AI & dev
What we learn automating processes, shipping products and making them visible on Google and inside AI answer engines. Methods, numbers and the mistakes to avoid.
What a repetitive task really costs: the hours and francs math
A task of a quarter of an hour a day costs about CHF 6,300 a year. The step-by-step maths: loaded hourly cost, errors, real gain, payback, and the cases where you should not automate.
Insurance brokers: renewals without a spreadsheet
A renewals spreadsheet never warns anyone. Here is the reminder sequence that replaces manual rereading, what the machine prepares, what stays with you, and when not to automate.
Which Task to Start With: Choosing Your First Process to Automate
Ten tasks to automate, only one to pick. Five criteria scored 1 to 3, an hours-based calculation, an example, common mistakes and the cases where you should install nothing.
Monthly reporting without a spreadsheet: what prepares itself
The month’s report often lands mid-month. Here is what prepares itself, what is commented with review, what stays with you, and when not to automate at all.
Wealth advisors: prepare a client meeting in ten minutes
Meeting preparation looks more like searching than thinking. Here is the one-page brief that removes the search, what you review, and what stays yours.
Categorizing SME expenses without losing your month-end
Expense categorization is often done in bulk at month-end. Here is what an AI proposes well, what a person should validate, and when not to automate.
Chasing unpaid invoices: sequence, tone and when a person takes over
An unpaid invoice is rarely a refusal. Here is the four-message sequence, the tone of each step, what an AI prepares and the moments when a person takes over.
Wealth advisors: track open files and chase missing documents
A wealth advisor carries dozens of files at different stages. Here is what can run alone in the follow-up and reminders, what stays your decision, and how to size the gain.
Targeted alerts: only hear about what leaves the ordinary
Too many alerts are worse than none. How to choose what deserves a message, set thresholds, what AI adds, and when to install nothing at all.
Wealth manager: from first meeting to a complete client file
Between the first meeting and a complete file, chasing and filing take weeks. Here is what can run on its own, what stays your decision, and how to put a number on the gain.
Summarising Customer Service Calls: One Week of Listening
Nobody replays five days of calls. Here is what an automatic summary surfaces in one week, the six-step flow, its limits, and when not to install it.
Daily summary: replace ten browser tabs with one morning message
Touring your tools takes time and gets skipped on busy days. See what a good morning summary contains, how to choose its figures and when to skip it altogether.
Wealth manager: prepare client reports without losing evenings
A client report takes little time alone and a lot once multiplied. See what can be prepared automatically, what stays your commentary, and how to measure the gain.
Site reports dictated by voice: from voice memo to document
A report dictated on site replaces evening typing: the AI transcribes and structures, the person reviews and signs. Steps, limits, an error table and a test before you start.
AI-generated quotes and contracts: the safeguards to set
An AI assembles a quote or contract from your templates; it decides neither the price nor the clauses. Here are seven safeguards, a table of roles and a test to run before launch.
Accounting firms: get through the close without data-entry overtime
The close crushes teams because everything arrives at once. Here is the four-phase calendar, what AI prepares, and when this is the wrong choice.
Supplier invoice entry: what AI reads well, and badly
An AI reads well what is printed plainly and badly what needs context. Here is where it is reliable, five rules for deciding who has the last word, and how to test on your own invoices.
Reactivating Old Clients When Your CRM Has Been Asleep for Two Years
A CRM asleep for two years hides deals. Here is where to start: sort first, three messages, a four-step calendar, and what an AI prepares while you approve.
Accounting firms: collect client documents without chasing ten times
Chasing a client ten times for one invoice is not inevitable. Here is the collection loop, the reminder cadence and the point where a person takes over.
Accounting firms: the tasks that really automate
Five tasks weigh on an accounting firm’s time. Here is which ones an AI handles reliably, which need approval, and which to start with.
Answering an Inbound Enquiry Fast, Even in the Evening
An enquiry left waiting goes cold. Here are the published orders of magnitude, a five-step workflow to reply the same day, and the part an AI prepares without deciding for you.
Following Up on a Quote Without Nagging: The Rhythm That Works
An unanswered quote is almost never a refusal. Here is the follow-up rhythm that holds, the structure of a useful message, and the part an AI drafts while you approve.
Swiss FADP and AI: what to check before sending data to an LLM
Sending customer data to an LLM engages your responsibility under the Swiss FADP. Three data levels, five questions for the vendor and a checklist before going live.
Why Your AI Automations Fail on Real-World Data
An AI automation that shines in a demo degrades on your real data. Six break points, a four-step diagnosis method and the fixes to apply, in order.
Choosing an AI Model for Your Business: The Criteria That Matter
Model leaderboards shift every six weeks. Here are the six criteria and the five-step method we use to choose an AI model that survives production, not just a demo.
AI Coding Assistants: What Changes in Code Review
AI-generated code is almost always well formatted, which disarms a reviewer's usual instincts. Seven failure modes to know, and a method for calibrating review depth by risk.
Testing an AI System Before the Demo: A 5-Step Method
A successful demo proves nothing about how an AI system really behaves. Here is how to build a systematic evaluation before presenting it to a client or shipping it to production.
The Real Cost of an AI Feature in Production
The API price tag is almost never the real cost. Four cost layers make up an AI feature's production bill, and one of them usually outweighs the other three combined.
RAG in 2026: What Still Holds and What Got Replaced
Long context, agents, cost math: in 2026 the RAG versus context window debate is framed wrong. What actually got replaced, what still holds, and the method for choosing right on an SME project.
AI Agent in Production: What Breaks First
An AI agent that passes every test can still break in production for five specific reasons, almost never tied to the model itself. The checklist to catch them before launch.
Forward Deployed Engineer: the engineer who replaces the consultant
An engineer who starts from the problem rather than the ticket, and ships software to production rather than a recommendation. What the role covers, and when it is the wrong choice.