
Agentic process automation (APA) is the automation of end-to-end business processes by AI agents that work towards an outcome instead of following a fixed script. The agent reads the incoming work, uses context about how the process runs to decide the next step, acts in systems such as SAP, Salesforce and Outlook, and hands the decision to a person when it falls outside the limits the business has set.
Most enterprises have already automated the predictable parts of their processes. Workflows run, integrations move data, RPA bots key in the standard cases. The work that is left arrives as an email, a PDF or a forwarded thread: an order with a missing reference, a change request on an order already in flight, a complaint that needs three systems to answer. People read it, look things up and retype the result. That is the work agentic process automation is meant to take over.
Agentic process automation applies AI agents to a whole process rather than to a single task. Three things set it apart from earlier automation:
APA is not a chatbot that answers questions, and not a copilot that drafts text for a person to send. Both are useful; neither completes the work. Gartner calls the relabelling of such tools as agents "agent washing", and in June 2025 estimated that only about 130 of the thousands of vendors claiming agentic AI were genuine.
An APA system runs as a loop. The steps below describe what happens to one piece of incoming work, from the moment it arrives to the moment the outcome feeds back into the next decision.

Robotic process automation (RPA) repeats steps that someone has recorded or scripted. It is fast and reliable when the input is structured and the path never changes. Agentic process automation handles the cases where the input is messy and the path depends on what the request says. Intelligent process automation (IPA) sits between the two: it adds machine learning, OCR or document extraction to scripted automation, but the process logic stays fixed.
APA does not replace RPA. A stable task with structured input is cheaper and more predictable with a script than with an agent. In practice the two work together: the agent decides what needs to happen, a bot or workflow carries out the fixed steps. The full comparison is in APA vs RPA.
The terms overlap, so it helps to separate them.
Put simply, process intelligence is the understanding and APA is the execution. Gartner's 2026 Hype Cycle for Agentic AI lists context graphs among the emerging practices for building and managing agentic systems: a live model of how work unfolds that agents can act on. Without that context, an agent automates the process as someone imagines it, not as it runs. APA versus process intelligence covers how the two connect.
APA fits processes where the work arrives as communication rather than as clean system data, and where the volume makes manual handling expensive.
Order intake. A purchase order arrives as a PDF attachment. The agent reads it, matches the customer and products against master data, creates the order in the ERP and sends a confirmation, routing it for review only when something does not match. Dossche Mills automated order routing and entry into SAP across its EMEA customer service operation, reaching over 95% classification accuracy and €100k in net annual savings. See sales order automation.
Order changes. A customer emails to move half of a delivery to next month. The agent finds the order, checks what the change affects and updates the ERP or asks for approval. See order change management.
Customer service routing. Requests are classified, prioritised and enriched with order and customer data before they reach the right team. BD, the medical technology company, handles 1.4 million inquiries a year in more than 15 languages. After classification and routing were automated, first response time fell from 3 days to 22 hours, and SLA compliance for product incident reporting rose from 13% to 97%.
Disputes and deductions. A short payment arrives with a one-line remittance note. The agent classifies the deduction, opens a case and gathers the evidence from across systems before an analyst decides. See deduction management.
Accounts payable. Supplier invoices and queries arrive in a shared inbox. The agent extracts the invoice data, matches it to the purchase order and goods receipt, and answers status questions from suppliers.
The first three sit in the quote-to-cash process, which is where most enterprises start because volumes are high and the manual work is easy to measure. How that works step by step is covered in quote-to-cash automation.
For a comparison of the types of tools on the market, see this guide to agentic process automation tools.
In June 2025 Gartner predicted that over 40% of agentic AI projects would be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. Three patterns sit behind most of those failures:
Start with one process where the work arrives by email in high volume, such as order intake, order changes or customer service requests. Map how it really runs first, including the inbox, so you know the volumes, the variants and the exceptions. Automate the most frequent, best-understood path, keep a person in the loop for the decisions that need judgement, and measure against the baseline. Then extend the same context to the next step of the process.
That is the order Tekst follows: process intelligence as the foundation, agentic automation on top. The agents act in the systems a team already uses, and every action adds to the picture of how the process runs.