We start with a specific problem and a small pilot; if it works, we expand. You do not need to know which metric to improve: we measure it in the diagnosis. The usual starting point is a process diagnosis (indicative range: €100–180).
How a proposal is written
Every proposal has three parts. If it cannot be measured or does not depend on you, it is not promised.
What changes
The expected result in time, errors, cost, capacity or missed calls.
Timeline
The agreed time for the pilot, based on scope and dependencies.
Condition
What the project needs to work: a data sample, access, daily use and one person to approve.
Documents and data
Teams that spend hours every week copying, checking and validating information from invoices, job sheets, orders or spreadsheets.
What changes
The team stops re-entering data and errors are caught before they reach the system.
How it is measured
% of documents that require manual review, time per batch and errors caught before accounting. The baseline is set in the diagnosis.
Condition
A representative sample of documents and one person to validate the criteria.
Includes
Field extraction from structured and semi-structured documents.
Rule-based validation (amounts, formats, reconciliation) with doubtful cases flagged.
Export to CSV, Excel or directly to your system via API.
Automated reports and summaries of what has been processed.
Does not include
Handwritten documents or low-quality scans without a prior sample.
Accounting or tax decisions: validation is technical, not advice.
Indicative pilot investment: €350–650
AI in existing processes
Classifying messages, summarising documentation or answering repeated questions takes time and depends on team memory.
What changes
Repeated questions are answered from approved information and doubtful cases reach a person with context.
How it is measured
% of questions resolved from the approved base, % escalated and time to first response.
Condition
An approved knowledge base and a person available for escalated cases.
Includes
Classification and summarisation assisted by language models.
Assisted extraction followed by deterministic validation.
Assistants connected to your information and actions, on a chat owned by your website or system, or on WhatsApp, Telegram, LINE and email.
Workflows that call an API or record results, with human review for sensitive cases.
Does not include
Automated decisions without supervision in sensitive cases.
Use of data that is not authorised or requires a specific contract.
Indicative pilot investment: €600–1,000
Integrations between tools
Information lives in systems that do not talk to each other and someone moves it by hand.
What changes
Information moves from one system to another without manual copying and with an alert if something fails.
How it is measured
Successful synchronisations, incidents, manual interventions and latency.
Condition
Access to the APIs or a reliable export from the systems involved.
Includes
Connecting applications through APIs with error handling and retries.
Data synchronisation between systems, with logs and notifications.
Integration with the APIs of your tools: CRM, calendar, WhatsApp Business, Telegram, LINE, email and your own systems.
Automation of internal tasks and notifications.
Python and SQL workflows on your existing databases.
Does not include
Systems without an API or a reliable export/import mechanism.
Full platform migrations (quoted separately).
Indicative pilot investment: €600–1,000
AI voice agent
Missed calls outside opening hours or repeated questions that interrupt in-person service. Only proposed after defining scope, telephony, limits, tests and operating costs.
What changes
Calls outside opening hours or at peak times are recorded and handed over instead of being lost.
How it is measured
Calls handled, requests recorded, % handed over and time to first response.
Condition
An approved script, a handover destination and a real integration if booking or writing is promised.
Includes
Answers based on information approved by the business.
Capture of contact details and recording of requests.
Handover to a person when the case requires it.
Call summary with the recorded data.
Testing with real scripts before activation.
Does not include
Bookings or writes into systems without prior real integration.
Promises of immediate availability: each case is validated first.
Indicative pilot investment: €1,500–3,500
Ongoing improvement (optional)
A delivered workflow needs adjustments, supervision or improvements, but you do not want unlimited maintenance.
What changes
The flow keeps working and every month you know what was processed, what failed and what was improved.
How it is measured
Incidents, cases reviewed, hours used and improvements delivered.
Condition
Only if the flow needs it; hours and response times in writing.
Includes
Support hours and response times defined in writing.
Error review and small adjustments.
Configuration changes and agreed improvements.
Monthly report with volume, incidents and proposed improvements.
Does not include
Unlimited maintenance or permanent urgent support.
New features outside the agreed scope.
Indicative pilot investment: €150–300/month
When it is not worth it
Saying no is also part of the diagnosis.
The process happens only a few times a month or is not repeated.
There is no representative data sample or access to the tools.
The decision must be 100% human and there is no repeatable part.
The destination system has no API or reliable export.
The process needs tidying first: automating disorder only makes it faster.
You are only looking for the cheapest option: there are templates that cost less.
Questions before we start
I do not know which metric to improve. Can I still start?
Yes. In the diagnosis we measure the current process and agree what is worth moving: time, errors, cost, capacity or missed calls.
Can I not just do this myself with Claude or a no-code tool?
Sometimes you can. If the process is simple and does not touch critical systems, I will tell you. Doing it yourself is usually not the right option when data lives in another system, nothing can be lost, an audit trail is needed or something has to respond when it fails in production.
Are you going to replace my team?
No. I automate the repetitive part and leave your team what needs judgement. In sensitive cases, the system flags and hands over; it does not decide.
What happens if the AI makes a mistake?
What can be verified is validated with deterministic rules; what is probable is checked and flagged for human review. If the system does not know, it says so and escalates. And there is always a manual fallback.
Do not see your exact case?
Describe the process you have in mind: how it works today, how often and which tools it uses. I will tell you honestly whether it is worth automating.