How I Cut Azure Costs by 78% at Royal Broker

In August 2022, I joined Royal Broker Solutions as an intern. Six months later, I was hired full-time with one main mission: take over and optimize our internal SaaS platform, InstaHR.

The problem? We were paying $600/month on Azure for an application used by about thirty people. It made no sense.

Here is exactly how we got down to $135/month.

The initial diagnosis

The first thing I did: audit the Azure bill service by service.

Azure App Service (Plan P2v3)         → $312/month
Azure SQL Database (Standard S3)      → $150/month
Azure Blob Storage                    →  $28/month
Azure Application Gateway             →  $85/month
Azure Active Directory Premium P1     →  $25/month
─────────────────────────────────────────────────
TOTAL                                 → $600/month

A few things jumped out right away:

1. The App Service P2v3 plan (4 vCPU, 14 GB RAM) was massively oversized for our actual load
2. Azure SQL Standard S3 (100 DTUs): our peak in concurrent queries rarely went above 5
3. The Application Gateway was being used as a plain load balancer for... a single instance
4. Azure AD Premium was enabled for features we weren't even using

The approach: Right-sizing + Serverless Architecture

Step 1: Analyze the real load

Before touching anything, I instrumented the app with Azure Monitor and Application Insights for 3 weeks.

Results:

• Average CPU: 8% (peak at 35% on Monday mornings)
• Memory: 2.1 GB used on average
• Max concurrent SQL queries: 12
• Traffic: roughly 200 requests/hour at peak hours

This data let me justify every migration decision to management. Never optimize blindly: metrics first.

Step 2: Migrate to a hybrid serverless architecture

Instead of a brutal all-at-once migration, I went module by module:

ASP.NET Core backend → Azure Functions (Consumption Plan)

The long-running operations (sending emails, generating PDFs, syncing with Zoho) were extracted into Azure Functions triggered by queues.

[FunctionName("SendCandidateEmail")]
public static async Task Run(
    [QueueTrigger("email-queue")] EmailMessage message,
    ILogger log)
{
    // Send email via SendGrid
    await _emailService.SendAsync(message);
}

Impact: These workloads now only run when they're triggered. Cost: nearly zero (covered by the free tier).

Azure SQL → Azure SQL Serverless

I migrated to the Serverless tier of Azure SQL, which scales automatically between 0.5 and 2 vCores, with auto-pause after 1h of inactivity.

-- Serverless configuration
CREATE DATABASE InstaHR
WITH
  (EDITION = 'GeneralPurpose',
   SERVICE_OBJECTIVE = 'GP_S_Gen5_1',
   MIN_CAPACITY = 0.5,
   AUTO_PAUSE_DELAY = 60);

Impact: We only pay for what we consume. Savings: $110/month.

App Service P2v3 → B2s

With the Azure Functions absorbing the asynchronous workloads, the main server no longer needed as many resources.

Before → P2v3 : 4 vCPU, 14 GB RAM → $312/month
After  → B2s  : 2 vCPU,  4 GB RAM →  $38/month

Application Gateway → Azure Front Door (Free Tier)

For our case (a single origin, SSL, no advanced WAF), Azure Front Door Basic was free.

The final architecture

Users
    ↓
Azure Front Door (Free)
    ↓
Azure App Service B2s ($38/month)
    ↓              ↓
Azure SQL        Azure Storage
Serverless       ($12/month)
($75/month)
    ↓
Azure Functions (Consumption)
Email / PDF / Zoho Sync
(~$10/month)
─────────────────────────────
TOTAL : $135/month (-78%)

The mistakes I would have avoided

1. Not having metrics from the start

I nearly scaled down the SQL plan without knowing the real load. If we'd had spikes of 200 concurrent queries, the serverless tier would have throttled and degraded the user experience.

2. Migrating everything at once

I went step by step: first the Azure Functions (no impact on the core), then SQL, then App Service. Each step was tested in staging for a week.

3. Ignoring network egress charges

Azure charges for outbound bandwidth. By moving the storage into the same region as the App Service, we avoided inter-region data transfer charges.

What this experience taught me

The cloud is powerful but dangerous if you're not paying attention. Providers make "provisioning" too easy: one click and you have a $300/month machine.

The practices with the most impact:

• Instrument before optimizing: metrics justify the decisions
• Serverless for intermittent workloads: you only pay for what you consume
• Regular right-sizing: needs change, and plans should follow
• Budget alerts: I set up alerts at 80% and 100% of the monthly budget

Cutting from $600 to $135/month freed up budget for other initiatives. And above all, it earned me technical credibility within the team, which made the following projects easier.

Are you working on a similar cloud optimization? Feel free to reach out, I'm always up for a chat about it.