Best Practices for Agentifying ERP Workflows
Last updated
Was this helpful?
Last updated
Was this helpful?
Legacy ERP systems like SAP, NetSuite, PeopleSoft, and Oracle EBS are notoriously difficult to integrate with modern automation platforms. GLIK enables a non-invasive, AI-native approach by embedding orchestration logic outside the ERP boundary.
Here are best practices to follow:
Most ERPs generate structured output as:
PDF invoices
CSV exports
Excel-based ledgers
Email-based approvals or exceptions
Use these as the entry point for your GLIK workflow instead of trying to interface directly with the ERP backend.
Blocks like Doc Extractor
and Question Classifier
let you turn PDFs or form exports into actionable signals. This is critical when dealing with:
Expense receipts
Supplier forms
Procurement logs
Instead of hardcoding policy logic, define your:
Thresholds
Blacklisted vendors
Approval hierarchies
in a Markdown or CSV file, then use Knowledge Retrieval
to pull and apply them dynamically.
This makes your workflows portable across departments — even if each one has a different policy schema.
Where ERPs traditionally rely on multi-step human review flows:
Use Loop
and IF/ELSE
blocks to simulate those steps
Add Agent
blocks to represent managers, reviewers, or compliance units
Use Template
or HTTP Request
blocks to output approvals or escalate issues via email or Slack
ERP replacements often fail because they lose visibility. Use:
Variable Aggregator
Code
block
Save Point Agent
(or custom Template → JSON output)
to generate a complete trail of decisions, timestamps, and policy sources.
Integration method
UI scripting, screen scraping
Agent-based logic via files, forms, memory, LLM
Deployment speed
Weeks/months
Hours/days
System risk
Modifies fragile interfaces
Wraps around existing systems (non-invasive)
Automation logic
Rules, flows, macros
LLMs + agents + knowledge-based policy execution
Resilience
Breaks on UI changes or updates
Abstracted from UI or code-level changes
Memory & policy reuse
Siloed automations
Scoped variables, reusable policy logic
Escalation path
Hard-coded, rarely dynamic
Agent delegation + human-in-the-loop compatible