Automation · Switzerland
Business process automation.
I automate the processes of companies that pay people to read, check, re-key or chase: case files arriving by email, documents to verify, rules applied by hand, double entry between two tools. The result is a workflow system — structured intake, a versioned rules engine, states and follow-ups, human review where it is due, an audit log — built fixed fee, usually CHF 20’000–50’000.
Updated
In brief
- For
- SMEs, fiduciaries, practices and operations teams that process case files
- Delivered
- Structured intake · versioned rules · states and follow-ups · human review · audit log · integrations
- Range
- Usually CHF 20’000–50’000, fixed fee; minimum CHF 20’000
- Typical timeline
- 4–8 weeks after scoping
- AI
- Only where it beats a deterministic rule, always reviewable — no product in production depends on it today
- Built and run
- Atamatax · Horvera · Vellaci · Neucelle (in production)
Automating does not mean handing everything to an AI model. Most of the gain comes from less spectacular things: a form that structures what used to arrive in bulk, a document checklist that keeps itself up to date, a rule written once instead of checked a hundred times, a reminder that goes out on the right day, a record of who approved what.
None of my products in production depends on a language model today. What they prove is the workflows: on Atamatax, a case workspace with a personalised document checklist, encrypted uploads, a review status per document, follow-ups and a check by a person before every delivery. AI is added where it beats a rule — extraction, classification — and stays reviewable; see AI developer.
01 — Use cases
What automates well
Repetitive tasks where errors cost money, where a rule exists, and where someone loses hours every week today.
- Case files arriving by email
- A form and an upload space replace the inbox: every case has a state, an owner, a document checklist and a history.
- Documents to verify
- The checklist is computed from the client’s situation; each document has a status — received, accepted, to replace — and a reminder goes out when it is missing.
- Rules applied by hand
- Eligibility, pricing, thresholds, deadlines: the rule is written once, versioned, tested against your real cases, and every decision keeps the version that produced it.
- Double entry
- What is copied from one tool to another goes through an integration — API, export, scheduled import — with a log and a retry on failure.
- Deadlines and follow-ups
- Deadlines live in the system, not in a calendar: dated reminders, alerts before the due date, a list of what is late.
- Checks and approval
- A review queue where a person approves, corrects or rejects. Nothing is marked “checked” that was not.
- Exports for the fiduciary or a third party
- A structured export the recipient picks up as is — as on Neucelle or ComptaNexo.
02 — Method
How I automate a process
I start from the real process, not the org chart: one case followed end to end, with the exceptions, the back-and-forth and what happens when a document is missing. The states, transitions and owners come out of that.
Rules become pure modules, tested against a set of real cases you give me, versioned so a past decision stays explainable after a change. Integrations are idempotent and logged: re-running a sync does not create duplicates.
AI only comes in after that structure, on a measurable task: reading a document to pre-fill fields, classifying a request. Its output is validated against a schema, its confidence is shown, and below a threshold the case goes to human review. If a rule does as well, the rule wins.
All of it ships as a custom application: accounts and roles, data in Switzerland or Europe as required, code and accounts in your name.
03 — Limits
What I do not automate
- A costly decision with no possible review
- If nobody can review it and an error is expensive, you do not automate the decision — you automate its preparation.
- A process that changes every month
- Until it is stable, a no-code tool or a well-kept spreadsheet costs less; custom comes later.
- What off-the-shelf software already does
- Standard invoicing, CRM, HR: configure what exists before building.
- Industrial automation
- Programmable controllers, SCADA, production lines: a different trade.
04 — Cost
What an automation system costs
Usually CHF 20’000–50’000, fixed fee, over 4–8 weeks. What moves the price: the number of states and rules, the number of integrations, the number of roles, the volume of documents. A call to an AI model adds a running cost per document, priced and capped before work starts.
The number that matters is the time given back: hours of reading, checking and re-keying per month, multiplied by their real cost. That number, not the technology, decides whether the project is worth it — it is set during scoping.
05 — Hire?
Should you hire instead?
Hiring to automate makes sense when automation is continuous: new processes every quarter, integrations to maintain, a team that depends on them daily. For a defined process, a project delivers the system and leaves you to decide later whether someone in-house should evolve it.
The hire-or-outsource guide compares both with the real employer cost in Switzerland.
Is this the right service?
This makes sense if
- People spend hours every week reading, checking, re-keying or chasing.
- Your case files follow rules someone checks by hand.
- Clients or partners need to upload documents and follow progress.
- You want a record of who decided what, with which rule.
It is not the right service if
- You want a chatbot dropped on your documents.
