At Boyer & Associates, customer conversations about AI are moving beyond enabling Copilot features and toward something more operational: retrieval-first AI agents layered over ERP data that answer business questions inside Microsoft 365.
In a recent episode of PartnerTalks, Lee Witkop, CEO of the Minnesota-based Dynamics Partner, said most mid-market customers are not asking for autonomous systems. They are asking how to apply AI to governed ERP and CRM data without expanding audit risk.
In practice, that has translated into retrieval-first AI agents. These tools surface information from systems such as Dynamics 365 Business Central inside Teams or Outlook. They retrieve data a user is already authorized to access, but do not post transactions or modify records.
Boyer has documented such a deployment alongside Business Central at ImageTrend, a SaaS provider serving emergency healthcare organizations. In a published ImageTrend case study, the Partner outlines how staff can query open invoices and account balances from within Microsoft 365, rather than logging in directly to the ERP system.
Becky Kingsberry, Assistant Controller at ImageTrend, said the objective was to reduce routine interruptions to the finance team while limiting direct ERP access for non-finance users.
Retrieval-first AI agents before autonomy
Witkop argues ERP environments can provide a structured starting point because financial systems typically operate under tighter governance and role definitions. That does not mean ERP data is inherently cleaner than CRM data, but it often carries clearer audit expectations.
Agents do not correct flawed data. Instead, they expose it.
Whether a human or an agent retrieves a record, the underlying information remains unchanged.
Deployment may require:
- Master data remediation aligned to defined agent queries
- Security role review and least-privilege enforcement
- Workflow normalization in quote-to-cash and procure-to-pay
- Governance definitions specifying what the agent can access and how outputs are validated
A practical foundation for Frontier positioning
Microsoft’s Frontier AI narrative emphasizes multi-agent orchestration and broader automation. Retrieval-first deployments can serve as groundwork for that ambition. Before orchestration becomes viable, data governance, permission discipline and process clarity must be established.
Rather than positioning agents as one-off AI projects, Witkop sees them fitting into managed services models that include ongoing accuracy testing, permission audits and configuration oversight.
For Partners evaluating Frontier positioning, the best bet might be to start with controlled retrieval over governed ERP cores and recurring services built around maintaining those boundaries.
