Skip to content

AI Consulting and AI Development

AI consulting and AI adoption in your company: rollout steps, cost factors, data protection and works council, AI literacy training, through to operations.

AI That Keeps Running After the Pilot

We advise on AI because we use it ourselves every day, and have since 2023 – in development, accounting, marketing. That practice tells us what holds up, where things stall, how people actually behave around AI and which levers matter. So our consulting starts where a plain Copilot rollout ends.

  • since 2023 AI in practice
  • >50% of our 2026 code changes are AI-assisted
  • >2,800 AI-assisted commits since 2025
  • 12 automated review agents before every merge
  • 13+ articles on AI

Measured on 22 September 2026 across 47 repositories and 13,835 commits, based on the AI attribution notes in commit messages; the share is a lower bound.

Why JASP?

  • Measured, not estimated: We count the AI share of our code changes through commit trailers and publish it
  • Control is built in: Twelve automated review agents check every change before merge: quality, security, tests, frontend, backend, DevOps – a team rule since May 2026
  • We write it down, failures included: We publish what we learn about AI, including the four-part series “Brilliant yet Clueless” on working with an AI agent
  • “Run smarter Projects”: AI is a tool, not a goal – we recommend it where it saves work, and say so when it doesn’t

What We Advise On

  • First use case: We score your ideas by frequency, data, risk, ownership and test cases, and cut a bounded start
  • Data traffic light: Which data may go into which AI system? We sort by sensitivity and record the rule per data type. This does not replace legal advice
  • Pilot to operations: Pilots rarely stall on technology. They stall on missing acceptance, unclear ownership and a process that doesn’t fit. We settle those before anything gets built
  • Solution path: Microsoft Copilot, RAG, agents, custom development or a mix – compared against your case, not the product promise

How Does an AI Rollout Work?

An AI rollout is not a licence purchase but a series of bounded decisions. We proceed like this:

  1. Choose a use case: A frequent, bounded task with existing data and correctable mistakes. How the scoring works is in “How Do You Choose Your First AI Use Case?”
  2. Set the data traffic light: Which data may go into which tool, before anyone types anything in. The traffic light is explained in “Which Data Can Go into AI?”
  3. Choose the solution path: Copilot, RAG, agent or custom development, compared against your case. The decision aid is “Copilot, RAG, AI Agent or Custom Development?”
  4. Start small and test: One responsible person, a fixed set of normal, edge and rejection cases, and stop signals defined in advance.
  5. Move into operations: Settle acceptance, ownership and process fit. What pilots otherwise get stuck on is covered in “Why Do AI Pilots Get Stuck?”

In parallel we train the teams who will work with the AI. How much freedom an agent gets in operation is described in “Steer AI Agents by Situation”.

What Drives the Cost?

We name concrete prices once we know your case. These factors determine the effort:

  • Number of users: A per-seat licence often doesn’t pay off for a task that concerns only a few people. A custom assistant on your own documents is then cheaper
  • Scope of the use case: One task, or a multi-step process across several systems
  • State of the data: With RAG, how the sources are prepared often matters more for quality than the language model
  • Connections and permissions: Which systems are involved and who may see what
  • Approval points and test cases: The more expensive a mistake, the more human review the process needs
  • Training and operation: Enabling teams, checking quality, approving changes

How Do We Handle Data Protection and the Works Council?

Before the first input, we settle tool, contract, storage location, training use, purpose and deletion with you. The result is the data traffic light: one rule per data type. RAG doesn’t solve data protection automatically either, because source quality, data minimisation and access rights have to be right. If the AI runs on personal data, it needs a purpose, a legal basis and data minimisation.

You involve the works council and IT from the start. That applies most when AI evaluates employee behaviour. We prepare the technical basis for it: which data flows, who sees the results, what stays local. The legal assessment is for your legal department and not part of our consulting.

JASP Products With AI

  • Widget Builder: AI editor that builds Staffbase widgets by chat, and an MCP server for AI assistants
  • Automatic Translator: AI-assisted translation of SharePoint pages with Azure AI
  • M365 Search Connector: Staffbase content findable in Microsoft Search and Copilot
  • JASP Events: API keys per tenant, the same interface as an MCP server for AI assistants

What We Build

  • Custom assistants on your own documents: vendor-open, and without a per-seat licence that wouldn’t pay off. For a workshop foreman, one finds spare parts for the machine park in supplier catalogues in several languages and versions. Every answer links the source document, so he can check the part in the technical drawings. A quarter of an hour of searching becomes two minutes
  • Workflow automation, among others with n8n, for recurring processes with a clear trigger and result
  • RAG on Microsoft 365 content, so answers come from your documents and stay traceable
  • Agents with approval points: A human decides wherever a mistake gets expensive
  • Custom applications when the standard isn’t enough – for example the AI editor in the Widget Builder, building widgets by chat since April 2026
  • Enabling your teams: Training on AI literacy under Article 4 AI Act, so people know the tool’s options and limits

How We Do It Ourselves

Our accounting has run with AI since August 2026: documents get assigned, bookings proposed, and in uncertain cases a second agent tries to disprove the first. The partners still approve every external payment individually. Agents write our marketing drafts, and nothing gets published without human approval.