- The process is not stable yet.
- Off-the-shelf software already covers most of it.
- It is industrial automation (controllers, SCADA).
Proof — products I built and run
My own products, with their actual state. No client project is published yet.
- atamatax.comIn production · payments liveAtamatax — US expat tax infrastructureA case workspace with a personalised document checklist, encrypted uploads, review status per document, follow-ups, and a check by a person before every delivery.Next.jsTypeScriptPostgresStripeVercelRead the case study →
- vellaci.chIn production · subscriptions liveVellaci — EU Cyber Resilience Act compliance for product manufacturersA regulated workflow with 24- and 72-hour legal deadlines, immutable revisions and an exportable evidence vault.Next.jsTypeScriptPostgresStripeCycloneDX · SPDXOSV · KEV · EPSSRead the case study →
- horvera.comIn production · payments liveHorvera — flight compensation infrastructureA versioned rules engine that delivers every decision with its trace, then carries a claim and escalation workflow.Next.jsTypeScriptPostgresStripePostHogRead the case study →
- neucelle.chIn production · subscriptions live · FR · DE · ITNeucelle — cash expenses and receipts for Swiss restaurantsA ledger shared between a restaurant and its fiduciary: a receipt per expense, VAT per line, a cross-client review queue, batch validation, export.Next.jsTypeScriptPostgresStripeVercelRead the case study →
How it works
Four steps, fixed scope, price agreed upfront.
- 01
Diagnose
20 minutes, free. Business model, workflows, risks and product scope. If I’m the wrong person, I say so.
- 02
Architect and propose
Within 48h: data model, system boundaries, integrations, UX — and one fixed fee with dated milestones. No “starting at”.
- 03
Build
4 to 6 weeks of controlled iterations. A demo every Friday on a Vercel preview. You watch the system grow live — no black box.
- 04
Launch
Payments, analytics, deployment, operational workflows. You leave with the repo, the accounts, the pipeline — and 30 days of 24h replies.
- 05
Improve
Measured iteration on real usage: a fresh fixed-fee scope per increment, or nothing if your team takes over. No hidden dependency.
Frequently asked questions
How much does process automation cost in Switzerland?
Usually CHF 20’000–50’000, fixed fee, for a complete workflow system — intake, rules, states, review, integrations — over 4–8 weeks. The price depends on the number of rules, integrations and roles.
Do you need AI to automate document processing?
Rarely to start with. A structured form, a computed checklist and written rules remove most of the manual work. AI then helps read unstructured documents, with validated output and human review when confidence is low.
What is the difference between an AI engineer and an automation consultant?
The consultant recommends tools and processes; the engineer builds the system and is accountable for how it works: data, rules, integrations, evaluation of AI output. I do the latter, fixed fee.
How is this different from Zapier, Make or RPA?
Those tools connect existing applications and fit when the process is simple and stable. When you need your own rules, permissions, an audit trail or a client space, a dedicated system costs less over time.
Where are the documents stored?
In private storage, in Europe by default or in Switzerland when required, never as email attachments. The choice is documented in the proposal.
Can a person still approve the result?
It is the default for any sensitive decision: the system prepares, a person approves, and the approval is logged with its author.
Further reading
- When should a SaaS use AI? And when it absolutely should notI have built three decision engines in production — tax, air passenger law, scoring — and none of them uses a language model to decide. Here is the rule I apply, the cases where AI belongs, and why RAG is often the wrong answer.AI product engineering8 minRead →
- The limits of AI for building a web productWhat a coding agent does well, what needs evidence, and the minimum evidence to get — task by task, from the interface to maintenance.AI product engineering8 minRead →
- Next.js architecture for a production SaaS: the decisions I make on every productWhat separates a Next.js project that holds from one that gets rewritten after a year comes down to about ten decisions. Here are mine, as applied on Atamatax, Horvera, Wabiro, SponsorsMetrics — and on this site.Next.js & full-stack9 minRead →
Other services
- Custom web applicationInternal tools, client portals, eligibility engines, back offices — the software that replaces the spreadsheet and the emails.
- AI developerAI features inside a real product — extraction, classification, structured generation — with validation, evaluation and costs under control.
- Architecture and product auditThe paid diagnostic before the build: v1 scope, target architecture, risks, plan and budget. CHF 2’400, seven working days, 100% credited against the build.
A process that costs hours every week?
Describe it: what comes in, who handles it, how often per month, what happens when it goes wrong. Six questions to check the fit.
danilson.daveigaramos.pro@gmail.com · Reply within 24h, Mon–Fri.