Concept Media · AI for service businesses and SMEs

AI assistants and agents

Build an AI assistant around a defined task, approved knowledge and selected system actions. Give your team a useful helper with clear limits and a route to a person.

A clear task and a practical delivery

The task

A general chatbot can answer questions, but your team needs help completing a specific part of the work with company context.

What we deliver

We design a focused assistant, connect the agreed knowledge and tools, and define when it may prepare an action or must ask for approval. Testing covers missing data, unsupported requests and failed system connections.

Possible starting points

  • An assistant for one team workflow
  • Approved knowledge and limited tools
  • Review, handover and test scenarios

Illustrative workflow

A coordinator asks the assistant to prepare a follow-up on an open case. It reads permitted case information and proposes a task with a source reference. The coordinator can edit the text and approve creation. The assistant cannot change prices, delete a case or message the customer. If it cannot identify the right case, it must ask rather than select one itself.

An illustration of a possible delivery, not a reported customer result.

An example to scope together

From input to useful work

  1. A person asks

    The user supplies a case or question within the assistant’s defined task.

  2. AI prepares a step

    The model retrieves permitted knowledge and prepares one proposal with the required details.

  3. Rules validate access

    The application checks user access, field values and the allowed action.

  4. A person approves

    A person approves the single write action; the outcome is recorded for review.

How AI, systems and people work together

The role of AI

The model interprets a request and may propose a tool action. The application validates the request and permissions before executing it. Important changes follow the agreed human approval process.

What we need before starting

A narrow task, an identified user group, permitted data sources and a documented list of actions. API access and approval responsibilities must be available.

Test the complete task

Track completed tasks, useful drafts, rejected actions, handovers and time spent reviewing the work.

Scope and price

Focused AI assistant

From DKK 24,900excluding VAT · total project price

One task using approved sources and read access, with at most one write action approved by a person. Includes the audit, action limits, failure tests and training. More autonomous workflows are quoted separately.

A starting price does not guarantee that your task fits this scope. We check access, examples and acceptance criteria before you order and give a fixed quote. Licences and usage are separate.

The DKK 4,995 workflow audit is included in the total project price and fully credited against a subsequent solution for the same workflow. For example, a DKK 24,900 project leaves DKK 19,905 after the paid audit. If the task is already defined, necessary analysis is included in the project quote.

See all prices and terms

Planning the delivery

A focused assistant is indicatively planned over 3–6 weeks after source and tool access is available, including testing and staff approval.

This is a planning estimate, not an agreed delivery date. Availability, review and supplier dependencies are confirmed in the quote before work starts.

Questions before you start

Can the assistant send proposals or change prices itself?

That is outside the base scope. It covers read access and at most one write action with human approval. Price calculations and more autonomous actions need separate scoping and testing.

What if the assistant chooses the wrong tool?

The application must reject tools and actions outside the allowlist. We test incorrect case references, insufficient permissions and failed system calls before handover.

Start with one useful change.

Describe the task you want to make easier. We will discuss the fit and a concrete next step.