The three cost models
Before comparing amounts, compare types of spend. A subscription is a predictable operating cost; a custom project is an investment followed by recurring fees; in-house development shifts the cost onto your team's time.
| Model | Cost structure | Time to use | Suitable when |
|---|---|---|---|
| Subscription product | Monthly, per user or per usage credit | Short | The need matches common processes: emails, quotes, calendar, follow-ups |
| Custom project | Scoping + development, then operations | After scoping | Specific process, internal tools, strong compliance requirements |
| In-house build | Salaries, infrastructure, maintenance | Variable | AI skills already present and durable in the company |
Comparing a monthly subscription with a project budget is like comparing rent with a purchase price. Always bring all three options back to a total cost over twenty-four months, maintenance and internal time included.
What a subscription AI assistant costs
This is the only segment where prices are public and verifiable. At Swiss Agent Network the grid is as follows, excluding tax, per month:
- Starter — CHF 99: one user, 500 SAN credits per month, Sania, calendar and email services plus one service of your choice.
- Business — CHF 249: one user, 2,500 credits, adds CRM and invoicing plus one service of your choice.
- Scale — CHF 590: 8,000 credits, all services included.
Two variables move the real bill: additional users, billed on a decreasing scale from the second onwards, and credit top-ups when volume exceeds the included amount. The full grid and conditions are on the pricing page.
At this price level the question is not "is it expensive?" but "does the target process repeat often enough?". A CHF 99 subscription saving one hour a week pays for itself in almost any Swiss SME. The same subscription on a monthly process never does.
What a custom AI agent costs
We do not publish a range for custom projects, and that deserves an explanation rather than silence: a figure displayed without a scope is either a safety margin or a promise the provider will discover along the way. What we can describe precisely are the variables that determine the number.
What pushes the budget up
- The number of integrations. Each connected tool adds development, testing and a dependency to maintain.
- No documented API. A closed industry application can multiply the effort needed to reach the data.
- The quality of existing data. Incomplete client records or unstructured documents are paid for in preparation time.
- The number of rules and exceptions. The most decisive factor, and the most often underestimated at quoting time.
- Traceability requirements. Fine-grained logging, audit review and contractual reversibility are real work.
What brings it down
- A deliberately narrow scope for the first deployment — one process, not a department.
- Standard tools with existing connectors.
- An internal person who knows the process and makes time to describe it.
- Accepting systematic human validation at the start, which reduces the need for complex safeguards.
Our working sequence and per-stage deliverables are described on the custom AI agents page. Quoting happens after scoping, and the scoping itself is a deliverable you keep even if you decide not to continue.
The items actually billed
| Item | Nature | What makes it vary |
|---|---|---|
| Scoping and design | One-off | Process complexity, number of stakeholders |
| Development | One-off | Business rules, exceptions, interfaces |
| Integrations | One-off | Number of tools, API availability |
| Licences and subscription | Recurring | Users, activated services |
| Model usage costs | Recurring variable | Volume handled, task length |
| Hosting | Recurring | Location, data volume, retention requirements |
| Maintenance | Recurring | Third-party changes, rule adjustments |
| Training | One-off, then repeated | Number of users, staff turnover |
The costs nobody announces
These are what derail budgets, and none of them appear on a quote.
- Internal scoping time. Someone on your side must explain the process, settle ambiguous cases and review results. That time is real and belongs in the calculation.
- Validation time during the first months. While the agent is under observation, checking its work takes time. Net gain only appears after that phase — and an automation that never leaves it costs more than it returns.
- Data preparation. If the needed information is scattered or incomplete, cleaning it up is a project in itself.
- Variable usage costs. Double the volume, double the consumption. Always check the behaviour when limits are exceeded.
- Third-party tool changes. A supplier API update can break an integration. That is maintenance, not an exceptional incident.
- Exit cost. Recovering your data and configurations. Negotiate it at the start, while you still have leverage.
Building a defensible budget in four steps
1. Quantify the current situation. How many times a week does the process occur, and how long does it take today? Without that measured baseline — not estimated from memory — no return calculation is solid. Our ROI method explains how to record it without inventing anything.
2. Add up the full cost over 24 months. Subscription or project, plus usage, plus maintenance, plus internal scoping and validation time. That total is what compares with the current situation.
3. Set a stop criterion before starting. Which result, by which date, would trigger the decision to stop? A project without an exit criterion never stops, it stretches.
4. Start with the narrowest scope. The best cost-reduction lever is not negotiation: it is refusing to widen until the first process has proved something.
Four common sales traps
- The "turnkey" flat rate announced before any scoping. Nobody can price a process they have not seen.
- Quantified gains promised in advance. "70% time saved" without measuring your baseline is a sales argument, not a forecast.
- Usage credits with no cap or alert. Ask exactly what happens when the limit is reached: stop, automatic billing or queueing.
- No reversibility clause. If nobody wants to write down how you leave, that is the most useful information in the quote.
Frequently asked questions
How much does an AI agent cost for an SME in Switzerland?
Three models. By subscription, expect a few tens to a few hundred francs a month — with us, CHF 99, 249 or 590 excluding tax depending on the plan, credits included. Custom-built, it is a project budget quoted after scoping plus recurring operating fees. In-house, the cost shifts to skills and maintenance.
Why do providers not publish prices for custom work?
Because a price without a scope has no value. Cost depends on the number of integrations, data quality, the number of business rules and the level of validation required. A flat rate announced before seeing the process contains either a safety margin or a bad surprise to come.
What are the hidden costs of an AI agent project?
Internal scoping and testing time, data preparation, variable usage costs, maintenance driven by third-party tools, training, and above all human validation time during the first months. That last item is almost always missing from return calculations.
Is an AI agent cheaper than an administrative employee?
The comparison is misleading. An agent does not replace a role: it takes on repetitive sequences inside several roles, under validation. The right unit is cost per process handled, compared with the time that process consumes today, checking time included.
Should you pay per usage or a flat fee?
Both exist. Usage keeps cost proportional to volume but unpredictable during peaks. A flat fee protects against variation but includes a margin. In both cases, check the behaviour when limits are exceeded.
At what budget does an AI project make sense?
The threshold is measured in recurrence, not francs. A process consuming a few hours a week, repeating identically and following describable rules usually pays for itself quickly with a subscription. A rare or always-different process will not pay for itself at any price.
Do prices vary by canton or company size?
Our subscription prices are public and identical across Switzerland. For a custom project, it is scope rather than geography that moves the price. We work with companies from Geneva and across French-speaking Switzerland.