H&R Block - Enterprise Tax Platform Running on Azure

H&R Block moved most of their infrastructure to Azure over several years while keeping tax season running. Migrating infrastructure during tax season is like changing the engine on a plane while it's flying. Their ops team probably aged 10 years in two.

What They Actually Run On

Everything is .NET on Azure because their VP of AI Platforms Aditya Thadani decided "Everything we're building now is in Microsoft Azure" and now they're locked into the Microsoft ecosystem for the foreseeable future.

The Azure Services That Matter:

• Azure OpenAI - The AI chatbot that knows more tax law than most CPAs (usually)
• Azure Cosmos DB - Stores all the tax data across the planet because distributed databases are fun until they're not
• Azure App Service - Runs the web apps that 69 million people use to file taxes
• Azure AI Document Intelligence - OCR that actually reads tax documents instead of puking garbage like most OCR systems
• Azure AI Content Safety - Makes sure the AI doesn't suggest deducting your entire mortgage as a business expense

AI Tax Assist: The Chatbot That Actually Works (Sometimes)

AI Tax Assist launched in December 2023 and it's surprisingly not terrible. Unlike most corporate AI chatbots that give you the digital equivalent of shrugging, this one was trained on H&R Block's Tax Institute knowledge base and decades of actual tax returns.

What It Actually Does:

• Response times are usually fast, except when the system decides to have an existential crisis during peak hours
• Handles basic tax questions without making you want to throw your laptop
• Sometimes gives weird answers about edge cases but hey, so do human tax preparers
• Integrates with their tax prep software instead of being a useless sidebar widget
• Doesn't leak your financial data (as far as we know)

The AI handles four types of questions: "What the hell is this form?", "How do I claim this deduction?", "Did the IRS change something again?", and "Please help me not go to jail." Microsoft's case study brags about processing 'thousands of conversations simultaneously' - like that's not literally what we pay servers to do.

The Great Data Migration That Somehow Didn't Kill Everything

H&R Block's data platform overhaul was basically "let's throw out 30 years of legacy systems and hope nothing explodes." Their CIO Alan Lowden diplomatically called it needing "buy-in from engineers and executives," which translates to "we argued for months about whether this would work."

The migration probably went something like this: January 2019, some executive said "we need to be cloud-first by 2022." Cue two years of planning, vendor negotiations, and engineers explaining why you can't just lift-and-shift a 30-year-old mainframe system. Then came the fun part: migrating live tax data during off-season while praying nothing broke before April.

The platform now handles 150 million tax returns annually across 12,000 locations. That's a lot of financial data flying around, and somehow they manage to not lose it all.

Real Performance Numbers:

• Peak season = January-April when everyone panics and files at once (46% file DIY)
• 60,000+ tax preparers all hitting the system simultaneously
• Nearly always up during tax season (except when it isn't), which means it goes down for exactly the wrong 2 hours on April 15th
• Auto-scaling that actually works (most of the time)

The architecture is typical microservices hell, but at least it doesn't crash when everyone files their taxes on the last day. Docker Desktop randomly stops working on M1 Macs and nobody knows why, but Azure Kubernetes Service handles container orchestration so individual developer laptop problems don't tank the entire platform.

The security and compliance stuff is handled properly because getting audited by the IRS is nobody's idea of fun, and H&R Block's legal department doesn't mess around either.

How 60,000+ Tax Preparers Don't Murder Each Other's Systems

H&R Block's tech runs 12,000 tax offices worldwide, which means 60,000+ people are all hammering the same infrastructure during tax season. The system gives preparers access to forms, catches obvious screw-ups before filing, and tracks client info. It's basically enterprise tax software that doesn't completely suck.

What actually works:

• Client file sharing that doesn't break when someone uploads a 50MB PDF of their receipts
• Multi-office oversight so corporate can see which locations are drowning during April
• Error checking that catches the obvious stuff before the IRS rejects your filing
• Audit trails because the government wants to know who touched what data when
• Training modules for seasonal workers who learned tax prep three weeks ago

Peak season means January through April when everyone panics about taxes. During 2024, they somehow kept the system running while thousands of preparers all tried to file returns simultaneously. Database queries stayed under a second most of the time, which is better than most enterprise software manages.

The great password reset incident of March 2024 when half the tax preparers couldn't log in for 4 hours right during peak filing season. Then there was that time Azure hit capacity limits in us-east-1 and everyone's filing queue backed up. The microservice that started memory leaking and took down document uploads for an entire afternoon probably cost them thousands of pissed-off customers.

