We’re in the middle of a vendor evaluation for our global ERP consolidation project and the AI capabilities angle has become unexpectedly important. Twelve months ago this was a secondary consideration. Now the board wants to understand what AI we’re actually buying, not just what’s on the roadmap slides.
We’ve had briefings from all three major vendors: SAP S/4HANA with Joule, Oracle Fusion Cloud with embedded OCI GenAI, and Microsoft D365 Finance with Copilot. All three demos looked impressive. But demos aren’t production.
I’d like to hear from practitioners who are actually running any of these in live environments in 2026—not pilots, not proofs of concept. Production. What is actually working and what’s still in ‘coming soon’ territory? Specific areas: anomaly detection in financial transactions, intelligent demand forecasting, AI-assisted procurement, and natural language querying of ERP data.
Microsoft Copilot in D365 Finance & Operations — Production Reality Mid-2026
Starting with what’s demonstrably in production on the Microsoft side, since that’s this forum’s lane, then flagging the cross-vendor comparison points worth pressure-testing.
What’s shipping and used in production environments:
Natural language querying via Copilot in D365 Finance is the most mature capability. The Copilot in Finance feature set — surfaced through the Finance insights workspace and inline chat in the UI — allows NL queries against ledger data, vendor balances, and cash flow summaries. This works against your own tenant data via Azure OpenAI Service with data residency controls. Production deployments are running this. Caveat: query depth is constrained by what’s exposed through Dataverse virtual tables and the Finance data model — complex cross-entity joins still fall back to Power BI or FO reports.
AI-assisted procurement via Copilot in Dynamics 365 Supply Chain Management includes PO draft generation, vendor response summarization, and contract clause flagging. The PO assistance features are GA (verify current GA status in your version/region). The contract intelligence pieces vary by whether you’ve licensed Microsoft Syntex alongside D365.
Anomaly detection in financial transactions is delivered through Finance insights — specifically Customer payment predictions and Cash flow forecasting — which use ML models trained on your transaction history. These have been in GA for multiple release waves. True anomaly/fraud-signal detection at the transaction line level is more limited; this overlaps with what Microsoft Sentinel or Compliance Manager handles rather than D365 Finance natively.
Intelligent demand forecasting uses Azure Machine Learning pipelines surfaced through Master planning / Planning Optimization. Production deployments exist, but model tuning requires your supply chain team to own the feature engineering — it does not self-configure.
Cross-vendor pressure-test questions for your evaluation:
Ask each vendor for referenceable production customers in your industry vertical, not just logos on slides
Clarify data egress and model training boundaries — is your transaction data used to train shared models?
For Joule and OCI GenAI, establish which capabilities require BTP or OCI AI services add-on licensing versus what’s included in core ERP
The demo-to-production gap is largest in anomaly detection and NL querying against complex schemas across all three vendors — that’s where your evaluation scrutiny should concentrate.
This draft is based on general ms-dynamics-365 knowledge. It has not been verified against your specific version and environment. Practitioners: verify the steps and share your experience below.
Running SAP S/4HANA with Joule in production for about seven months now. The natural language interface is genuinely useful for querying ERP data—finance team members who never used the standard reports are pulling their own analysis now. The weak spots are financial anomaly detection (it surfaces obvious outliers but misses subtler patterns that experienced reconcilers would catch) and anything involving complex multi-entity queries. The Joule roadmap shows improvement in both areas, but what we have today is better described as smart search than true AI reasoning.
We’re on Oracle Fusion Cloud with the embedded OCI GenAI capabilities. The demand planning AI—Oracle Demand Management with AI-driven demand sensing—has been in production for about a year and the results are genuine. Forecast accuracy improved about twelve percentage points over our previous statistical model. Where Oracle is weaker: the Oracle Digital Assistant natural language interface feels like it’s still catching up to what SAP Joule offers, and the anomaly detection in AP has too many false positives at our transaction volume.
D365 Finance with Microsoft Copilot here. The integration is better than I expected for natural language summarization—I can ask it to summarize a vendor’s payment history, explain a reconciling difference, or draft a collections email based on an aging report. It’s uneven: good on structured data queries, shakier on anything requiring business context. Where I’d warn you: the hallucination rate on anything involving formulas or calculated fields is higher than I’d want for finance use cases. We had it confidently explain a VAT calculation that was just wrong. Build human review into any copilot output that feeds a decision.
The honest answer on all three: the embedded AI in major ERP platforms in mid-2026 is best for augmentation, not automation. Natural language querying is the most mature use case across all three vendors. Autonomous decision-making—procurement approvals, automated journal posting, inventory rebalancing without human confirmation—is available but requires extensive configuration and governance investment. If your board is expecting to buy AI capabilities out of the box, set that expectation correctly early. What you’re buying is a foundation that requires your data, your governance, and your implementation investment to actually deliver value.
One dimension that’s rarely in the vendor demo: total cost of AI ownership. The AI features in all three platforms aren’t always included in base license. SAP Joule advanced capabilities are add-on. Oracle AI features are bundled but some require OCI consumption credits. Microsoft Copilot for D365 Finance has its own per-user pricing. One enterprise client I work with was shocked to find that Copilot added 30% to their D365 Finance per-user cost. The AI value proposition needs to survive that math before you commit.