AI automation services connect the tools you already use so a repeatable task happens without someone retyping it, and add an AI step only where judgement over unstructured text is genuinely needed. The price depends far more on the number of systems, exceptions and handoffs than on the AI.
What is usually included
- Mapping the current process and agreeing the workflow on paper.
- Building the workflows in a tool such as n8n, Zapier or Make, or in custom code.
- Integrations with your CRM, forms, calendar, WhatsApp, email and spreadsheets.
- An AI step where it fits (classifying, extracting, drafting or summarising) with a human review path.
- Testing, error handling and alerts for when a step fails.
- Documentation and a handover, so the system is yours.
- Monitoring and support, which is often optional and should be priced separately.
What drives the cost
We do not quote numbers here. Published price ranges for AI automation differ widely and rarely state their assumptions. The drivers below are what actually move a quote, and each one is a question you can ask.
| Driver | Why it changes the price | What to ask |
|---|---|---|
| Systems connected | Every integration is built, tested and maintained | List every tool the workflow touches |
| Exceptions | Cases that do not follow the standard path usually take more effort than the main path | What are the three most common exceptions? |
| Data quality | Messy or duplicated data has to be cleaned before it can be automated | Where does the data live, and who owns it? |
| Human steps | Approvals and handovers need interfaces and rules | Which steps need a person, and who? |
| AI model use | Per-call model cost, plus the effort of reviewing wrong answers | How many items a month, and what happens when the model is wrong? |
| Compliance and consent | Messaging and personal-data rules differ by country | Which countries are your customers in? |
| Ownership and hosting | Your accounts versus the agency’s; self-hosted versus SaaS | Whose account does it run in, and can you take it away? |
| Support and change | Workflows break when a connected tool changes | Who fixes it, how fast, and what is included? |
A scoping worksheet
Copy these into a document and answer each in a sentence before you talk to anyone. If you cannot answer one, that is the first thing to find out.
| Question to answer | Why it changes the scope |
|---|---|
| What triggers the workflow? | The trigger decides which system has to send the first signal |
| What is the outcome, in one sentence? | Without it nobody can say the project is done |
| Which systems are involved? | Each one is an integration to build and maintain |
| Roughly how many items a month? | Volume affects tooling, model cost and monitoring |
| What are the three most common exceptions? | They usually decide the build effort |
| Where must a person approve? | Defines the human-in-the-loop design |
| What data is sensitive? | Drives hosting, access and retention rules |
| Which countries are your customers in? | Sets the messaging and consent rules |
| Who owns the accounts and the data? | Decides whether you can leave without losing the system |
| What number will show it worked? | Pick one you already track, so the result is checkable |
Red flags in an automation quote
- No written scope, only a price.
- A fixed price with no list of exceptions.
- An AI step with no human review path.
- The agency owns the accounts the system runs in.
- No mention of testing, error handling or monitoring.
- A result promised before anyone has seen your process.
- The word “unlimited” attached to anything.
What we do and do not claim
Provixon builds automation: n8n workflows, CRM and document automation, API integrations and WhatsApp flows. We do not publish an automation result yet, so this guide is method, not a track record. What we publish are documented ad results on our case studies page. If you would like to talk through your own worksheet, see our AI automation services page.