How much an agent may decide alone depends on the situation. A tool of our own gives every AI session context and rules at the start, matched to the current state. At the desk, the AI asks back and shows results in a local preview before anything moves on. In meetings for a few hours, it may decide more within an extra set of rules. Overnight it takes a task through all checks and automated tests to a staging environment on its own. That takes about an hour and is right when nobody is waiting. We set autonomy per situation, not globally.

Have an AI idea and want to know whether it holds up? Get in touch. We look at task, data and risks, and tell you where to start.

Your Situation

  • You have plenty of AI ideas and don't know which one to tackle first
  • A pilot has been running for months but never reaches productive operations
  • You don't know which data may go into which AI system
  • You're deciding between Microsoft Copilot, a custom solution and a mix of both
  • Your teams are supposed to use AI but know neither its options nor its limits

Our Services

  • Selecting and scoring a bounded first use case
  • Data traffic light: which data may go into which AI system
  • From pilot to operations: acceptance, ownership, process fit
  • Solution comparison: Microsoft Copilot, RAG, agents, custom development
  • Workflow automation, among others with n8n
  • Custom assistants on your own documents, vendor-open
  • RAG on Microsoft 365 content and agents with approval points
  • Training on AI literacy under Article 4 of the AI Act

Examples

  • Custom assistant for a workshop foreman: finds spare parts in multilingual supplier catalogues – from a quarter of an hour of searching to two minutes, and every answer links the source document
  • Individual chat assistants for companies on their own documents, vendor-open
  • AI editor in the Widget Builder since April 2026: build widgets by chat, with live preview and your choice of Anthropic, OpenAI or Azure OpenAI
  • Our own accounting has run with AI since August 2026: document assignment and booking proposals automated, every external payment approved individually
  • Local AI without data leaving the machine: image generation with Draw Things and Flux 2, code review with LM Studio and Gemma

Frequently Asked Questions About AI Consulting and AI Development

Where do we start if we have done nothing with AI yet?
With a small, frequent task whose data already exists and where a mistake stays correctable. We score your ideas together by frequency, data, risk, ownership and test cases, and cut a bounded first use case out of them. We wrote up the method in “How Do You Choose Your First AI Use Case?”.
Buy Microsoft Copilot or build something ourselves?
That depends on the task, not on the vendor. Microsoft Copilot plays to its strength when the work happens inside Microsoft 365 anyway. If a task concerns only a handful of people, a per-seat licence often doesn’t pay off for it, and a custom assistant on your own documents is the cheaper route. If a multi-step process is involved, agents with approval points fit better. Often the answer is a mix, and that is exactly what we compare against your case.
How do you handle GDPR and the AI Act?
We sort out with you which data may go into which AI system and record the rule per data type. That is the data traffic light. For AI literacy under Article 4 of the AI Act we train your teams. This is explicitly not legal advice: the binding assessment stays with your legal department, and we supply the technical basis for it.
Does all of this have to run in the cloud?
No. Part of our own AI work runs locally, without data leaving the machine – image generation with Draw Things and Flux 2, code review with LM Studio and Gemma. Local models are slower and weaker than the large cloud models, and for clearly bounded tasks they are enough. We tell you per use case what makes sense locally and what does not.
How does an AI rollout in a company work?
In five steps: choose a first use case, set the data traffic light, choose the solution path, start small with fixed test cases, and move the pilot into operations. Training for the teams runs alongside. The section “How Does an AI Rollout Work?” on this page walks through each step.
What does AI consulting or an AI rollout cost?
We don’t quote a flat rate, because it would be a guess. The price is driven by the number of users, the state of the data, the connection to your systems, the approval points, the training and the operation. If you know these before the first call, you get a quote you can rely on. The factors are listed in the section “What Drives the Cost?”.
Do we have to involve the works council?
We recommend involving the works council and IT from the start and agreeing together which data the AI may process. That matters most when AI evaluates employee behaviour, such as reading behaviour in the intranet. It avoids having to rebuild trust later. Which co-determination rights apply in your case is for your legal department to decide. This is not legal advice from us.
Is AI training for employees mandatory?
Article 4 of the AI Act requires providers and deployers of AI systems to promote AI literacy among the people who work with these systems on their behalf. According to the European Commission, the measures depend on prior knowledge, context of use and risk. We train your teams accordingly. The binding assessment stays with your legal department.
Which JASP products use AI?
The AI editor in the Widget Builder builds widgets by chat. The Automatic Translator translates SharePoint pages with Azure AI. The M365 Search Connector makes Staffbase content visible to Microsoft Copilot. JASP Events offers its interface as an MCP server for AI assistants as well.

Ready to Talk?

Let's talk about your project – no obligation, completely free.

Free Initial Consultation