The question usually comes up at a familiar moment. Someone on the team spends the end of each day copying orders into a spreadsheet, quote requests sit waiting for a reply, and someone has heard that artificial intelligence could fix all of it. Then comes the obvious follow-up: what will it cost?
At Digital Swiss Agency, AI automation starts from CHF 1'200, in 2 weeks, for one automated process. Beyond that benchmark, this guide deliberately quotes no market prices. "Automating invoices" can mean a simple transfer between two programs, or a project that reaches into accounting, payment reminders and every customer's exceptions. An average figure would tell you nothing about your situation. What you can do is understand what pushes the bill up or down, compare proposals properly and pick a sensible first project.
What you are actually paying for
An automation project is made up of several items, and they don't all appear on the first line of a quote:
- Scoping: working out how the task runs today, with its rules and special cases.
- Build: connecting the tools, writing the logic and testing with real data.
- Subscriptions: automation platforms and AI services are often paid for by subscription or by usage, on top of the service itself.
- Maintenance: fixing what breaks and adapting the system as your business changes.
The first two are generally one-off costs, while the last two keep coming back. A realistic budget looks at both the setup cost and the running cost over time.
What makes the price vary
How many tools need connecting
Linking a form to an email inbox is simple. Linking your website, CRM, accounting software and calendar takes more work, because each connection adds testing and another possible point of failure. The tools themselves matter too. Recent software with an API usually connects without much trouble. An older program with no way in may force you to rely on file exports or custom development.
How much volume runs through it
Volume mainly affects recurring costs. An automation that runs a few times a week uses very little. The same one, triggered by every order or every incoming email, may push you into a higher subscription tier. AI follows the same logic: the more text you ask it to read or write, the more you consume. Ask for an estimate based on your current volume and on the volume you are aiming for.
The quality of your data
This is the factor people underestimate most. Picture an imaginary accounting firm in Vaud whose customer records exist in duplicate across several files, with addresses written in different ways. Before automating anything, someone has to clean that data and decide which version is correct. Nobody will notice that work, yet an automation plugged into messy data simply repeats the mistakes faster. Clean files directly cut setup time, and your own team can often handle much of this part.
How many exceptions there are
A rule with no exceptions is easy to automate. If every customer has special terms, or some orders need a verbal sign-off, those cases have to be described, built and tested. One simple question helps you gauge the complexity: could a new hire do this task with a single page of instructions? If so, the project is likely to be fairly light.
Maintenance
A software vendor changes its interface, a supplier alters its access, a colleague invents a new order type: any connected system needs looking after. Maintenance can be included in a plan, billed per intervention, or taken over by your team after some training. None of these is wrong in itself. What matters is knowing which one applies and who gets the alert when an automation stops.
The role you give to AI
Classic automation follows fixed rules: when a condition is met, an action fires. AI comes in when something has to understand free text, sort a request or draft a reply. Its output can contain mistakes, so anything important needs checks and human approval. That means extra work during scoping and testing. Where a simple rule does the job, it will often cost less to run.
How to compare quotes
Two proposals can only be compared if they cover the same scope. The total alone tells you little, and a modest quote may be exactly right for a well-defined need. Ask every provider the same questions:
- Which tasks are covered, exactly? Ask for a description of the outcome, from the trigger to the final action.
- What costs recur each month or year? Are licences and AI usage included, or do you pay for them yourself?
- Who owns the accounts? Make sure tool access stays in your name, so you can change provider without rebuilding everything.
- What happens when something goes wrong? Does the system alert you? Is there a manual fallback?
- What documentation will you get? A diagram and a clear note let someone else pick up the work.
- Where does personal data go? Under the nLPD, Switzerland's data protection act, you need to know which services process your customers' and employees' data, and in which countries. Ask for the list of subprocessors and, if in doubt, consult a lawyer.
- What maintenance is planned, and on what terms? Ask what is covered after delivery.
If an answer stays vague, ask for it to be written into the quote. Then line the proposals up in a table, with one row for setup and one for recurring costs. The differences become much easier to explain.
How to tell whether the spend makes sense
Rather than waiting for someone to promise a return, run your own numbers. Over a normal week, write down the time actually spent on the task. Add what errors cost you (an invoice redone, an order entered wrongly) and what delays cost you, such as an enquiry left unanswered until the next day. Compare that total with the setup cost and the recurring fees.
Some gains won't fit the sum, and that's fine: a manager freed from a chore can finally call prospects back. Note those separately rather than inventing a figure for them. If the numbers don't support the project, a simpler fix may be enough, such as an email template or a better form.
Where to start
A good first project is small, repetitive and easy to check. This approach keeps the risk low:
- List the repetitive tasks in your week and ask your team which ones weigh on them most.
- Pick just one, frequent and without serious consequences if it goes wrong.
- Write it down step by step, including the exceptions.
- Test on a small scope, keeping a person in the loop at first.
- Measure, then extend only what works.
Imagine a cleaning company in Lausanne that receives enquiries through its website, by email and on WhatsApp. A realistic first project gathers those requests in one place, alerts the right person and prepares a draft reply to review before it goes out. The owner quickly sees whether the system holds up, and the business learns what is worth automating next.
Leave for later anything that calls for judgement or carries liability: a dispute, a negotiation, personal advice. AI can prepare the file, and a person in your company still makes the decision.
How we set the price at DSA
At Digital Swiss Agency, automation and AI work starts from CHF 1'200, in 2 weeks, for one automated process (see our pricing). The factors covered in this guide show where your project stands relative to that starting point: we go through the task, your tools and the state of your data with you, then give you a firm price before we start. The quote is free, comes with no commitment and is stated in CHF. You deal with a single point of contact, supported by specialists when needed. Everything happens remotely, by video call, WhatsApp, email or phone, in French or English, for businesses in Geneva, Lausanne and across French-speaking Switzerland.
To prepare, just note the task that takes up the most time and the software involved. You can send us that through our contact page, and you'll get a reply in under 24 hours.