Nearshore ERP support: the AI-augmented model
Nearshore ERP support — managed L2 and L3 services for stacks like SAP, Oracle, NetSuite, Dynamics 365 and Salesforce — is in the middle of a quiet reset. AI agents now handle the work that used to consume the first hour of every ticket: classification, runbook retrieval, regression checks, and draft replies. Done right, this cuts manual hours 40–60% and lifts response SLAs at the same time. Done wrong, it produces confident-sounding nonsense that wastes more time than it saves. This essay maps where AI belongs and where humans stay in the loop.
The AI-Augmented ERP Support Model
We organize AI in ERP support across four layers, each with a clear human checkpoint.
1. Intake
An agent reads incoming tickets, classifies by module and severity, deduplicates against open incidents, and pulls the relevant runbook section. Humans approve any P1 routing.
2. Diagnosis
Retrieval over your config, transports, change history, and prior tickets surfaces likely causes. The agent drafts a hypothesis. A senior consultant accepts, edits, or discards.
3. Action
Routine actions (password resets, role assignments, master-data corrections inside a guardrailed schema) execute via approved workflows. Anything that touches financials, security, or production transport requires a human.
4. Regression
Before any change ships, an eval harness runs scripted regressions against a sandbox. AI summarizes the diff and any failures. Humans sign off.
What you can expect numerically
- First-response time: down 50–70%.
- Mean-time-to-resolve on P3/P4: down 30–50%.
- L2 hours per ticket: down 40–60%.
- Net SLA improvement, headcount roughly flat.
Where AI does not belong
Anything irreversible without a human checkpoint. Anything touching close-of-period, payroll, or customer-facing financials. And anywhere a wrong answer carries audit risk — AI drafts, humans sign.