Microsoft's AI and data tools, and how we choose between them
Five technologies that come up in almost every Zooz-tech project. For each one: what it is, when it is the right choice, what we build with it, and where it already runs for clients.
Microsoft Copilot Studio
Copilot Studio is Microsoft's platform for building AI agents inside Microsoft 365 and Teams. We use it mainly for internal employee processes, and connect the agents to core systems through Power Automate and MCP layers that we build.
When the agent needs to live in Teams and Microsoft 365, authentication is through Entra ID, and you want to launch quickly on ready-made connectors. It can also be published on a website.
- Agents that take actionThey read, check and write to SAP, Priority ERP and Dynamics 365, with user approval before every write.
- MCP for on-premises systemsA controlled layer that exposes to the agent only defined entities and actions, and is reused in every new process.
- End-to-end processesFrom an incoming email, through understanding the document, to finished documents and an email to the next person in line.
- Properly indexed knowledgeWe do not rely on the built-in indexing over SharePoint. We connect the agent to Azure AI Search.
Azure AI Foundry
Azure AI Foundry is Azure's platform for building, running and governing AI applications and agents, with model choice, evaluation and monitoring. For us, it is usually the choice for external-facing bots: your own design, custom authentication, voice, multiple channels and full control.
When you need authentication such as OTP against a CRM, a branded interface, Generative UI, a voice and phone bot, or a specific model, whether the users are customers or employees.
Copilot Studio agent development →Full comparison: Copilot Studio vs. Azure AI Foundry →
- Voice and branded botsOn the website, on WhatsApp and by phone, with cards, forms and buttons generated within the conversation.
- Grounding gatesAn answer with no source, or with weak grounding, is blocked before it is sent.
- Operational agentsIntent recognition, access to ERP and external services, and handoff to a person when needed.
- Models in your cloudDeployed in the organization's own Azure, with data residency and cost control.
Azure AI Search
Azure AI Search is Azure's search and indexing service, the foundation of every serious RAG system, and the vector database we work with. We index the website, documents, SharePoint and the content management system into it ourselves, because planned indexing produces more accurate answers than the tools' built-in indexing.
Always, when a bot or agent needs to answer from organizational knowledge, and especially when different users have different permissions.
- An existing website becomes a knowledge baseCrawled, indexed and kept up to date along with the website. This is the engine behind our voice bots.
- Search that understands HebrewHybrid search, synonyms and recognition of Hebrew prefixes such as ve-, ha- and be-.
- Security trimmingEach user gets answers only from documents they are authorized to see.
- Source citationEvery answer points to the document, chapter and page it came from.
Azure AI Document Intelligence
Document Intelligence is the Azure service for understanding documents: text, handwriting, tables, fields and structure. We combine it with vision models and business logic to read even the documents that come in from the field: skewed photos, faded receipts and handwritten invoices.
When a process starts with a document: an invoice, a form, a receipt, a delivery note, a claim file or a photo of a license plate.
- Handwriting and print togetherHebrew and English, numbers, amounts and identifiers, in the same document.
- Understanding contextNot just recognizing 282.00, but understanding whether it is the total, the VAT or an item price.
- A vision model as backupWhen legibility is poor, a vision model completes the extraction.
- Straight into the processStructured JSON to the ERP, Excel, SAP or an agent that takes it from there.
Microsoft Fabric
Microsoft Fabric is Microsoft's unified data and analytics platform: OneLake, Data Factory, Lakehouse and Warehouse, with Power BI on top. We use it when you need a single data layer for reports, dashboards and models, not for every bot: most of our agents run in Copilot Studio or Foundry and read from the data.
When data is scattered across systems, management needs a single view in Power BI, and you want to prepare the data for AI use as well.
- A single data layerCollecting data from CRM, ERP and operational systems into OneLake, with managed load pipelines.
- Power BI on top of everythingManagement dashboards and reports on the same single source of truth.
- AI-readyClean, documented data that agents and models can work with.
- Governance and permissionsPermissions, data lineage and control in one place.
Reference architecture
Frequently asked questions about the technologies
How do you choose between Copilot Studio and Azure AI Foundry?
Choose by what the agent needs: Copilot Studio is quick to set up, with ready-made connectors, and is especially suited to agents inside Microsoft 365 and Teams. Azure AI Foundry gives full control over the interface, authentication and model choice, and is especially suited to customer-facing bots, voice and WhatsApp. See the full comparison: licensing, security and costs.
How do you connect Copilot Studio to SAP or Priority ERP?
Through an MCP layer that Zooz-tech builds against the system, even when it runs on-premises. The layer exposes to the agent only defined entities and actions, and every write action goes through validation and user approval. This is how the Gil Car invoice agent works with SAP.
How does Hebrew search work in Azure AI Search?
Hebrew requires handling prefixes (ve-, ha-, be-), synonyms and everyday language, and Hebrew and English mixed in the same sentence. We combine semantic search with keyword search and synonyms, with permission-based filtering. We cover this in detail in the RAG guide.
Can Document Intelligence read handwriting?
Yes. In the Tadiran project, we extract data from handwritten documents, and add a vision model when a document's legibility is poor.
When do you need Microsoft Fabric?
When data is scattered across systems, management needs a single view in Power BI, and you want to prepare the data for AI use as well. It is the foundation of the data platform for local authorities.
Guides: how it works in a real project
The tools on this page, in practice: how to index data, how to connect agents to systems and how to build secure portals.