SmartApps Across systems

One process across four systems, one app, and the data stays where it is.

Most processes worth an app touch several systems: the order in the ERP, the account in the CRM, the history in the lake, the ticket in the service desk.

SAP S/4HANApurchase order · vendorSalesforceaccount · opportunitySnowflake12 months of historyServiceNowchange ticketDATA HANDLERS · ONE PER OBJECT · READ IN PLACEdh_podh_vendordh_accountdh_historydh_ticketPROCESS APP · SUPPLIER PRICE CHANGEBuyer reviewsscreen 1Sales confirms impactscreen 2Finance approvesscreen 3IT records the changescreen 4Price updated in SAPwrite-backwrite-back via BAPI · acknowledged · logged with the decision
A supplier price change: the buyer reviews the order in SAP, sales confirms the impact on the account in Salesforce, finance approves against 12 months of history in Snowflake, IT records the change in ServiceNow, and the new price is written to SAP through a BAPI and acknowledged.
4
systems in the worked example, each read in place
illustration
1
app, one log, one process ID
product
0
copies made or synchronised
product
5
data handlers, one per object
illustration
2.1 s
for a question spanning four systems on the showcase estate
showcase estate
Step by step

The supplier price change, screen by screen

Screen 1 · Buyer
Review the order in SAPThe screen reads the open purchase order lines and the vendor from SAP through two table handlers; the buyer sees the proposed price beside the current condition and the last three orders.
Screen 2 · Sales
Confirm the impact in SalesforceThe sales lane reads the account and its open opportunities from Salesforce; the seller marks which quotes carry the affected part and confirms the customer impact.
Screen 3 · Finance
Approve against the history in SnowflakeThe controller sees 12 months of purchase history and margin for the part, read from the conformed tables the Backbone landed, and approves within the band or returns with a note.
Screen 4 · IT
Record the change in ServiceNowA change request is created through the Table API with the app's process ID, so the change calendar and the app refer to each other.
Write-back · SAP
The new price condition is writtenThe app calls the SAP pricing interface with validity dates, stores the returned document number in the log, records the decision in the lake's decision table and closes the instance.
Why it works

Each system stays the owner; the app is where the process happens

Every screen reads through a data handler bound to the system that owns the object, so the buyer sees the SAP order as it is at that moment and the seller sees the Salesforce account as it is. The access rules of each system apply when the handler reads: a seller sees their own accounts, a buyer their own purchasing groups. The app adds the process on top: who acts, in which order, with what criteria, within which clocks.

When the flow ends, the write goes to the system that owns the object, through its released interface, and the acknowledgement is stored. The decision is also recorded in the lake beside the history it was made on, so an analyst can ask next quarter how many price changes were approved and against what margin.

Building it takes the same steps as any app: five handlers registered from the catalog, four lanes, four screens drafted from the handlers, five criteria, two clocks. The catalogue estimates 4 to 6 working days for a production-grade version.

Handlersdh_po · dh_vendor · dh_account · dh_history · dh_ticket5
Lanesbuyer · sales · finance · IT4
Screensone per lane, drafted from the handlers4
Criteriastart, three arrows, one band5
Clockssales 1 day · finance 2 days2
Write-backSAP pricing interface · Snowflake MERGE · ServiceNow Table API3
Patterns

Four shapes a cross-system app takes

ERP plus CRM

Quotes, credit limits, complaints and returns: the customer and the opportunity in Salesforce, the order, the invoice and the credit exposure in the ERP. Customer credit limit change and Quote discount approval are built this way.

ERP plus lake

Reviews on history: excess stock, supplier scores, budget variance. The decision needs a year of data, read from the lake, and the action goes to the ERP. Excess and obsolete inventory review and Supplier scorecard and corrective action.

ERP plus service desk

Changes and incidents: the equipment in the ERP, the change or the incident in ServiceNow, the maintenance order back in the ERP. Health, safety and environment incident and Change advisory board.

Two ERPs

Two divisions, two ERPs, one process: a shortage room or a material creation across SAP and Oracle, with a handler per object per system and one screen showing both. Critical component shortage room.

Terms on this page

Read in placeThe rows stay in the system that owns them. The app sends the filter and the columns it needs, and the system returns the matching rows.
Write-backA change made by the app in the system of record through that system's released interface, with the acknowledgement stored beside the decision.
Data handlerA named connection from an app to rows in a system: a federated table, a saved query, a pipeline result, an API or a file. Screens, expressions and decision points use it to read and write.
Decision logThe append-only record of every step, value and decision in every app instance, with the person, the time and the acknowledgement of any write-back.
Questions

What if two systems disagree?

The screen shows both values with their sources, and a rule can decide which one the process uses, or route the case to a person. The Master Data capabilities of the platform resolve the disagreement at the record level when it is a master-data problem.

How are secrets and identities handled?

Each connector holds its credentials in the vault; each write runs under the acting person's identity where the system supports it, and under the integration user otherwise, with the person's name in the log.

Can the process start from an event in one system?

Yes: a Change Data Capture message from Salesforce, a business event from Fusion, a goods receipt in SAP or a row landing in the lake can be the start criterion.

Does the app slow the systems down?

A screen is a filtered query on indexed columns, the same load as a person opening a transaction. History at scale is read from the lake, where it belongs.

See a SmartApp built on your data

Bring one process and the systems it touches. In the first session we register the data handlers, draw the levels or the lanes and run the first screen on your live rows.