CRM and ERP AI integration engineer: what the job involves

What does the work look like day to day?

Most of the job is building safe, well-described tools an AI agent can call. In Salesforce, that often means an Apex class exposed as an invocable method. Salesforce's developer guidance says the agent's reasoning engine plans using the labels and descriptions on those methods and their inputs and outputs, so vague naming leads to wrong tool choices. You write those descriptions as carefully as the code, and you keep the Apex and the action configuration in sync.

The other direction matters too. Salesforce now lets developers trigger an active Agentforce agent from Apex, from flows, and from external systems over REST. That means an ERP event, such as a delayed shipment, can start an agent in the CRM. On the Microsoft side, Dataverse can act as a Model Context Protocol (MCP) server, giving agents tools to search, create and update records in the tables that sit under Dynamics 365.

Where SAP and NetSuite fit

ERP projects lean harder on integration plumbing. A Databricks posting for a senior SAP developer gives a clear picture: build extensions on SAP BTP that keep the S/4HANA core clean, connect SAP and non-SAP systems through SAP Integration Suite, OData v4, events and change-data feeds, and add agentic features to finance workflows with tool calling and human-in-the-loop controls. The same ad asks for governance and auditability, which is typical of finance work.

Oracle offers a NetSuite AI Connector Service that supports MCP, so AI assistants can reach NetSuite data through a standard protocol. Connecting NetSuite to an outside AI platform usually starts with an integration record in NetSuite and OAuth credentials. SAP has Joule and Joule Studio for building agents, and n8n has announced a managed environment inside Joule Studio for visual AI workflows.

Approvals, audit trails and clean records

The hard part is rarely getting an agent to talk to a CRM. It is making sure it only does what the user is allowed to do and leaves a trace. Salesforce advises running agent Apex with the with sharing keyword and in user mode so object and field permissions apply, verifying users before private actions, and limiting fields returned by public actions. Microsoft's Dataverse MCP server lists record deletion as allowed only after explicit user approval.

Record quality is the other half. When an agent writes to an ERP, bad data mapping spreads into invoices, inventory and reports. Expect to spend real time on field mapping, deduplication, partial-failure handling, and returning clear error messages the agent and the user can act on. Salesforce's examples use partial saves with per-record results for exactly this reason.

Who hires, and how CRM and ERP specialists move in

Employers include companies running these platforms internally, software vendors building packaged agent features, consultancies and systems integrators serving SAP and Salesforce customers, and data platform companies that integrate with ERP, as the Databricks SAP role shows. Microsoft and SAP have also announced integration between Microsoft 365 Copilot and Joule for SAP customers, which creates partner and client project work.

If you already work as a Salesforce developer, SAP ABAP or BTP developer, NetSuite SuiteScript developer or Dynamics 365 consultant, you hold the scarce half of this job: knowing the data model and business process. Add the AI half by building one agent action end to end in your own platform, with permissions, error handling and tests. Then learn tool calling, retrieval and evaluation well enough to explain why the agent picked the wrong tool and how you fixed it.

Pay

United States, 25th to 75th percentile (BLS software developers)

$108,500–177,400

United States median (BLS software developers)

$140,200

California, 25th to 75th percentile

$141,300–223,500

Washington, 25th to 75th percentile

$136,400–218,500

New York, 25th to 75th percentile

$131,200–212,000

Texas, 25th to 75th percentile

$106,600–167,600

USD a year, before tax. US figures from BLS OEWS May 2025 for software developers, moved to 2026 with the Employment Cost Index. No official series covers CRM and ERP AI integration work: this is the nearest series and it covers all software developers, not AI or integration specialists alone. Location, platform depth and seniority move the figure. Refer to the AI Integration Engineers salary guide for figures from live adverts.

What you need

Deep knowledge of one platformKnow the data model, permissions and extension points of Salesforce, SAP, NetSuite or Dynamics 365. Employers like Databricks ask for years of hands-on SAP development for senior roles.
Platform codeApex and flows for Salesforce, ABAP, CDS and BTP for SAP, and the matching languages for your other platforms. Agent actions are written in these.
APIs and integration patternsREST, OData, events, change-data capture and OAuth. Most CRM to ERP automation runs through these rather than through the AI layer.
Agent and tool-calling conceptsTool descriptions, retrieval, grounding, MCP and human-in-the-loop design. You need to know why an agent picks a tool and how to constrain it.
Security and audit disciplineEnforce user-level permissions, limit exposed fields, require approval for destructive actions and log what agents change.
Testing and deliveryUnit and integration tests, CI/CD and code review. The Databricks SAP posting lists these alongside transport pipelines for ABAP and BTP work.

How to get there

Years 1-3Master one CRM or ERP platformWork as a developer or technical consultant on Salesforce, SAP, NetSuite or Dynamics 365 and learn its data model, security and integration options.
Years 2-4Own integrationsTake on API, middleware and CRM to ERP sync work, including data mapping, error handling and monitoring.
Months 1-3 of the switchBuild your first agent actionExpose a real business task, such as an order status check or record update, to an agent with proper permissions, descriptions and tests.
Months 3-6Add approvals and auditDesign human approval steps for risky writes and logging that shows what the agent did and why.
Year 4 onwardLead cross-system agent projectsConnect agents across CRM, ERP and data platforms, and own evaluation, governance and rollout with business teams.

Open now on AutomationRoles.ai

SAP NS2 AI Consultant SpecialistSAP IT Business Systeme · Herndon, Virginia

Principal Inbound Product Manager, CRM Voice AIServiceNow · Santa Clara, California

Senior Inbound Product Manager, CRM Voice AIServiceNow · Santa Clara, California

AI Engineer - SAP S/4HANABayer · Heredia, Heredia Province

Research Scientist - Salesforce AI ResearchSalesforce · Palo Alto, California

See all →

Questions people ask

Is this the same as a Salesforce or SAP developer?

It overlaps heavily. The difference is the added AI layer: you design tools an agent calls, control what it can change and check whether it chose correctly. Many people in this work started as platform developers.

Do I need a degree?

No source we read sets a degree requirement for this specialty. The postings we read focus on platform experience, integration skills and AI fluency. BLS lists a bachelor's degree as typical entry education for the broader software developer occupation.

Which certification helps?

A platform certification in your own system shows employers you know it. The Databricks SAP posting lists SAP certification in ABAP, BTP or SAP Build as a preferred qualification. Certifications help you get noticed for a screen, but no certification guarantees you a job.

What is MCP and why does it matter here?

Model Context Protocol is an open protocol for connecting AI applications to data and tools. Microsoft Dataverse can act as an MCP server, and Oracle's NetSuite AI Connector Service supports MCP. That makes protocol and permission design part of the integration job.

How much does it pay?

No source gives a pay figure for this specialty. The nearest BLS series, software developers, has a 2026-adjusted US median of $140,200, with the middle half between $108,500 and $177,400. That series covers all software developers.

What goes wrong most often?

Agents with too-broad permissions, vague tool descriptions and weak data mapping. Those lead to wrong records, exposed fields or duplicate transactions. Strict user-mode access, clear error results and approval gates prevent most of it.

What does a CRM and ERP AI integration engineer do?