The artificial intelligence term is now part of everyday language in finance departments. But between marketing promises and actual deployments, the gap often remains wide. At CashOnTime, we have chosen a rigorous approach based on clear governance principles and a fully in-house AI team. Here is how we concretely design AI to serve the accounts receivable.
AI: a technology much older than ChatGPT
Before getting into the core of the topic, a reminder is in order. Artificial intelligence was not born in 2022 with the emergence of ChatGPT in public debate. Its theoretical foundations date back at least to Alan Turing’s work after World War II and some prefer to go even further back to Ada Lovelace, a 19th-century mathematician often considered the first programmer, who had already formalized the idea of algorithmic reasoning on paper.
Why this reminder? Because it shapes how we approach the subject. AI is not a sudden rupture that must be tamed urgently. It is a long, profound evolution that deserves thoughtful integration, especially when it involves sensitive financial data.
Data governance: the real issue behind AI
Before talking about features, you must talk about governance. Too many organizations make this the last question when it should be the first. Our CPTO & RSSI puts it bluntly: his role as head of information systems security makes him, by definition, “a deeply paranoid data person.” And it is precisely through this lens that every design choice in our AI approach was guided.
Regulatory framework: GDPR, the AI Act and the duty to justify
Since 2018, the GDPR requires companies to account for their processing of personal data. France, despite being a forerunner together with Germany on these issues, was one of the last European countries to enforce it—a paradox that illustrates how hard it is to translate these texts into operational reality.
The EU AI Act follows the same regulatory logic. It has not yet been transposed into national law, but its intent is already clear: the regulation does not prescribe how AI systems must be built, but it requires vendors to be able to explain and justify their design choices. In other words: to answer precisely the question “why did you do it this way, and who is responsible?”
At CashOnTime, we have anticipated this requirement since our founding. Every design decision around AI is documented, traceable, and fits within a rationale we can defend to clients and regulators alike.
First principle: no data pooling between clients
This was a foundational choice we enforced from the start: CashOnTime does not use multi-tenant databases. Practically, this means each client’s data is isolated in its own environment; it is never mixed with another client’s data, neither for optimization nor for performance reasons.
This choice has a cost: it is more demanding to maintain and less infrastructure-efficient. But it delivers three decisive advantages:
- Security: a breach at one client cannot affect another client’s data.
- Client rights: each client retains full sovereignty over their data.
- Certification: this compartmentalization makes it easier to obtain strict security certifications.
This principle naturally extends to our AI architecture: your debtors’ data never feed a model shared with other client companies.
Second principle: fully internalized models, hosted on our infrastructure
This is the second pillar of our governance and arguably the most structural. We chose to invest in AI internally by building a dedicated team of three people: two data scientists and one developer. All our models are designed, trained and maintained by this team.
The direct consequence: we do not use any public third-party models, no GPTs, no models hosted on servers subject to other jurisdictions to process our clients’ data. Our models are based on royalty-free foundations that we host and run on our own infrastructure.
This decision requires more investment than an approach based on third-party APIs. But it ensures that your financial data and your debtors’ data never transit through a system over which we do not have full control.
Third principle: transparency toward the end user
Every action performed by AI on our platform is explicitly identified as such. The user always knows when they are looking at an automatically generated suggestion. This is not just an ergonomic choice: it is an ethical and regulatory requirement. Humans remain the decision-makers; AI proposes, the user validates.
This philosophy is embodied in our confidence score system associated with automatic matching suggestions: the reliability level is displayed, and it is the user, or their explicit settings, that determine the threshold above which an action may be automated without manual approval.
Artificial Intelligence in CashOnTime
Refined cash receipt forecasts
Cash receipt forecasts have long existed in our platform, calculated mathematically. AI now allows us to refine these projections significantly: where we used to forecast receipts one month ahead, we now target a fifteen-day window, even down to one week. The goal is to give finance teams more granular cash visibility for better working capital management.
Account matching with confidence scoring
This is one of the most impactful features. When your major payers send payment advices in various formats, or when payment notices arrive by email, our AI analyses that data and proposes automatic matching of the relevant invoice lines.
Each suggestion is accompanied by a confidence score. This score lets the user assess the reliability of the suggestion and, if the threshold is high enough, allow fully automated processing. In all cases, the interface clearly identifies operations handled by AI; transparency toward the user is a non-negotiable design principle.
Discover CashOnTime’s automatic account matching features.
Intelligent prioritization of collection tasks
Our AI analyses users’ workloads and suggests a prioritization of collection actions based on multiple criteria: probability of collection, the debtor’s historical behaviour, and urgency of due dates. The goal is simple: focus accounts receivable teams’ efforts where they have the greatest impact.
Ongoing AI developments at CashOnTime
Contextual assistant and industry chatbot
We are working to transform our online help into a conversational assistant capable of understanding natural language requests. The user will be able to describe their need or issue; the assistant will direct them to the right feature or help with configuration, always respecting the principle that the human decides.
Sentiment analysis on client communications
Under experimentation, this feature will analyse the tone of emails from your debtors. A contact who normally replies neutrally but suddenly responds in a tense or stressed manner: the AI detects it and generates an alert. This type of indicator enables real-time adaptation of collection strategy.
Adaptive recovery strategies
Today, a collection strategy is defined statically: call, then email, then letter, etc. Tomorrow, our algorithms will analyse each debtor’s history to dynamically suggest the most effective steps or recommend skipping those that have never worked for that profile.
AI as a companion, not a substitute
We want to be clear: AI will not replace accounts receivable teams. It is, and will be, their best amplification tool. Like any industrial revolution, it transforms roles, changes some skills and renders others obsolete. But tasks that require judgement, relationship and contextual decision-making remain, and will remain, human.
Our development philosophy reflects this: every AI feature we deploy is designed to free time from low-value tasks and allow teams to focus on complex cases that require real domain expertise.
The right reflex when any vendor says “we do AI”
One final recommendation, useful well beyond CashOnTime: whenever a vendor presents an AI-enabled solution, ask these questions every time:
- How are the data stored? Are they isolated per client or pooled?
- Who are they shared with? Are the models hosted in-house or outsourced?
- Are the models open or closed? Do your data feed a model shared with other companies?
These questions are not trivial. They will determine your future compliance with the AI Act and they protect today your financial data and that of your debtors.
Would you like to learn more about our CashOnTime accounts receivable automation solution? Talk with one of our experts who will support you in your considerations.
