Automation and AI
The gains almost never come from a more powerful model. They come from three tasks that stop being done by hand. We work out which ones, we automate them on the tools you already have, and we leave in-house whatever has to stay there.
What we do
- Automation auditYou can see that things repeat, without knowing where to start.
- Connecting your toolsThe same data copied from one system to another, every week.
- Internal assistantThe same question thirty times, and the answer asleep in a document.
- Agents that actAnswering is not enough, the task has to get done.
- Self-hosted and compliant AIYour data is not allowed to leave the building.
A concrete example
An enquiry comes in through the website. It looks like a single event, but behind it six or seven things still have to happen before anyone really answers. That is where the time goes, and that is where enquiries get lost.
Today
7 manual steps
- The website form sends an email to someone.
- Create the contact in the CRM.
- Retype the same details into the management software.
- Create the folder in Drive or SharePoint.
- Tell the colleague who has to handle it.
- Draft the reply.
- Set a reminder. If anyone thinks of it.
With the flow
2 manual steps
- The enquiry comes in.
- The customer is matched, or created if they are new.
- The folder is created, named by your own rules.
- The details go into the CRM and the management software.
- The owner is assigned and told.
- The reply is drafted, ready to go.
- The follow-up is scheduled.
- Someone reads it, decides, and sends.
10 minutes per enquiry, 2 minutes after− 80 % of the time spent on each enquiry.For 40 enquiries handled a month, 64 hours of work freed up for other things.
More ordinary cases
| Today | With a flow |
|---|---|
| An invoice arrives by email, someone files it. | The document is recognised, named, filed and dropped in the right folder. |
| A prospect fills in a form, someone retypes it into the CRM. | The contact and the opportunity are created on their own. |
| An order changes the stock in one tool, but not on the website. | The two stay in agreement, without anyone thinking about it. |
| Someone hunts for a procedure through two hundred documents. | They ask the internal assistant, which answers and cites its sources. |
| Every Friday, someone gathers the same figures. | The report is produced and sent on its own. |
A company of twenty people does not have an artificial intelligence problem. It has orders copied from an email into a spreadsheet and then into accounting, a catalogue kept up to date twice because it lives in two places, and the same question asked thirty times a month whose answer already exists somewhere. That is where the time goes, not in the choice of model.
This is not only our own observation. Gartner expects up to 40% of enterprise applications to carry a task-specific agent by the end of 2026, up from less than 5% the year before, and at the same time expects more than 40% of agent projects to be cancelled before the end of 2027, for want of demonstrated value or controlled costs. Both are true together: the wave is here, and what separates a project that holds from one that is abandoned is decided before the first line of code, when the task is chosen. That is exactly what our audit settles.
We work on your tools rather than alongside them. A flow linking your management software to your website costs less to build and to maintain than one more piece of software, and it keeps running without us. The building blocks we use are open, n8n first among them, and they install with a host or on your own server just as readily.
For a public tender, or for an organisation handling personal data, the question is not performance but proof. Where the data lives, under what legal basis, who has access, and how to demonstrate it. The staged entry into force of the European regulation on artificial intelligence adds a classification of systems and transparency obligations. Those constraints are dealt with at design time, not after go-live.
One last thing, and it sometimes costs us a sale. Many tasks are not worth automating: those that come round three times a year, those whose rules change every month, those where an error costs more than the time saved. The audit serves as much to rule those out as to find the good ones.
A task that keeps coming back, and you are not sure it can be automated? That is exactly what the audit settles.
Common questions
Do we have to change our software?
No, and that is the point. We work on the tools you already have, because a flow that connects them costs less to build and to maintain than one more piece of software.
How long before it runs?
A simple connection between two systems takes a few days. An assistant on your documents takes weeks rather than days, most of the time going into collecting the sources and setting the access rights.
What happens if a flow stops?
It says so. Every flow we build reports when it ran and warns when it fails, on the channel you already read. A silent connection that stops on a Friday is discovered on Tuesday, when the numbers no longer add up.
Are we then dependent on you?
No. The flows are built on open tools, n8n first among them, documented, and you keep the keys. Taking over or handing the work to someone else does not require our agreement.
Our data has to stay in house, is that possible?
Yes, and it is a choice made use by use rather than all at once. Both the flow engine and the language model can run on a machine you control, at the price of lower performance for the same budget.