Processing invoices and documents
Incoming invoices, purchase orders and contracts are read out and placed in the right fields of your accounting system or case file. A staff member only checks the doubtful cases.
AI workflows
Less copy-and-paste work around your platform.
AI automation for repetitive tasks around your existing platform. With logging, and your team stays in control.
AI is valuable when it takes over the digital busywork people repeat every week. Not because it sounds good in a presentation.
We start with one process and only go further if it saves time or improves quality.
What is AI automation?
For us, AI automation means handing over one repetitive task around an existing system, such as mailboxes, forms, documents, spreadsheets or reporting. Zenjoy works for SMEs and organisations in Belgium and starts with a clearly scoped pilot, with human review, logging and a measurable result. No autonomous black box, no grand transformation promise, no chatbot nobody is accountable for.
When AI automation saves time, and when it doesn’t
We retype forms into the system.
AI converts the content of documents into structured fields on its own. A staff member only checks the doubtful cases.
The inbox eats our whole day.
AI classifies messages, drafts a reply and links the right record. A staff member sends the reply.
Making reports is copy and paste.
AI produces a first version from your raw data. A staff member checks and publishes. Reporting becomes reliable and repeatable.
We want AI, but not a black box.
Then start with one workflow, not with the choice of a model.
Each task looks small, but together they keep people from smooth follow-up, thorough research and good service.
One workflow with human review. Only then do we scale up.
The difference between using a standalone chat tool and building AI into your work process:
| Standalone chat | Integrated workflow | |
|---|---|---|
| Where the work happens | The user transfers information and results by hand | Data and actions are tied to the work process |
| Which data AI sees | Depends on the chosen tool, sources and settings | Agreed data sources, permissions and checkpoints |
| Who catches errors | The user checks the answer before using it | Validation, human review and logging in the process |
| What you manage afterwards | Management of accounts, sources and usage agreements | Management of integrations, model behaviour and product; also possible with Copilot Studio |
Good first workflows
- Automatically triaging a mailbox for membership operations or customer support
- Transferring unstructured information from documents into the right fields
- Structuring data for reporting and analysis
- Spotting patterns and errors in large data streams and suggesting improvements
What we don’t do
- Autonomous clinical diagnosis, triage or wellbeing decisions
- “AI replaces your team”
- Generic chatbot without a knowledge base or someone accountable
- A trial without a plan for rollout and maintenance
A workflow only works well when it is part of a platform we can build and maintain. That is where we differ from a pure AI specialist. Web platforms · Ethic for policy and governance.
Which business processes do we automate with AI?
The biggest time savings sit in the work that comes back every week and that nobody enjoys doing. These are the business processes we set up AI automation around most often.
Triaging and answering the mailbox
Messages are classified, linked to the right case file and given a suggested reply. Whoever follows up the case sends it.
Getting data out of forms
What people fill in on paper or in a web form ends up structured in your system, without retyping.
Preparing reporting
Figures from different sources are brought together into a first version of your report or dashboard. You check and publish.
Chatbots that cite their sources
An assistant that answers based on your own manuals and case files, and always shows where the answer comes from.
Not seeing your own process here? The question is not which field the task comes from, but whether it comes back often enough, follows clear rules and stays checkable. We look at that in a short audit of your work process, after which we choose between existing tools such as n8n and a custom integration.
AI automation for SMEs: what can you expect?
An SME has no team of data people and no year to spare. That is why we keep the first step small and measure what it delivers. Not sure which task should come first? Start with an AI Pitstop: one hour for one question.
Time savings
We count the gain in hours per week, not as a percentage on a slide. We estimate up front how much time the process costs today and measure again after the pilot. If it delivers too little, we say so.
Error margin
No model is right every time. We agree up front how often it may go wrong and what happens then: which results go through automatically, which ones someone checks and what ends up in the logging.
Training your staff
The colleagues doing the work today become the people who follow up and adjust the automation. We train them during the pilot and set out how they report an error. That way the efficiency holds up once we are gone. Want the whole team to learn to work with AI more broadly? That is what our tailored AI training is for.
What it costs
What a pilot around one workflow costs depends on the number of source systems, the state of your data, how strict the review has to be and whether an integration with your platform is part of it. We start with an exploratory call and then put the pilot into a quote with a fixed scope. After the pilot you know what the rollout costs and what it delivers, before you decide to go further.
What we have already built around models and data
Making an existing insect model usable in the field
We built a platform around the research model: mobile uploads, processing, review, corrections and export. The AI came from the research; Zenjoy turned it into a usable product.
Collecting speech data for AI research
A citizen science app for Flemish speech data. Proof of accessible data collection, not of an automated workflow or measured time savings.
From paper and Excel to usable data
Digitising food waste measurements. An example of the data foundation automation needs; not an AI case.
Pilot for one AI workflow
We take one process and test it with real historical examples from your organisation.
Map the process
Who does what, with which documents and systems, and where the time goes.
Review sample data
Real files or emails from the past, including the tricky cases.
Assess privacy and risks
Which data may go to an AI model and what goes wrong if it gets it wrong.
A prototype with review by your team
A working first version in which a colleague checks and corrects every result.
Analyse errors and estimate the time savings
We measure how often it fails, how serious that is and how much time it really saves.
Advise to roll out or to stop
An honest recommendation with numbers. Sometimes stopping is the best outcome.
Questions we often get
Not without an explicit agreement. We choose the model, the hosting and the data processing based on how sensitive the data is. No GDPR slogan, but a clear data flow.
Ordinary automation follows fixed rules: if this happens, do that. That works as long as everything is predictable, but it gets stuck on an invoice with an unusual layout or a mail that asks two questions at once.
AI automation fills exactly that gap. A model reads, classifies or summarises whatever does not fit a fixed rule and passes the result on to your existing process: sorting mail, reading documents, checking data or preparing a report. The software does the preparatory work. Someone from your team reviews it.
We start with one workflow. If that saves time, we only then go further.
Yes. That is why we provide thresholds, review by a staff member and logging, with a clear boundary for what the workflow may never do on its own.
Sometimes Microsoft Copilot is enough. Copilot Studio can also connect data sources and actions. We look at what is already available and what is missing: integrations, access rules, review and management. Custom work makes sense when that combination does not fit well in the existing tools.
That can be perfectly fine. For an integration between two tools with a clear trigger, n8n or Zapier are often faster and cheaper than custom work, and we will say so.
Custom work becomes interesting as soon as you need permissions per user, want results checked before they go through, want to be able to trace everything in logging, or the automation becomes part of the platform your customers and staff use every day. Then you do not want a stray flow nobody manages, but software with an owner.
No. Start with the workflow. The model follows from what that workflow needs.
No. We explore what AI can do, but we don’t push it. Sometimes a good data model or a plain integration is the better solution.
Which repetitive workflow costs you time every week?
One sentence is enough. We tell you honestly whether a pilot makes sense.
Book a call
You talk to someone who thinks along on the substance. No obligation, no standard pitch.
Not ready for a call yet? See what we have built