2026 Microsoft Partner of the Year Awards
Innovate with Azure AI Apps and Agents
Lighty, the Eurolux Smart Sales Agent Platform. First Technology turned a manual sales quoting process at Eurolux, one of South Africa’s leading lighting distributors, into a fast, intelligent and scalable workflow using a multi-agent platform orchestrated on Microsoft Azure AI Foundry. It went live in June 2026 and is now Eurolux’s primary quotation platform.
This is the same content our AI has been briefed on. Talk to it about any of this →
Overview and approach
Eurolux sales representatives prepared quotations with an Excel-and-macros tool, looking up products, stock and pricing by hand across catalogues and the ERP. Matching a client’s need to the right products depended on senior reps, which created bottlenecks, slowed the sales cycle and made quotes inconsistent. We agreed a business case with Eurolux senior management to move beyond the first product chatbot and rebuild quoting as an agentic workflow.
- Manual look-ups across catalogues and the ERP
- Key-person dependency on senior reps
- Quotes filed in local spreadsheets and email
- Inconsistent quotes, no single source of truth
- About an hour to produce one quote
- Natural-language quoting in the web app and Microsoft Teams
- Live product, stock and price checked automatically
- Templated, professional quotations every time
- Grounded in live ERP data and approved knowledge
- About 18 seconds per quote
Architecture on Azure AI Foundry
In production, deployed in Eurolux’s own Azure tenant, Azure South Africa North, in-region and POPIA-aligned, designed to the Cloud Adoption Framework and the Well-Architected Framework for AI. Lighty’s orchestrator agent interprets the rep’s intent and delegates to two specialist agents: a stock agent fetches live product, stock and pricing from the ERP, and a knowledge agent answers from Eurolux’s indexed lighting knowledge using retrieval-augmented generation over Azure AI Search. All three agents share one Azure OpenAI GPT-4o model. Lighty then applies business rules, composes the answer and generates the quotation.
- 1. A rep signs in with Microsoft Entra ID and asks Lighty a product or quoting question, in the web app or Microsoft Teams.
- 2. The orchestrator interprets the intent and routes to the stock agent for live product, stock and price from the ERP.
- 3. The knowledge agent retrieves relevant lighting knowledge from Azure AI Search, vector plus semantic, over approved content.
- 4. The shared GPT-4o model reasons over the grounded context and Eurolux’s business rules.
- 5. Lighty composes the answer, generates the templated quotation and logs the conversation.
Every answer and quote is grounded in two places: Eurolux’s own indexed product knowledge through Azure AI Search, and live, structured ERP data through the stock agent. The model is constrained to approved enterprise content, and customer data is never used to train foundation models. In practice this materially reduces hallucination and keeps pricing correct.
Infrastructure is defined as code in Bicep and promoted through CI/CD pipelines in Azure DevOps. Application Insights and Log Analytics carry full telemetry and audit logging, and Azure Cost Management holds spend to a budget.
Built on the Microsoft stack
| Azure service | Role in the solution |
|---|---|
| Azure AI Foundry | Orchestrates the multi-agent solution; model catalogue picks the most cost-effective model per workload. |
| Azure OpenAI (GPT-4o) | Shared foundation model for reasoning and generation across all agents; embeddings for retrieval. |
| Azure AI Search | Vector and semantic index over Eurolux product catalogues; the retrieval layer for grounding. |
| Azure App Service | Hosts the web chat interface and the backend API. |
| Microsoft Teams | Puts Lighty in the sales team’s daily flow of work. |
| Azure Cosmos DB | Conversation history and sentiment, for continuity and oversight. |
| Azure SQL Database | Application state and background-task scheduling. |
| Azure Blob Storage | Source product documents and assets for ingestion. |
| Microsoft Entra ID | Single sign-on; only authenticated, authorised Eurolux staff get access. |
| Azure Key Vault | Secrets, API keys and connection strings. |
| Application Insights + Log Analytics | Monitoring, telemetry and audit logging. |
| Microsoft Defender for Cloud | Threat protection and security posture across the subscription. |
| Azure Cost Management | Budgets and cost-optimisation alerts. |
| Azure DevOps + Bicep | Infrastructure as code and CI/CD across environments. |
The numbers
The business case uses Eurolux’s own figures from the approved scope of work (ref. LIGHTY250319). Production figures come from Azure Monitor in the Eurolux tenant.
8 reps × 3 quotes/day × 22 days = 528 quotes a month, at one manual hour each (R100/hr): R52,800 a month, or R633,600 a year.
Build R149,050 one-off; support R136,800/yr; Azure R144,000/yr. Year-1 R429,850, ongoing R280,800. Payback ~8 months, ongoing net benefit R352,800/yr (126% on run-cost).
One framework, many industries
Lighty was built on FirstAi, our reusable Azure AI agent and RAG framework. It packages the parts of enterprise AI that never change: an AI-workload landing zone, Entra ID authentication, retrieval grounding over Azure AI Search, model selection from the Foundry catalogue, prompt guardrails and content safety, Cosmos DB conversation history, and infrastructure as code with Bicep and Azure DevOps. What changes per customer is the persona, the knowledge and the business rules. That is what makes it repeatable across an industry, a functional role and a customer segment.
| Deployment | Industry | Pattern on Azure |
|---|---|---|
| Lighty (Eurolux) | Distribution | Multi-agent orchestration, RAG and live ERP data on Azure AI Foundry |
| Portia (City of Cape Town) | Government, legal | Fine-tuned model and RAG agent over thousands of prior legal opinions |
| Harvey (in-house legal) | Legal services | RAG assistant on FirstAi, deployed in about a week |
| African Alliance | Financial services | Multi-model RAG, intent routing and AI tools on Azure Functions |
| Inverter monitoring | Financial services | IoT ingestion with generative-AI insight and outlier detection |
| Retail Intelligence | Retail | Copilot stack and Azure AI Foundry adoption programme |
| Health Intelligence | Health and life sciences | Copilot stack and Azure AI Foundry adoption programme |
Responsible AI, security and governance
Design decisions made on day one, aligned to Microsoft’s Cloud Adoption Framework and the Well-Architected Framework for AI, not patches applied later.
- In-tenant. The whole solution runs inside Eurolux’s own Azure tenant. No customer data leaves that boundary.
- In-region. All processing and storage run in Azure South Africa North, supporting POPIA-aligned data residency.
- Identity-gated. Access is limited to authenticated, authorised Eurolux staff through Microsoft Entra ID single sign-on; the API is key-secured.
- Grounded and guarded. RAG constrains answers to approved knowledge and live ERP data; prompt guardrails and content safety sit at the boundary.
- Secrets and encryption. Secrets live in Azure Key Vault; data is encrypted in transit and at rest.
- Threat protection. Microsoft Defender for Cloud watches posture and threats across the subscription.
- Private by design. Customer data is never used to train foundation models.
Customer evidence and working with Microsoft
“There is much excitement about the Lighty sales assistant agent. First Technology has deployed the most innovative emerging technology solution into Eurolux, something few of our other vendors was ever able to do.”
“This is the fastest any innovative application has gone from POC to approval by the City board for release to production. Well done and thank you.”
“A belated thanks to the First Technology team for delivery of the Natural Language AI solution! Super exciting space and thanks again!”
Microsoft funded the original Lighty proof of value. First Technology is a Microsoft Solutions Partner for Data & AI on Azure, the first African-headquartered partner to achieve Microsoft’s Advanced Specialisation in AI on Azure, and our team worked directly with Microsoft technical experts on the City of Cape Town legal-opinion agent.
Still curious?