Part of Process automation.
Your process has clear steps.
The information does not
always fit a form.
We use AI at that point in the process. It interprets the message, document or image. Your business rules then decide what happens next, with a person involved where the task needs their judgement.
A customer explains a request in their own words. A supplier sends a document in a new layout. Before the process can continue, someone has to work out what the information means.
Use AI where the next step depends on understanding meaning or recognising a pattern. If the information is already structured and the rules are clear, ordinary automation can handle it.
A customer asks to change an order in their own words. AI identifies the requested change. The workflow finds the order, checks what is allowed, makes the change and confirms it.
Missing or conflicting details go to a person.
Suppliers send order confirmations in different layouts. AI finds the products, quantities and delivery dates. The workflow matches them to the purchase order and updates expected deliveries.
Price or quantity changes follow your review rules.
AI checks inspection photos for signs your team has defined, such as a missing label. The workflow sends possible issues for review and records the final result against the job.
People confirm findings. We test with your own images.
These are examples of possible projects. We agree the tools, access and checks with you before building.
A customer emails to ask where their order is. AI helps the workflow understand the question, even when the wording changes. The answer comes from your order and delivery systems.
Save the customer’s email and keep the original message.
AI identifies a delivery question and picks out any order reference in the message.
Match the customer to the right order, then get its delivery status and tracking link.
Send the verified delivery update when the request meets your rules. Other requests go to a colleague.
Save the response against the customer’s order so the team can see how the enquiry was handled.
If the customer cannot be matched, the request has several meanings or delivery information conflicts, a colleague gets the original message and the facts found so far. The workflow does not invent a delivery date to finish the reply.
AI helps understand the question. Your systems supply the order facts, and your rules decide what can happen next.
We look at the point where the process needs to understand something, then decide what can continue automatically and what needs a person. The checks should match the consequences of getting it wrong.
| Situation | What should happen |
|---|---|
| Different wording, same request | Recognise the intent and follow the agreed process. |
| Information in a new layout | Find the needed details and check them against the original. |
| Missing facts or several possible meanings | Ask for clarification or pass the task to a person. |
| A request outside your rules | Keep it for review rather than making a new policy decision. |
| Structured information with clear rules | Use ordinary automation if AI adds no useful interpretation. |
These projects show two ways we have explored AI in everyday work: understanding customer feedback and testing visual inspection.
Support conversations and public reviews feed a weekly report, giving the product team a way to look at recurring patterns.
See the feedback project AI vision / PilotThis pilot used the company’s own images to test models and data preparation. It established requirements for further work, rather than a production-wide guarantee.
See the inspection pilotWe agree what the first version should do, how to check it and who will look after it. Then we work through it with your team.
We follow the process with your team and identify what someone has to understand before the next step can happen. We check whether ordinary rules are enough.
We try the typical requests, unusual wording and difficult cases together. We check both the interpretation and the time your team still spends reviewing it.
We add the rules and approvals needed before a result can move to another tool or affect a customer.
We agree who reviews quality, handles problems and tests changes to the AI or the instructions it follows.
We agree which documents the AI can use, where they are processed, how long they are kept and the provider’s terms for using them.
We agree who reviews unclear results. Accounting decisions, contracts and decisions about people need the appropriate person involved.
A new AI model or a change to its instructions can change the result. We test it against the agreed examples before using it in live work.
We track missing details, wrong categories and corrections, so your team can see when the setup needs attention.
First projects start from CHF 4,800. We test the interpretation step on examples from your work before agreeing the full build. The proposal covers scope, timing, AI usage and the checking your team will still need to do. A workflow means one complete business task, or a connected set of tasks, automated from start to finish. It may take several n8n workflows, other tools or pieces of code to deliver that result.
When someone needs to interpret changing language, documents or images. If the information is already structured and the rules are clear, ordinary automation may be simpler and cheaper.
We find out by testing examples from your work. We agree the formats and languages to support and what happens with anything outside them.
We test a shared set of examples and look separately at wrong categories, missing values and invented details. We also measure how much checking your team still needs to do.
We agree that around the consequences of a mistake. A draft summary can need different checks from an accounting change or a message to a customer.
The original is kept. The task can wait, retry safely or go to a person, depending on what we have agreed. Missing information should not be filled in with a guess.
We check the actual AI provider, connected tools and contract, including storage and any use for training.
We review whether it meets the agreed checks and is worth using. Before launch, we agree connections, approvals, support and who watches the quality of real results.
Walk us through one process and the messages, documents or images it receives. We’ll look at where AI could help it move forward.
Book a first call