Welcome Erika! Great to have you here, and sounds like you’re on a fantastic career path. My videos on here and my YouTube videos focus on financial data in SQL, Power Query and Power BI which I believe is a strong foundation to build on. Good luck with your projects!

@Erika Estrada Great questions Erika — and honestly, locked-down data access is more common than people expect. When you can't access the database directly, the skill that pays off isn't live querying — it's building a reliable reconciliation layer on whatever you CAN export: 1. Get fluent in Power Query first. It connects straight to CSV/Excel exports, cleans them, and lets you reconcile two sources on a repeatable refresh (e.g. sub-ledger vs GL). That's most of the value with zero DB access, and it's a fast win. 2. Master the GL flow before the variances. You can't reconcile what you don't understand — so learn how AR/AP transactions actually post and roll up into the GL. Inventory valuation, unbilled receipts (GRNI), PPV and WIP variances are cost-accounting territory, great to understand eventually, but the transaction flow comes first. 3. Use SQL/BigQuery on the exports you already have. Practice joins, grouping, and matching a sub-ledger to a GL balance. The logic is identical the day fuller access opens up — you just point it at a bigger source. My honest advice : start where you already own the data — AR/AP — and get ONE reconciliation fully automated end to end (export → clean → match → flag differences). That single build will teach you more than ten tutorials, and it's something you can show your team without needing anyone to unlock the ERP for you.

@Anna Nova great question — and you're right that nothing is ever 100% guaranteed. What I meant is keeping AI inside your company's own secure environment rather than pasting data into consumer tools. The practical baseline most finance teams work to: 1. Keep data in your own tenant — e.g. M365 Copilot's enterprise data protection means your prompts and data aren't used to train the model. 2. Least-privilege access — the AI only sees what that user is already allowed to see. 3. Governance & logging — approvals, audit trails, no client data leaving the environment. Connecting AI to data via MCP isn't automatically "secure" — there have already been real vulnerabilities out there. It's only as safe as it's configured correctly - you're right to stay sceptical.