According to my company’s recent survey of 550 finance professionals, 77% of finance leaders see “a U.S. recession as likely, possible, or already underway in their sector,” while 65% are pouring 10% or more of their budgets into AI anyway.

In other words, finance teams are being asked to cut costs and invest in AI at the same time—and brace for a downturn while they scale. However, one challenge in scaling AI is that isolated tools don’t give you the intelligence you need to respond, especially when the economy stops cooperating.

Recent McKinsey research puts a number on the gap: Nearly two-thirds of organizations haven’t yet scaled AI across the enterprise, with most stalled pilots failing because they remain “poorly integrated into core processes.”

What often separates the organizations actually moving the needle isn’t how many AI tools they’ve deployed but rather how those tools are built and connected underneath.

Why Single-Purpose AI Tools Aren’t Enough Anymore

Based on my experience in the industry, the first wave of AI in accounts receivable (AR) was about task-level automation: getting humans out of the loop on routine reconciliation, speeding up invoice matching and reducing manual outreach. That work had real value. According to my company’s 2025 report, 82% of the 500 enterprise decision-makers surveyed scaled operations by 11% or more without adding staff once they got foundational automation right.

But task-level automation has a ceiling. Generic AI tools don’t understand the nuance of B2B payment relationships. They don’t know that a customer who’s paid within 15 days for three years but went quiet at day 25 is a different problem than one who chronically pays at day 45 regardless of reminders.

AI only delivers meaningful outcomes in AR when it can see and act on the full order-to-cash picture, because payment behavior, disputes, credit risk and customer communication are all interdependent signals, not isolated events.

In practice, that means connecting data and workflows across invoicing, payments, collections and cash application so decisions are made with context instead of fragmented snapshots.

The Three Layers That Make AR AI Actually Work

In working with finance teams deploying AI at scale, I’ve found that the organizations getting real results have built or deployed systems with three distinct layers working in concert:

The Data Layer

Effective AR AI starts with a unified customer financial profile drawing from every relevant source: ERP data, payment history, bank feeds, dispute logs and communication records.

Many organizations are still operating with this data fragmented across systems that don’t speak to each other, and AI working with incomplete information can’t deliver intelligent recommendations.

The Intelligence Layer

This is where architecture either earns its value or doesn’t. Retrieval-augmented generation (RAG), an approach that lets AI query a company’s own historical data at the moment of decision, is what makes domain-specific intelligence possible.

Rather than relying on a general model’s training, a well-built AR system pulls from years of payment behavior, dispute history and relationship context to surface recommendations grounded in your specific portfolio. AR teams can ask questions of their own data in plain language, without routing everything through a data analyst.

The Action Layer

This is where the field is heading. Agentic AI systems don’t just surface insights; they execute workflows within defined guardrails. One agent handles invoice exceptions, another monitors payment patterns, a third optimizes working capital decisions.

At this stage, though, the design principle matters: AI should act autonomously within defined thresholds, recommend for human approval in gray areas and escalate when signals conflict.

We’re not trying to replace human judgment. We’re trying to get finance teams out of the work that doesn’t need it so they can focus on the work that does.

The Economic Case For Getting The Architecture Right

Where I’ve seen teams succeed, the difference isn’t just investing in AI, it’s being disciplined about how decisions are handed off between people and systems.

Many teams either over-automate too quickly, which erodes trust when exceptions aren’t handled well, or keep humans in every loop, which limits scale. The real value sits in the middle: clearly defined thresholds for when AI can act, recommend or escalate.

Another common misstep is treating AI outputs as static instead of continuously improving them through operator input. The teams seeing the strongest results build feedback loops into workflows so overrides and corrections sharpen future recommendations. They also measure success through outcomes like reduced days sales outstanding, faster dispute resolution and better customer experience, not just activity.

In practice, this is less about deploying a tool and more about redesigning how work gets done. Finance leaders who treat AI as an operating model shift, not a technology purchase, are the ones who consistently translate architecture into results.

What The Architecture Can’t Solve On Its Own

Even with the right stack in place, 89% of finance leaders believe they won’t fully capitalize on AI until their teams shift how they think about the work itself.

The organizations extracting the most value have redesigned workflows around what AI makes possible. They’ve defined clear decision boundaries and built feedback loops, so specialist overrides improve the system over time.

And the people doing their best on AI-augmented AR teams aren’t the most technical ones. They’re the ones who can look at what the system is flagging and say “that doesn’t match what I know about this customer”—and then explain why.

The goal is to move from reactive reporting to forward-looking, context-aware decision-making. AI built on AR data should learn from payment behavior, disputes and customer interactions to prioritize actions, adapt as patterns shift and point teams to what will actually move cash.

When that foundation is in place, your data stops telling you what happened and starts helping you decide what to do next.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

Source link