The SaaS playbook built your finance stack. Finance teams spent a decade assembling best-of-breed tools. One for accounts receivable. Another for accounts payable, treasury, FP&A, and so on. The software era rewarded this specialization: find a workflow, own it, expand from there. Point solutions won by going deep on single workflows.
Building billing systems for finance teams requires a certain perspective. That perspective shifts when you also build enterprise AI solutions for the same departments. The change in viewpoint affects how the existing stack is assembled. Decades ago, structured integrations were necessary because systems could not read. An invoice was opaque to software. Developers built APIs, schemas, and data pipelines to translate documents into something machines could process. The best point solutions were those that built the best translation layer for their workflow.
Today, AI can read the invoice all by itself. No need for a translation layer. That changes everything about how the stack should be assembled. The industry is moving from APIs to agents. The goal is no longer just connecting systems, but allowing software to understand the business logic across them.
Four Categories of Agents
Finance organizations need four agent categories to function effectively. Assist agents draft collection emails, summarize payment history, and pull aging reports. They are useful, but human bottlenecks remain. Automate agents process invoices, match purchase orders, code to the general ledger, and queue for payment without a human in the loop. These agents handle the routine, repetitive tasks that slow down operations.
Related: AmEx and Apple Pay Miss Gas Savings
Advise agents surface what nobody was looking for. They identify vendors whose pricing is quietly drifting, customers whose payment patterns predict churn six months out, or expense categories outpacing the revenue they support. Configure agents go further, observing approval workflow bottlenecks and restructuring them. They can also restructure vendor-tiering based on spend patterns that no longer reflect the business. Most teams never get to Advise and Configure. The problem is architecture.
Advise agents that only see your AR platform cannot connect a customer’s payment slowdown with declining order frequency and a change in their AP contact. They can only see within one system. Configure agents that only touch your invoice workflow cannot restructure the approval hierarchy that spans finance, procurement, and operations. They remain bounded by the same silos their SaaS predecessors were.
The value proposition of Advise and Configure agents depends on context that crosses the boundaries of existing point solutions. The insight that a vendor’s pricing is drifting requires visibility across every invoice, every contract, and every historical negotiation. That data does not live in one system. It lives in five.
The instinct to restrict data access across systems is legitimate. Finance carries real regulatory exposure, and even model behavior introduces risks that blunt access controls can’t fully address. But the answer isn’t fragmentation as a proxy for governance. It’s policy-based access with full observability. Agents must operate under defined permissions, with every action logged and every decision auditable. That’s a harder architecture to build than a firewall between systems. But it’s the only one that actually scales.
Related: How Christmas parties help to improve your work environment
This is why rebundling is happening across enterprise software. Companies winning with AI recognize that owning one workflow means owning none of them. The AI that creates real value needs to see the full picture. Foundation-model companies understand this better than anyone. They are not selling APIs any more. They are selling trust, and they’re accumulating the cross-workflow context that makes that trust deserved.
That’s the AI growth playbook, and it runs directly counter to how finance teams have been buying software for the past decade. If you’re a finance leader, the individual tools in your stack are probably fine. Your liability is the absence of a context layer connecting them. An Automate agent processing invoices in one system, an Advise agent watching vendor patterns in another, a Configure agent restructuring approval workflows in a third—these agents need to share context to work as a system.
Most finance leaders sense this. Their tools are good, but the results are underwhelming. The gap isn’t motivation or budget. It’s architecture. Fragmented systems share exports, not context. And agents without context can only Assist.
Leave a Reply