AI Use-Case Build
One AI use case, built and put into production on a foundation that is already ready.
We design, build, and deploy a single AI use case on your existing Azure foundation, evaluated against your real data and classified for the EU AI Act. You get a system in production with the monitoring that keeps it accurate, not a demo.
- Investment
- From €30,000
- Duration
- 6–10 weeks
- Format
- Remote-first, on-site for design
At a glance
- Investment
- From €30,000
- Duration
- 6–10 weeks
- Format
- Remote-first, on-site for design
- Markets served
- DACH and Benelux
Who it's for
- A discovery sprint ranked your use cases and you want the top one built, on a foundation already in good shape.
- A pilot proved out in a demo and now has to run in production, governed and monitored.
- You have a clear use case (retrieval over documents, document processing, or an agentic workflow) and want it delivered, not assessed.
- Multiple teams are experimenting with the same kind of AI use case independently, and you want one production-grade build instead of three fragile ones.
- Your board approved budget for exactly one AI system this year, and it has to work.
Scope and format
Six to ten weeks, remote-first with on-site sessions for design. We design, build, and deploy one AI use case on your existing Azure foundation: retrieval over your document estate, intelligent document processing into a downstream system such as Dynamics 365, or an agentic workflow. We evaluate retrieval or processing against your real data rather than assuming it, classify the system for the EU AI Act, and build in the monitoring. The team leads, with six-figure Azure AI delivery, intelligent document processing turning PO PDFs into validated SAP orders, and 500+-user enterprise search behind them. This build assumes the foundation is ready; if it is not, the AI-Ready Azure Platform builds both. Where the use case is an agentic workflow, it runs on the Microsoft Agent Framework with MCP and A2A for tool and agent connections, the same protocols behind the team's multi-agent orchestration across Claude, GPT, and Gemini.
What you get
The use case in production
One use case designed, built, and deployed on your foundation, running against your real data.
Retrieval or processing evaluated
The quality measured against your data, not assumed, with the numbers on the table.
EU AI Act classification and Annex IV documentation
The system mapped to its risk category and documented as an engineering deliverable.
MLOps and monitoring
The pipeline, monitoring, and the retraining or rollback path that keeps it accurate after go-live.
Integration into your systems
Wired into the downstream system the use case feeds: Dynamics 365, Microsoft 365, or your line-of-business app.
Evaluation harness
A repeatable test set your team can rerun after every model or prompt change, so quality regressions are caught before users notice them, not after.
Handover and runbook
Documentation and a working session so your team runs and extends it.
Pricing
Value-based and scoped per engagement, from €30,000. We price against the use case and the systems it touches, and fix the figure before we start.
If you ran an AI Discovery Sprint with us in the last eight weeks, that fee comes off this engagement.
Where your data sits
The build runs in your tenant, in the Azure region you are bound to. Access is least-privilege and scoped to the engagement. Where data residency matters, models and processing stay in the region you are bound to, and we sign your data processing agreement before we start.
Governance & compliance
- The use case classified to its EU AI Act risk category before design.
- Annex IV technical documentation produced as an engineering deliverable.
- Built in your tenant and your region, with data residency enforced.
- Post-market monitoring designed in, not promised for later.
Includes EU AI Act positioning and how we classify AI systems.
Read our governance approachCommon questions
How long does it take?
Six to ten weeks, depending on the use case and the systems it touches.
How is this different from the AI-Ready Azure Platform?
The platform builds the governed foundation and the first use case together. This build assumes the foundation is ready and delivers the use case alone. If you are not sure the foundation is ready, the Microsoft Landscape Assessment tells you.
Which use case do you build?
Retrieval over your documents, intelligent document processing, or an agentic workflow. If you have not picked one, the AI Discovery Sprint ranks the candidates first.
Do we own it afterwards?
Yes. Code, documentation, monitoring, and a working session for your team.
Who actually builds it?
Saša Pavicevic leads delivery: the same engineer behind the intelligent document processing that turns PO PDFs into validated SAP orders and the multi-agent orchestration across Claude, GPT, and Gemini. You work with the person who ships the code.
We already have a pilot. Do you rebuild it?
No. We take what proved out in the demo, evaluate it against your real data, and rebuild the parts that do not hold under production load: retrieval quality, error handling, and the monitoring that was never part of the pilot.
What happens after go-live?
The monitoring we build tracks accuracy against your data, not just uptime, and triggers the retraining or rollback path when it drifts. Your team owns and runs it after the handover session; we are available for a separate engagement if you want to keep us on it, not folded into this fee by default.
Which AI models do you use?
Whichever fits the use case and your data residency requirements, typically Azure OpenAI models, sometimes a smaller open model for a narrower task. The choice is made during design, not fixed upfront.
What if the use case involves multiple systems talking to each other, not just one model call?
That is the agentic workflow case: multiple agents or tool calls coordinated through the Microsoft Agent Framework. The build and the evaluation harness both account for that complexity, not just a single request-response check.
Is this a fixed-fee engagement like the assessments?
No. It is value-based and scoped per use case, because the systems it integrates with vary too much for a single fixed number. We fix the figure before kickoff once the use case and integration points are clear.