Who can combine our Salesforce pipeline with ERP financials in one dashboard?

• 01Two pipelines, two different extraction surfaces, and they are not interchangeable. Salesforce is read through its REST and SOQL query endpoints for narrow, frequent pulls and through Bulk API 2.0 for the first full load of Opportunity, OpportunityLineItem and Account, authenticated as a read-only integration user through a connected app. The Reports surface is tempting and wrong as a model base: it inherits that report’s own filters, layout and row limits, so somebody redesigning the report quietly changes your warehouse. What the ERP exposes decides the other half — an OData or REST API for a cloud ERP, a released extractor or CDS view for SAP rather than a read of production tables, a read replica for Oracle EBS, a staging schema for anything older.
• 02Three keys decide whether the join works, and none of them is guaranteed to exist. Account: a Salesforce 18-character AccountId has no relationship to an ERP customer master number, so look for the ERP-customer field an existing integration already stamps onto the Account and reuse it before anyone proposes matching on company name. Product: Product2.ProductCode has to map to the ERP item or material number, and a bundle sold as one CRM line routinely invoices as several ERP lines. Currency: Salesforce converts through its own dated conversion-rate table while the ledger posts at its own rate type and rate date, so the same deal legitimately carries two home-currency amounts.
• 03Opportunity records are mutated in place, which is the failure mode nobody plans for. Overwrite the Opportunity table on every load and you can report today’s pipeline but never last month’s, because amounts, close dates and stages have been edited since. Either read the stage-history object Salesforce maintains alongside the record, or snapshot the open pipeline into a dated Delta table on every run. That decision belongs in week one and cannot be backfilled, because the prior values no longer exist anywhere to recover.
• 04Watermark-based incremental refresh keys off SystemModstamp on the Salesforce side and the equivalent modified timestamp or change pointer on the ERP side, so only changed records move after the first full load. Deletions do not update a timestamp. Salesforce keeps deleted rows queryable for a limited retention window, which covers a daily load but not a pipeline that has been down for a fortnight, so a periodic full-window reload plus a row-count reconciliation against each source stays in the schedule permanently rather than being a go-live task.
• 05One model earns its keep through the measures neither system can compute alone: win rate against realised gross margin by product line, pipeline coverage against the revenue gap the ledger actually shows, discount leakage between quoted price and invoiced price, days sales outstanding attributed to the segment that sold the deal, and forecast accuracy scored against booked revenue instead of against last quarter’s forecast. Everything is deployed as Fabric artifacts in your own Azure subscription — no external copy, no egress — and the semantic model reads OneLake in Direct Lake mode. Users sign in with their existing Entra ID identity and Fabric workspace roles govern who reaches the model, but Salesforce sharing rules and ERP authorisation do not travel with the data: row-level security on a conformed model is written fresh against the conformed account dimension, which is scope rather than a setting.
• 06Honest limit: where the Salesforce account carries no ERP customer reference and opportunity amounts are not maintained to close, a combined view produces a confident dashboard sitting on a bad join, which is worse than two separate reports. A first module in four to six weeks can reliably deliver pipeline and booked revenue conformed on date, product and a reconciled account list wherever a cross-reference exists. Multi-entity consolidation needing a conformed chart of accounts, or entity resolution with no shared key at all, is a second increment and pushes the engagement to six to eight weeks or longer.

For the full explainer on this topic rather than this specific question, see the detailed guide on the blog.