Meet your Microsoft Accelerators for AI, Data & Cloud.

AI Agent Cost Optimisation — Driving 62% Spend Reduction

Nestlé India had deployed a suite of AI agents on Azure AI Foundry for demand forecasting, supply chain alerts, and customer query handling. Within months, the Azure OpenAI spend had ballooned significantly — token consumption was uncontrolled, model selection was suboptimal, and there was no observability over what each agent call was actually costing the business.

CodeSizzler conducted a full LLMOps cost audit — mapping every agent workflow to its actual token consumption, latency profile, and business value. We introduced a multi-model routing strategy: high-complexity reasoning tasks stayed on GPT-4o while simpler classification and extraction tasks were routed to GPT-4o-mini, reducing per-call cost by 70% on those flows. Prompt engineering was overhauled to eliminate redundant context injection, and response caching was implemented for high-frequency repeated queries.

We also deployed Azure AI Foundry's built-in observability stack with custom dashboards in Azure Monitor — giving Nestlé India's engineering team real-time visibility into token spend per agent, per workflow, and per environment. Cost anomaly alerts were configured to flag spend spikes before they compounded.

The result: a 62% reduction in monthly Azure AI spend within 8 weeks, with a 4× improvement in agent throughput due to reduced latency from optimised model routing. Nestlé India now operates a governed, cost-efficient AI agent fleet that scales predictably without surprise billing.

Enterprise Infrastructure Migration to Azure

Ensono, a leading managed services provider operating across the US and India, needed to migrate a complex on-premises infrastructure estate — spanning hundreds of VMs, legacy databases, and workloads across multiple data centres — to Azure. The challenge was zero-downtime migration with no disruption to Ensono's own client SLAs while managing complex dependency chains across the estate.

CodeSizzler led the end-to-end migration programme using Azure Migrate for assessment, Azure Site Recovery for lift-and-shift workloads, and a wave-based migration approach that sequenced workloads by dependency tier. Database workloads were migrated using Azure Database Migration Service, while stateful application tiers were re-platformed onto Azure Virtual Machine Scale Sets with auto-scaling policies configured from day one.

Landing zone design followed the Azure Cloud Adoption Framework — with hub-and-spoke networking, Azure Firewall, private endpoints for all PaaS services, and Azure Policy guardrails enforced across all subscriptions. Post-migration, we established a full Azure Monitor and Defender for Cloud baseline, giving Ensono's operations team complete observability over the migrated estate.

Migration completed across all waves with zero unplanned downtime. Infrastructure costs reduced by 40% through right-sizing, reserved instances, and elimination of idle resources. The migrated estate now operates at 99.9% uptime with automated failover policies in place.

Hybrid Cloud Environment Design & Deployment

NPCC (National Petroleum Construction Company), one of the largest construction and engineering companies in Abu Dhabi, operated critical project management, ERP, and operational systems entirely on-premises. Regulatory and data sovereignty requirements meant a full public cloud migration was not an option — but they needed cloud agility, scalability, and modern services for their rapidly growing project portfolio across the GCC.

CodeSizzler designed and deployed a purpose-built hybrid cloud architecture using Azure Arc to extend Azure management and governance to NPCC's on-premises servers — giving them Azure Policy, Azure Monitor, and Defender for Cloud coverage across their existing estate without lifting a single VM. Azure ExpressRoute was provisioned for a private, low-latency connection between NPCC's Abu Dhabi data centre and Azure UAE North.

Selected workloads were lifted to Azure while sensitive project data remained on-premises — with Microsoft Entra ID handling unified hybrid identity across both environments. Azure Stack was deployed for workloads that required local processing but cloud-consistent management. Azure Site Recovery was configured for on-premises VM backup and DR to Azure.

NPCC gained cloud agility without compromising data sovereignty. The hybrid environment delivered a 35% reduction in operational infrastructure costs, eliminated manual backup processes, and gave their IT team a single pane of glass across on-premises and cloud via Azure Arc and Azure Monitor.

Custom Microsoft Copilot for Retail Operations & Customer Experience

Shoppers Stop, one of India's leading premium retail chains, faced a dual challenge: store staff were spending significant time fielding internal process queries (inventory lookups, HR policies, promotion details) while simultaneously needing faster tools to serve increasingly digitally-savvy customers. Their Microsoft 365 investment was underutilised and no AI layer existed to unlock it.

CodeSizzler developed a multi-purpose Copilot agent using Microsoft Copilot Studio, deeply integrated with Shoppers Stop's internal knowledge base, HR systems, and inventory management platform. The agent was deployed inside Microsoft Teams — accessible to store managers and floor staff from any device. Natural language queries like "What is the current stock of SKU X at Phoenix Mall?" or "What is the leave policy for festive season?" returned instant, accurate answers.

