where an integration pays back first and ends in fixed prices for the top three.
Evidence
What this article rests on — a measurement we took, a dated source, or a decision we made and what it cost.
- Decision
AI enters an existing system as one bounded job per integration — one task, one quality check around it — and writes happen behind the check.
What it cost: The boundary caps what a single engagement can promise: no AI-in-everything projects, and every further system is its own priced job where others sell one large one.
Common questions
- Do we have to replace our ERP to use AI?
- No. A language model connects to the systems you already run over their APIs — SAP, Dynamics, accounting systems and document stores. Swivrr publishes bands for exactly that work, and the whole point of the integration shape is that your system landscape stays put.
- Can AI read our documents and answer from them?
- Yes — the shape is called RAG: the documents are indexed in a vector database and the model answers with references to the sources. The published Swedish market prices the build at 100,000–180,000 kr per Swivrr. What it asks of you is that the documents are gathered and machine-readable.
- Is AI allowed to write directly into our systems, or only read?
- Both exist, and the order is a per-flow choice. Reading jobs — classification, drafts, search — are the risk-free place to start. Writing jobs are built behind a quality check, so whatever reaches the ERP has passed a review, human or automated, before it is written.
- How long does an AI integration take?
- The published Swedish market quotes two to four weeks for a system integration with API sync. A simpler reading integration is faster, an integration against a heavy ERP takes longer — and every extra system is its own job with its own timeline.
- What is a bounded job, concretely?
- A task you can describe in one sentence and check against an answer key: read this invoice and hand over structured data, classify this ticket, draft a reply from these documents. The narrowness is what makes quality measurable and errors visible before they cost.
- Does this work with older systems that have no API?
- Often, via a detour: many older systems can export files or receive them, and a flow can be built around that channel. Where the line sits is settled in the mapping — if no machine-readable way in or out exists, the review says so, up front and in writing.
Sources
- Vad kostar AI-integration — Swivrr
Evidence for the bands 50,000–80,000, 80,000–130,000, 100,000–180,000 and 150,000–250,000 kr, and the 15,000–40,000 kr per extra system.
- Vad kostar AI-projekt — Swivrr
Evidence for the 30,000–100,000 kr band for LLM integration with existing systems.
- Priser — AI Enterprise
Evidence for the 20,000 Danish kroner integration add-on against CRM or ERP.
- Vad kostar automation — Snabba Pucks
Evidence for the two-to-four-week timeline for a system integration with API sync.
Next steps
Read next
- Which processes to automate first?The method that decides which processes get automated first: measure hours, frequency and error cost per process, rank the list, and leave the bottom alone.
- The most common automations to hire an agency for in 2026Search data from ten markets and official statistics from nine countries show which automations companies seek help with in 2026 — and where agencies come in.
- Automating in-house or hiring: where the line sitsWhat you can build yourselves in Zapier, Make or n8n, what the platforms cost in 2026, and where the line to hiring an agency sits in practice.