Why Christy Johnson thinks AI's future in receivables is built on trust 

The receivables teams under the most pressure right now are at companies growing faster than the AR function can scale, with invoice volumes climbing quarterly, and headcount plateauing. When these teams evaluate artificial intelligence (AI) in accounts receivable, their questions are practical: what problems it solves, how it takes work out of the team's day, how it uses their data, and how they will know when it has acted on their behalf. 

Versapay's Chief Product Officer, Christy Johnson, starts her answer with judgment.

Growth pressure exposes something about the work. From outside, receivables reads as data entry: invoices go out, payments come in, someone matches the payments to the invoices and closes the month. What determines whether a company gets paid on time is how well a receivables team can read its customers and predict their behavior.

A collector choosing which of their 40 overdue accounts to call first is judging which relationship will bend and which will break, as well as which customers are at risk of nonpayment and which will respond to a timely reminder.

A credit manager asked to extend more trade credit to a customer that has hit its limit can look past the aging report for signs of good payment behavior: a linked bank account or card on file, auto-payments turned on, regular payment portal visits to review invoices. Those signals show intent to pay and intent to keep buying, which makes a higher limit a strategic way to protect the relationship.  

That is the constraint. Judgment of this kind lives with individuals, moves when they move, and varies from one collector to the next. Adding headcount adds capacity to a receivables team, but it does not add consistency, and it does nothing for the nearly three-quarters of organizations whose teams spend meaningful time every week chasing late payments. 

The AI gap in receivables is a data problem, and data problems can quickly turn into trust problems. Versapay's 2026 Cash Flow Clarity Report surveyed 400 finance leaders at companies between $100 million and $1 billion in revenue: 50% anticipate using AI to automate repetitive accounts receivable tasks, and 37% expect to adopt predictive analytics within receivables.

Both figures describe plans. Whether those plans pay off depends on the data artificial intelligence has to work with, and Christy points to two places where that data falls short. 

The first is fragmentation. When call notes live in the ERP, customer messages in Outlook, and trade credit allocations in Salesforce, an AI system working from any one of them has partial context, so its recommendations are harder to believe and the work it suggests can't be completed.

AI needs the full buyer context in one place to draw useful inferences and support a team's judgment at scale. That is why the choice of vendor and integration type matters, and why the ERP integration complexity that 47% of finance leaders cite as a top barrier holds back AI as well as AR software. 

The second is that reporting still depends on manual work to combine the numbers with what the team knows. Teams download aging reports, add notes by hand, collect context from the people who manage each buyer relationship, and send the result to the CFO's office, then repeat the process because the data only reflects a single point in time. Enterprise AI tools can speed up the analysis, but without connected data, someone still has to pull information from several systems and add the human context afterward.

This is where Versapay Intelligence helps. Because Versapay reconciles to and from the ERP, contact, invoice, and payment data stay current in near real time. Versapay also captures each buyer's comments, login history, payment history, and payment methods, and teams can push dispute details into Salesforce. Versapay's AR-trained AI runs on that full context to summarize an account, recommend next steps, and draft a response. 

Trust is the most used and least specified word in AI marketing, usually treated as a feeling to be managed with a confidence score and an explainability panel. Christy uses it as a threshold. Each stage is something the system reaches only after the previous one has held up.

The ladder she describes runs in four rungs. 

At the first, Augment, AI does the research, surfaces signals in data the team would otherwise dig for, and recommends what to do next. A person still decides. At the second, Assist, the system drafts the work: the email, the comment, the order of a task queue, the move of an account from one collections workflow to another. A person still approves. At the third, Action, it acts inside boundaries a person configured in advance. At the fourth, Autonomous, agents coordinate with other agents, set priorities, and close loops within trusted permission structures.

Christy's phrasing about how a receivables team moves up is worth quoting precisely: they get there, she says, "when you've gained that trust with credible recommendations." Gained, not granted. The condition is a record. 

Augment is the stage that looks like a starter tier. AI reads the data, says what it sees, and waits. For a team under pressure to show a return, it can read as the least ambitious option on the menu.

It is closer to the foundation for everything above it. Christy describes this stage as augmented intelligence: the system surfaces a pattern, recommends a decision, and shows the reasons and evidence behind it, and a person agrees or disagrees based on experience, intuition, or information the system didn't have. Each of those responses trains the system, so it improves as the team works with it and holds on to patterns of knowledge that would otherwise leave the organization.

Every recommendation the system makes while a person still decides shows the team how the system reads its customers, and that familiarity is what gives a team the confidence to let it do more. Skip this stage and a team arrives at autonomy without that confidence, which becomes a problem the first time a manager, a CFO, or a customer asks why an account was escalated. 

Receivables teams already measure judgment when it belongs to people. Promise-to-pay kept rates, collections effectiveness index, and average days to pay all exist to score decisions after the fact. Those same outcomes are where AI's contribution shows up as it takes on more of the work. 

AI readiness in AR comes down to a question few software vendors answer clearly: how a system takes on more responsibility over time, and who decides when it does. Autonomy claims describe what a system is permitted to do. This question describes whether a team stays in control as that changes.

For finance leaders setting requirements for AR software right now, this is criteria worth setting. A system built for the agentic future should be able to explain its recommendations and act only within boundaries the team has set.

Data belongs on that list too, because AI is only as good as the data it can reach, and AR teams are often short on it when their systems lack native integrations. Ask how a vendor keeps data in sync with the system the team works in, since current data is what makes recommendations, buyer summaries, and work-lists relevant. Versapay's native integrations sync data in near real time, and its adaptive algorithms are trained on a decade of AR experience.

This is also the clearest read on why 50% of finance leaders are still describing what they anticipate doing with AI rather than what it did for them last quarter. The evidence has not caught up with the expectation, but asking for it is how that changes.