What the agents handle
| Workflow | Systems | What the agent does |
|---|---|---|
| Order-to-cash exceptions | Salesforce, SAP | Reconciles opportunity, order and billing data; explains mismatches; proposes corrections for approval |
| Customer case resolution | ServiceNow or Salesforce Service Cloud, SAP | Gathers order, delivery and invoice context, drafts the resolution and updates the case |
| IT and HR service requests | ServiceNow | Classifies and routes tickets, fulfils standard requests, and escalates the rest with a summary |
| Master data upkeep | Salesforce, SAP | Detects duplicates and inconsistent records, proposes merges and fixes for data stewards to approve |
| Account and renewal preparation | Salesforce, SAP, ServiceNow | Compiles usage, open incidents and billing status into an account brief |
How agents connect to each platform
- Salesforce: REST and Bulk APIs or Platform Events, through a connected app and integration user with a permission set limited to the objects and fields the agent needs.
- SAP: OData services or BAPIs/RFCs, usually through SAP Integration Suite or your existing middleware, with a technical user holding restricted authorisation roles.
- ServiceNow: Table and Import Set APIs or IntegrationHub, with a dedicated integration account and scoped roles.
The audit trail
Every agent action is logged append-only with the work item, the input, the records retrieved, the model and prompt version, the tool call and parameters, validation results, the target system's response and any human approval. Entries are keyed to Salesforce record IDs, SAP document numbers and ServiceNow sys_ids, so they reconcile with Salesforce field history and Setup Audit Trail, SAP change documents and ServiceNow sys_audit.
Controls
- Read-only first: agents start with read access and gain write access per workflow once shadow-mode results are reviewed.
- Deterministic validation of every write before it is sent.
- Approval gates by value, risk or record type, with evidence attached for the approver.
- Rate and scope limits per agent and per tool.
- Versioned prompts, models and tools with rollback.
- Data handling rules that keep regulated data inside your environment where required.
How an engagement runs
We start with one cross-system workflow that has a measurable baseline, such as case handling time or order exception volume. You get a working prototype on your data within 3 to 5 weeks and production within about 60 days, moving from shadow mode to approval-gated actions. You own the code and models; work is scoped as fixed-price milestones.
Frequently asked questions
Do you use Salesforce Agentforce or ServiceNow's built-in AI agents?
Where a platform's native agents cover the workflow well, we configure and extend them. When work spans several systems or needs custom validation and audit controls, we build agents that sit across the platforms and call each through its APIs. We recommend based on the workflow, not a preferred product.
Can agents write to SAP?
Yes, through OData services or BAPIs with a technical user whose authorisations cover only the transactions the workflow needs. Writes are validated first and can require approval by document type or value.
How do auditors see what an agent did?
Through the agent audit log, which records inputs, retrieved context, model version, tool calls, validations and approvals, keyed to record and document IDs so it can be matched with each platform's own change history.
Where does our data go?
Data stays in your systems, and agents retrieve only what a step needs. For sensitive workloads we can run models in your cloud tenancy or on private infrastructure so records never leave your environment.
What if one of the platforms is unavailable?
Agents queue work and retry with backoff, never partially applying a multi-system change. Workflows are designed so each step can be retried safely or rolled back.