- According to the Institut de la statistique du Québec (November 2025), 12.7% of Quebec businesses used AI for production purposes, and annual growth is less than half of Ontario's pace.
- As of April 2026, SAP counted more than 30 specialized Joule agents and over 2,500 skills, including agents dedicated to invoice disputes and payment advice processing.
- Industrial group Wieland raised its incoming payment automation from 61% to 83% with SAP's machine learning matching (SAP case study).
- PlanAxion recommends starting with a high-friction finance process, such as cash application, rather than a broad rollout.
It is 8:30 a.m. on the third of the month. The accounts receivable team at a B2B distributor opens the bank statement and finds the same puzzle as last month: bundled payments, partial amounts, no invoice references. Meanwhile, ERP vendors are announcing AI agents that handle those tasks on their own.
Both realities coexist. The gap between them defines 2026.
According to the Institut de la statistique du Québec (November 2025), 12.7% of Quebec businesses used AI for production purposes. Year-over-year growth reached 3.3 points in Quebec, versus 7.8 points in Ontario.
What is agentic AI in an ERP?
Agentic AI refers to software agents that run a complete business process inside the ERP, from analysis to action, instead of suggesting an answer for a human to apply. A copilot answers your questions. An agent closes the file.
The shift has accelerated among major vendors. As of April 2026, SAP counted more than 30 specialized agents and over 2,500 Joule skills, including an agent that analyzes the root cause of invoice disputes and automated payment advice processing. For the vendor-by-vendor detail, our review of AI agents in ERP systems covers what SAP, Microsoft and Oracle are shipping.
The distinction with classic generative tools matters for your architecture choices. We covered it in our article on the difference between traditional AI and generative AI.
Why is finance the first proving ground for AI agents?
Because finance processes combine the three conditions an agent requires: documented rules, high volumes, and results you can measure in dollars and days. Few functions offer terrain that clean.
Take cash application. The problem is not receiving the money. It is knowing exactly where each payment belongs when a customer settles 40 invoices with a single wire, deducts two credit notes, and sends no remittance advice.
In the accounts receivable mandates PlanAxion leads, that manual sorting is what consumes the team's hours, not collections itself. The cash application solutions PlanAxion implements target that exact friction point: removing repetitive matching so the team can focus on exceptions and collections.
Documented results exist. German industrial group Wieland raised its incoming payment automation from 61% to 83% by deploying SAP's machine learning matching across three group companies. SAP itself reports a 71% reduction in AR matching effort with this type of tool (2025). Vendor figures, to be taken as orders of magnitude, but the mechanics repeat: less manual sorting, more time for exceptions.
Where do Quebec businesses stand compared to Ontario?
Quebec is moving forward, but at less than half of Ontario's pace, and intentions for the coming year hold that gap in place. The Institut de la statistique du Québec figures, drawn from the Canadian Survey on Business Conditions, paint a precise picture.
- AI use in production: 12.7% in Quebec, 13.3% in Ontario (Q2 2025).
- Year-over-year growth: +3.3 points in Quebec, +7.8 points in Ontario.
- Planned use over the next 12 months: 13.1% in Quebec, 16.5% in Ontario.
- Businesses with 100 employees or more: 26.1% adoption, versus 12.2% for those with 1 to 4 employees.
- Main reported barriers: implementation cost (27.2%), uncertain return (19.3%), lack of expertise (14.3%).
Those barriers say something other than a rejection of AI. Companies mostly doubt their ability to pick the right project. A SAP and Oxford Economics study of 1,600 executives (2025) measures an average 16% return on AI investments, and 78% of executives believe agents can transform their operations. Again, vendor figures rather than a guaranteed average.
An AI agent rarely fails on the technology. It fails on accounts receivable data nobody validated.
How do you go from idea to a first use case in production?
Pick one process, validate the data that feeds it, and quantify the expected gain before writing a single line of code. That is framing work, not a technology program.
PlanAxion's quick-win AI solutions workshop structures that framing over four weeks and five steps: prepare, identify, prioritize, validate the data, then decide and deliver. Every idea goes through the same filter: problem to solve, expected value, available data, required effort.
The deliverable is a prioritized roadmap, not another report. The investment varies with scope and is confirmed during a short exploratory call. For eligible companies, Investissement Québec's ESSOR program can support the studies leading up to this type of project.
Should you wait for agentic AI to mature before acting?
No. Agents improve every quarter, but preparing your data and choosing your first use case do not depend on tool maturity. Companies that test a narrow scope now learn on their own data, at their own pace. The first AI use case should be specific enough to test, measure and adopt.
Frequently asked questions about agentic AI
What is agentic AI in simple terms?
Agentic AI covers software agents able to carry a business process from start to finish: read the data, decide according to rules, execute the action in the ERP, and document the result. Unlike a chatbot, the agent does not wait for a human to apply its recommendation. It acts, under defined supervision.
What is the difference between an AI copilot and an AI agent?
A copilot assists a person: it summarizes, drafts, suggests. An agent completes a full task autonomously, such as matching a payment to open invoices or analyzing a dispute. They complement each other, but the agent demands more: clean data, clear rules, and supervision defined from the start.
Which finance processes should you automate first with AI?
Start with high-volume repetitive processes whose results are easy to measure: cash application, payment advice processing, invoice dispute resolution. These tasks combine documented rules with measurable gains in recovered hours, cash applied faster, and a shorter month-end close.
How long does it take to identify a first profitable AI use case?
A structured approach takes about four weeks: prepare the team, identify friction points, prioritize by value and effort, validate the available data, then decide. The result is a costed roadmap. The time to deploy the first agent then depends on data quality and the chosen scope.
