Most AI-assisted workflows for SMEs cost somewhere between a few hundred and a few thousand baht per month to run. The variance comes from how often the workflow runs, how much AI it uses, and which tools it connects.
The build cost of an automation is easy to quote. It is fixed up front and visible on the invoice. The running cost, the part you pay every month after the build is live, is harder to estimate, and it is the part most vendors gloss over until the first surprise invoice lands.
This is what those numbers actually look like for the kinds of builds we deliver to small and medium businesses in Thailand.
The three cost lines
A typical AI-assisted workflow has three monthly cost lines:
- The LLM API itself. Per-token costs from whichever model you use. This is usually the most variable line, because it scales with how much your system thinks.
- Your orchestration tooling. The platform that runs the workflow on a schedule, watches for events, and connects pieces together. n8n, Zapier, Make, or self-hosted alternatives.
- Anything else. Database hosting, email-sending limits, file storage, occasional API costs from the tools the workflow connects to.
For most SME builds, the first two dominate. The third tends to be small unless the workflow is moving large volumes of data.
Realistic ranges
For the kinds of workflows we typically build, here is what the monthly running cost looks like, all in:
- A small lead-qualification pipeline classifying around five hundred enquiries per month, drafting personalized replies: roughly 600 to 1,200 baht per month.
- A guest-segment email nurture flow shipping a few thousand emails per week, with AI-drafted personalization per segment: 1,500 to 3,500 baht per month, mostly driven by sending volume more than AI cost.
- A content production pipeline producing forty long-form pieces per month with AI-drafted first passes and fact-check passes: 2,500 to 5,000 baht per month, dominated by the heavier model usage on the drafting step.
- A simple reminder automation with no language generation, just notifications and routing: under 500 baht per month, sometimes effectively free.
The lower end of these ranges is what you will pay if the build is tuned with care. The upper end is what you will pay if it is not.
Where the variance lives
Three things drive variance more than anything else.
How often the workflow runs. This sounds obvious but is often miscounted. A workflow that runs every minute of every day racks up costs differently than one that runs in batches. Batching, where it is possible, is almost always cheaper.
Which model you use. The frontier models cost roughly ten times what the smaller fast models cost, sometimes more. For most workflows that involve classification, simple drafting, or extraction, a smaller model performs the job equivalently. For heavy reasoning steps, the frontier models earn their cost. A well-designed workflow uses each model where it makes sense.
How much memory the workflow carries. Long context windows are expensive. A workflow that loads a 50-page document into context with every call will cost dramatically more than one that uses retrieval to pull in only the relevant paragraph. Retrieval design matters more than people expect.
How to estimate before you commit
Before signing on a build, ask the vendor for three numbers:
- What is the expected monthly volume of LLM calls?
- Which model will each step use, and why?
- What does the projected monthly cost look like at typical volume, and at twice that volume?
If they cannot answer those, they have not designed the system carefully. If they can, the answers will usually fall inside the ranges above.
What this is not
The numbers above cover running cost only. They do not include the build cost, which is fixed and quoted up front. They also do not include any underlying SaaS subscriptions you may already be paying for and folding into the workflow, which is your existing cost rather than a new one.
In practice, most clients we work with see their total monthly running cost land in the low thousands of baht, returning savings of tens or hundreds of thousands of baht per month in reclaimed team time. The math, when you sit down and do it honestly, is rarely close.