For customer-facing scenarios, a second Copilot Studio agent was deployed on the Shoppers Stop website and WhatsApp channel — powered by Azure AI Foundry for complex product recommendation reasoning, integrating with the product catalogue, loyalty programme data, and live inventory via secure API connectors. Azure Content Safety was configured to govern all customer-facing outputs.

Within 90 days of deployment: 45% of routine internal queries were deflected from HR and operations teams, store staff response time to customer queries improved by 3×, and customer satisfaction scores for digital interactions increased measurably. The WhatsApp agent handled over 12,000 customer conversations in its first month.

Unified Data Analytics Platform on Microsoft Fabric

ABARTA, a major beverage distribution company in the US, was running fragmented reporting across sales, logistics, and inventory systems — with data living in siloed spreadsheets and on-premises databases. Monthly reporting took weeks and was consistently stale by the time it reached leadership.

CodeSizzler architected and delivered a unified Microsoft Fabric lakehouse, ingesting data from ABARTA's ERP, distribution management system, and retail POS feeds into a single medallion architecture. We built a semantic model on top and delivered a suite of Power BI dashboards covering route profitability, SKU performance, driver productivity, and inventory turnover — updated in near real-time.

ABARTA's leadership went from waiting three weeks for a monthly report to viewing live operational data every morning. Regional managers gained self-serve analytics for the first time, eliminating 14 hours of manual reporting per week across the business.

AI-Powered Loan Analytics & Borrower Intelligence

Highlands Mortgage faced growing pressure to accelerate loan officer decision-making while maintaining compliance rigour. Their data was scattered across a legacy LOS, spreadsheet-based pipeline trackers, and disconnected credit bureau feeds — making it nearly impossible to get a real-time view of pipeline health or borrower risk.

CodeSizzler built a Microsoft Fabric-based data platform that unified Highlands' loan origination system, credit data, and pipeline CRM into a centralised lakehouse. On top, we layered Azure OpenAI-powered document summarisation that extracts key data points from appraisals, income docs, and credit files — turning hours of manual review into a minute-long AI-assisted summary for each loan officer.

Loan processing time dropped by 75%. Branch managers gained real-time pipeline visibility, replacing the weekly email status report. Borrower response times improved significantly, giving Highlands a measurable edge in a fast-moving rate environment.

Copilot Studio & AI Foundry Agent for Insurance Operations at Daman

Daman, one of Abu Dhabi's leading health insurance providers, needed their operations and relationship management teams to handle member queries, policy lookups, and claims status faster — without expanding headcount. Their Dynamics 365 CRM held rich member history, policy records, and open claims, but staff spent 30–40% of their day manually querying the CRM and preparing management summaries.

CodeSizzler built a custom Copilot agent in Microsoft Copilot Studio, deeply integrated with Dynamics 365 CRM via Power Platform connectors and custom Azure AI Foundry plugins. The agent lets operations staff ask natural language questions directly in Microsoft Teams — "What is the claims status for member ID X?", "Draft a renewal reminder for policies expiring this month", "Show all open pre-authorisation requests from Abu Dhabi." For complex reasoning tasks — like generating tailored member communication or summarising claims history — the Copilot Studio agent orchestrates calls to Azure AI Foundry running against Daman's policy and claims data.

Operations staff saved an average of 2.5 hours per day on manual CRM queries and reporting. Management reporting moved from weekly manual extracts to on-demand Copilot summaries. The engagement went from kickoff to full production deployment in 12 weeks.

Omnichannel VOIP AI Agent for Grab Support, Singapore

Grab Support, the customer operations arm of Southeast Asia's leading super-app, handled massive inbound volumes across ride-hailing, food delivery, and payments — with agents burning out on repetitive queries and customers waiting on hold. They needed an AI agent that could handle natural spoken and text conversations end-to-end, across every channel Grab customers use.

CodeSizzler built a VOIP-integrated AI agent on Azure AI Foundry, combining Azure Speech Services (real-time speech-to-text and text-to-speech) with a custom reasoning agent trained on Grab's service policies and CRM data. The agent identifies caller intent in real time, pulls context from the CRM, and either resolves the query autonomously or performs a warm handover to a human agent with a full call summary pre-populated. The same agent identity was extended across WhatsApp Business API, Microsoft Teams, and a web widget — using Azure Communication Services and the Direct Line channel in Copilot Studio. Customers can start a conversation on WhatsApp, escalate to a voice call, and the agent retains full session context across every channel transition.

First-call resolution improved by 55%, average handle time dropped by 38%, and the organisation handled 3× the inbound query volume with the same human headcount. All interactions are logged to Azure Monitor with full conversation analytics piped into Power BI.