AI assistants & automation
AI for your daily work
I build and test an AI assistant for a specific task. Your team assesses whether the results are useful in everyday work.
Start with a defined use case
A useful starting point is a task you can evaluate in practice. We compare the current effort with an initial test and establish which results a person needs to check.
- Pricing
- Quoted following a feasibility review
- Timing
- After reviewing the workflow, data access and approvals, you receive a proposal with a timeline and fixed price for the agreed scope.
Before work begins, you receive a proposal with a clear scope and costs.
Test AI against a specific task
Before the pilot, we define a usable result and who decides when details are missing or sources conflict.
Example of a bounded pilot
Prepare a draft customer response
This is an illustrative workflow. Data, tools and checks are agreed before a pilot begins.
Example details
- Source material
- Approved information about your own services
- Output
- A draft referring to the information used
- Approval
- A person checks the claims and sends the response
Identify tasks that can use fixed rules
An acknowledgement or the transfer of a known form field can often use fixed rules. AI may suit variable free text or searching documents. We choose the approach based on the task and compare its effort with the current process.
Test incomplete and incorrect answers
We use examples with missing details, conflicting sources or no relevant information. The assistant should make those limits apparent. Access and human approval for sensitive actions are established before production use.
A simple automation may be sufficient
When inputs and decisions are clearly defined, we assess a workflow with fixed rules. An AI pilot is useful to investigate where those rules do not adequately cover the task and the results can be meaningfully checked.
Review workflows and tools togetherYour AI pilot
- Feasibility review
- We assess the task and access to the data it requires. You learn which approach could work, what costs to expect and what needs resolving before a start.
- Initial pilot
- I implement one agreed work step. Real examples show how the assistant handles complete, missing or conflicting information.
- Review and approval
- We agree who checks results and which actions need approval. Typical failures have defined feedback and a way to stop the workflow.
- Evaluation and handover
- You receive the test results and guidance on operation and limitations. This gives us a basis for deciding whether further work is worthwhile.
Delivery and handover
Choose the task
We examine a recurring work step and agree how you will judge whether the assistance is useful.
Develop the pilot
I connect the data and tools required. Access is limited to the agreed purpose.
Assess the results
Your team tests the pilot with realistic examples. The results guide decisions on changes or further development.
Before commissioning
Does this fit my business?
Delivery fits SMEs with a recurring, bounded work step in which information is found, structured, prepared or passed on after an approval.
The workflow occurs regularly and creates meaningful manual effort.
Inputs, intended output and responsible approval can be named.
Required data is lawfully accessible and has an accountable source.
What do I need to provide?
Real, sanitised examples from the selected workflow
One subject-matter owner for rules, tests and approvals
Documented access to the required systems or a technical contact
What is agreed separately?
Exactly one clear use case is scoped first.
Legal advice is outside the technical implementation scope. Legal questions, any data protection impact assessment and workplace agreements require separate expert review where applicable.
Sensitive or irreversible actions are not automated without suitable approval.
Licences, model usage and third-party services are disclosed separately in advance.
Which process makes a good first pilot?
A frequent, clear step with known inputs and a verifiable output. Good candidates prepare decisions; they do not immediately replace critical approvals.
Does our data need to be perfect?
No. But its source, owner and minimum quality must be known. If data conflicts or cannot be accessed reliably, that foundation is addressed first.
Do we need MCP immediately?
Not necessarily. MCP is one way to connect approved tools and data. We first check whether an existing API or a limited data import is enough for your task.
Can AI send emails or change data itself?
That can be technically possible. In the first stage, sensitive actions are designed with preview, approval, limited permissions and auditable logging.
How is value assessed?
Before building, we agree test cases and a specific goal, such as fewer manual transfers or more complete reply drafts. We then compare the pilot with the existing workflow.
What ongoing costs should we expect?
That depends on the model, usage, hosting and connected services. External costs are shown separately from implementation and support in the proposal.
What if the pilot is not convincing?
It is not expanded automatically. You receive the documented findings, limits and a clear recommendation to adjust, choose another approach or stop.
Your offer as an app in ChatGPT
Beyond internal assistants, a clearly defined part of your offer can become usable directly in a ChatGPT conversation – through an MCP connection to approved data and tools. The pilot starts at €2,900 net if a suitable interface exists.
When the pilot fits
Customers or staff repeatedly select, compare or look up information from your offer, and the data is accessible through an interface.
What the pilot delivers
A defined workflow in ChatGPT with a data and permissions plan, an MCP connection, tests covering normal and edge cases, and documentation for a possible submission.
What is clarified first
Interface, data quality, permissions and approvals. Missing interfaces are quoted separately beforehand; approval by OpenAI cannot be guaranteed.
Sensitive actions such as bookings or orders receive limited permissions and explicit confirmation in the conversation.
Discuss the app pilotWhich task could AI help your team with?
Describe a recurring task and the tools you use for it. We can assess whether an AI pilot or another technical solution is appropriate.