The Future of ERP Is Agent-Driven - Here's How We're Building It

ERP is evolving - from systems of record to intelligent systems of action.

The Future of ERP Is Agent-Driven - Here's How We're Building It
Mads Kalør
May 5, 2025
Product

Enterprise Resource Planning (ERP) systems are changing. What used to be static systems of record—designed to track transactions and enforce structure—are becoming intelligent, adaptive platforms. With advances in artificial intelligence, especially large language models (LLMs) and agent-based automation, the ERP of the future will not only tell us what happened but also what to do next—and increasingly, will do it for us.

At Kaunt, we believe that ERP processes will be executed by AI agents. These agents will understand, interpret, act, and improve—handling finance workflows, posting journal entries, managing exceptions, and learning from feedback. But autonomy doesn’t mean loss of control. In fact, we believe trust, transparency, and configurability are the cornerstones of making this future work.

We're not just imagining this future. We’re building it—one agent at a time.

Why We Built Kaunt as API-First

Let’s start with how Kaunt fits into the ERP ecosystem. We are not a platform you bolt onto an ERP system. Instead, ERP vendors integrate into Kaunt through APIs.

This decision reflects a fundamental belief: the best user experience is an embedded one. ERP vendors know their customers, their workflows, and their design language. With Kaunt, they retain full control of the UI and product experience—while outsourcing the complexity of AI, machine learning, and model iteration.

This API-first model means we can evolve quickly, continuously improve our models, and serve ERP vendors across verticals and geographies—without depending on heavy custom integrations or release cycles. It’s future-proof, scalable, and flexible. But most importantly, it makes the AI feel like it belongs inside the ERP system, not just added on top of it.

What We’re Already Delivering

Our vision is ambitious, but it’s also grounded in what we’re already doing today.

Kaunt is live with ERP partners who use our services to automate critical, high-volume processes in accounts payable and finance.

Here’s what’s already in use:

  • Invoice Coding: Our AI predicts the correct GL account, VAT code, and cost center for each invoice line—based on language, numbers, and context.
  • Anomaly Detection: We flag deviations from expected coding patterns and suggest corrections—preventing errors before they hit the ledger.
  • Confidence-Based Automation: When the model is highly confident, and the amount is low-risk, invoices can be coded and posted automatically. When confidence is lower or the stakes are higher, we defer to the user.

This is not just AI as an assistant—it’s AI doing the work, safely, when the system knows it can be trusted. It’s the beginning of ERP autonomy.

Our Roadmap: Building Autonomy in Layers

We’re building Kaunt’s AI services as modular agents—each focused on a specific task or workflow, each getting smarter over time. These agents evolve from suggestions to actions to orchestration, depending on the needs and comfort level of the ERP vendor and their customers.

We’re currently in Phase 1, where foundational agents automate high-volume, low-risk tasks. This work is live today, helping ERP vendors improve efficiency, reduce manual workloads, and unlock new value for their customers. From here, we build upward—layering in foresight, decision support, workflow automation, and eventually, full process orchestration.

Each phase of this roadmap is shaped by how organizations adopt new technology: they start by automating the known, then gradually move toward trusting systems to act, guide, and adapt. That’s exactly how we’re building Kaunt.

Phase 1 - Task Automation: Agents That Act on Known Patterns

We start with agents that solve known, well-bounded problems:

  • Predicting GL accounts and cost centers
  • Matching invoices to POs
  • Extracting data from documents
  • Normalizing vendor and master data

These are areas where the structure is clear, the data is rich, and the risks are manageable. It's where trust in automation is easiest to build.

Because these tasks repeat thousands of times each month in every ERP system, the impact is immediate. Workloads shrink, errors decline, and finance teams can focus on the exceptions instead of the routine.

And that trust is key. Once users see that the system can code correctly, explain its reasoning, and defer decisions when needed, they become open to a new question: what else could it do for me?

Phase 2 - Contextual Guidance

Once automation is trusted, we introduce agents that interpret context and guide decisions before they’re made. These agents are explainable, configurable, and designed to keep humans in control:

  • Detecting policy violations or anomalies
  • Suggesting approval paths based on behavior
  • Flagging risks or inconsistencies
  • Summarizing patterns, KPIs, and historical actions

Using predictive models and LLMs, these agents don’t just process data — they narrate it. They help users understand what’s happening and what to do next, without navigating deep configuration menus.

This phase also prepares the ground for automation to go further. Agents that understand policies, patterns, and user preferences are far better equipped to act responsibly—and vendors gain the ability to fine-tune how much autonomy to give, and where to keep approvals in place.

Phase 3 - Workflow Execution

With that foundation in place, agents can begin to orchestrate actual workflows. Not just a suggestion here or a forecast there - but real actions:

  • Managing expense reports, invoice workflows, or reconciliations
  • Running approvals, escalations, and follow-ups
  • Taking user commands in natural language

What changes at this point is the relationship between the user and the system. Instead of simply receiving outputs, users interact with agents in natural language—asking them to generate reports, delegate tasks, or walk through a decision. The ERP starts to feel like a partner, not just a tool.

Phase 3: Flexible workflow rules in natural language.

Because the groundwork has already been laid—through explainability, risk thresholds, and user configuration—vendors can offer this functionality without compromising on compliance or oversight.

Phase 4 - Adaptive Orchestration

Ultimately, we’re working toward ERP systems that run more of themselves. Not invisibly, and not without supervision—but with agents that can carry out entire processes:

  • Procure-to-pay
  • Order-to-cash
  • Month-end close

At this stage, agents can learn from usage, adjust to updated policies, and even apply new coding practices without being explicitly trained. They can generate audit-ready documentation for every action taken, explain their logic, and route only the edge cases for human review.

It’s a future where ERP systems are continuously optimizing themselves—not because someone told them to, but because they’ve learned how to do it well.

We’re not there yet—but we’re building toward it, with each agent and each integration designed to make that future both achievable and trustworthy.

Why ERP Vendors Work With Us

ERP vendors don’t want to build and maintain their own AI stacks. They want to deliver intelligent, modern experiences to their customers—quickly, safely, and without enormous R&D overhead.

Kaunt makes that possible.

By integrating into Kaunt:

  • Vendors get a stream of evolving AI capabilities with minimal effort
  • Their customers get a deeply embedded, intelligent ERP experience
  • They stay in control of workflows, UI, and autonomy levels

Because our APIs are modular, they can start small—then grow.

Where We're Headed

We believe that autonomy will come from layered trust. Not all processes will become autonomous overnight—and they shouldn’t. But piece by piece, task by task, ERP systems will evolve into intelligent systems that don’t just record what’s happening—they help make it happen.

The future of ERP is agent-driven. And we're already on the way there.

Mads Kalør

CTO