Copilot for Finance is not strictly Generative AI.
Copilot for Finance is not strictly Generative AI.
While the "Generative" part (LLMs like GPT-4) gets the most attention, Copilot for Finance is actually a hybrid system that combines three distinct types of AI technology to work within Dynamics 365 Finance.
1. Generative AI (The "Creative" Layer)
This is what most people recognise. It utilises large language models to handle unstructured data and facilitate human interaction.
- What it does: Drafts collection emails in Outlook, summarises complex workflow histories, and explains "why" a variance occurred in plain English.
- The Benefit: It saves time on manual writing and helps non-technical users understand dense financial data.
2. Analytical & Predictive AI (The "Brain" Layer)
This is "Traditional" AI or Machine Learning. It is deterministic, meaning it follows logic and math rather than "predicting the next word."
- What it does: Scans for anomalies in expense reports, predicts which customers are likely to pay late (Payment Predictions), and identifies patterns in historical data to suggest budget forecasts.
- The Benefit: Financial data requires 100% accuracy. Generative AI can "hallucinate" numbers, but the Analytical layer ensures the math is grounded in the actual SQL database of D365FO.
3. Agentic / Autonomous AI (The "Execution" Layer)
As seen with the Account Reconciliation Agent, this layer focuses on task completion.
- What it does: It uses a set of instructions to proactively monitor the system. It doesn't just wait for you to ask a question; it "patrols" your ledgers and flags issues.
- The Benefit: It moves the software from being a "passive tool" (like a calculator) to an "active participant" (like a digital employee).
How they work together in a real scenario:
Imagine a Subledger Discrepancy (eg. Account Reconciliation Agent):
- Analytical AI detects that the AP subledger and GL are out of balance by $500.
- Generative AI looks at the voucher, sees the vendor name is "Contoso," and writes a summary: "The mismatch is due to a manual journal entry posted to the GL without a corresponding vendor transaction."
- The Agent presents this to you in the workspace with a button that says "Post Correcting Journal."
|
AI Type |
Role |
Example in D365FO |
|
Generative |
Communication &
Summary |
Drafting a
"Notice of Default" letter |
|
Analytical |
Pattern
Recognition |
Finding a
duplicate invoice that humans missed. |
|
Agentic |
Automation |
Running the daily bank
match while you sleep. |

Comments
Post a Comment