Business Tax Software: For When Your LLC Actually Makes Money

H&R Block has small business tax software for business owners who don't want to pay a CPA $300/hour but also can't figure out Schedule C on their own. The software handles most business scenarios without completely screwing up your taxes:

Business types it won't destroy:

• Schedule C - Solo freelancers with income from 47 different clients
• Form 1065 - Partnerships where everyone argues about K-1 distributions
• Form 1120S - S-Corps that elected that status to save on payroll taxes
• Form 1120 - C-Corps with actual employees and complicated deductions
• Multi-state chaos - When your business operates in states with different tax rules

They partnered with Wave Accounting so you can import your bookkeeping data instead of typing everything manually. In practice, the Wave integration works great until Wave randomly categorizes your office supplies as 'Entertainment Expenses' and you spend two hours fixing it. The API integrations are solid but good luck getting enterprise-level access unless you're paying mid-six figures.

APIs: Good Luck Getting Access Unless You're Paying Big Money

H&R Block has APIs buried somewhere in their enterprise portal, but the docs were clearly written by someone who's never actually used them. It's like IKEA instructions translated through Google Translate twice. They'll talk to you if you're a big enterprise client spending six figures annually through their business services division. Everyone else gets file exports and manual data entry.

Who actually gets API access:

• Payroll companies - ADP and Paychex need to generate W-2s and 1099s
• HR platforms - Corporate systems that distribute tax documents to employees
• Big accounting software - Not your QuickBooks, we're talking enterprise-level integrations
• Banks - Financial institutions offering tax prep to their customers

The APIs use OAuth 2.0 and return JSON, so at least they didn't invent their own data format. They have sandbox environments for testing, which is nice because nobody wants to debug API calls with live tax data.

Scaling Architecture: When 69 Million People File Taxes at Once

H&R Block moved to Azure Kubernetes Service because scaling their old infrastructure was like trying to add more lanes to a highway during rush hour. The new system handles seasonal chaos through several layers of "please don't break":

How they keep it running:

• Microservices - Individual pieces can die without killing everything else (Martin Fowler's microservices guide)
• Docker containers - Spin up more instances when usage spikes, pray they don't all crash (Docker architecture docs)
• CDN - Tax forms served from Azure CDN so downloads don't crawl during peak season
• Database sharding - Split user data across multiple databases because one database can't handle 69 million users (Azure SQL sharding patterns)
• Redis caching - Frequently used forms cached in memory so the database doesn't scream (Redis documentation)

Azure Application Insights monitors everything, which means they can see the system dying in real-time and maybe fix it before taxpayers notice. Of course, monitoring doesn't prevent the inevitable April 15th meltdown when 10 million people simultaneously realize they forgot to file their taxes.

The auto-scaling works great until it doesn't - like when Azure hits resource limits in their data centers, or when some microservice starts memory leaking and consuming all available RAM before anyone notices. Then it's 2 AM and some poor ops engineer is getting paged to restart containers while tax preparers across the country can't access the filing system. The .NET migration broke their custom forms builder and nobody noticed for two weeks until customers started complaining about missing deduction fields.

Security: All the Checkboxes Because the IRS Doesn't Mess Around

H&R Block has all the security checkboxes filled because getting audited by the IRS is nobody's idea of fun. Their platform is locked down tight with enough compliance acronyms to make a lawyer happy:

Security stuff that actually matters:

• Everything encrypted because leaving financial data in plain text is a career-ending move
• Azure Key Vault for storing secrets (no more passwords in config files)
• Multi-factor authentication for admin access (because someone will eventually click a phishing email)
• Regular penetration testing to find holes before the bad guys do
• SOC 2 Type II certification to satisfy enterprise customers who ask for it

Compliance alphabet soup:

• IRS Publication 1075 - Federal tax information handling requirements
• CCPA/GDPR - Privacy regulations for different states/countries
• GLBA/FFIEC - Financial industry regulations that nobody understands but everyone requires
• PCI DSS - Payment processing security standards

They use Azure AI Content Safety to prevent their AI from suggesting you deduct your entire mortgage as a business expense or accidentally leaking someone else's tax data. The Azure OpenAI rate limiting becomes a problem when 10,000 people ask about crypto taxes simultaneously, and the Document Intelligence API occasionally returns confidence scores of 99.9% for documents that are clearly pictures of cats. But they haven't been publicly hacked yet, which is better than most companies can claim.