Organisations are investing heavily in automation. From robotic process automation (RPA) and workflow platforms to artificial intelligence and autonomous AI agents, the promise is attractive: fewer manual tasks, faster execution, lower costs and more efficient operations.
Yet there is a fundamental question that often receives far less attention than the technology itself:
Do we actually understand the business process we are trying to automate?
It is surprisingly common for organisations to begin an automation initiative by identifying a task, selecting a platform and discussing technical implementation. Someone notices that employees spend hours entering information, processing requests or transferring data between applications. The immediate reaction is to automate that activity.
But the task is only one part of a larger process. It has inputs, outputs, dependencies, business rules, decisions, exceptions, responsibilities and relationships with other activities. Automating an individual task without understanding these connections can create a solution that works technically but delivers disappointing business results.
This is where Business Process Management (BPM) becomes essential.
BPM is not simply a way to document how work is performed. It is a management discipline for understanding, analysing, designing, executing, monitoring and continuously improving business processes.
And as organisations move from conventional automation towards AI-driven and agentic systems, the importance of BPM is becoming even greater.
Successful automation begins with understanding, analysing and improving the business process — not with selecting an automation tool.
The automation trap: starting with technology instead of the process
Imagine an organisation where employees receive hundreds of supplier invoices every month.
The accounts payable team manually checks invoice details, compares them with purchase orders, obtains approvals, resolves discrepancies and enters information into the accounting system.
Management sees an opportunity to automate invoice processing. A new platform is introduced, invoices are scanned using intelligent document processing, and an automated workflow routes them for approval.
The project appears successful. Data entry is faster, fewer documents are handled manually, and invoices move through the system more quickly.
But a few months later, problems begin to surface.
Invoices are routed to the wrong managers. Some approvals are delayed because the workflow does not account for employee absences. Duplicate invoices occasionally pass through. Discrepancies between purchase orders and invoices still require manual investigation. Employees develop workarounds because the automated process does not reflect how certain transactions are actually handled.
The technology is functioning as designed. The problem is that the organisation automated an incomplete understanding of its process.
Perhaps the original process had unnecessary approval layers. Perhaps responsibility for resolving discrepancies was never clearly defined. Perhaps different categories of invoices required different approval rules. Perhaps the organisation had never established a consistent approach to handling exceptions.
Automation did not solve these problems because they were not primarily technology problems.
They were process problems.
This distinction is important. Technology can execute instructions, transfer information and support decisions, but it cannot automatically compensate for an organisation that has not clearly defined how its work should be performed.
In some cases, automation makes existing inefficiencies more visible. In others, it embeds them more deeply into the organisation's operations.
The result is an automated process that may be faster in certain activities but remains fundamentally inefficient.
BPM provides the understanding that automation needs
Before an organisation decides what to automate, it needs to understand the process itself.
That means looking beyond individual tasks and examining the complete sequence of activities required to achieve a business outcome.
A proper BPM approach typically involves several connected stages:
1. Process identification
Identify the business processes that exist across the organisation, understand their purpose, and determine which ones are important to business objectives.
2. Process discovery
Understand how the selected process actually works. This involves identifying activities, participants, systems, information flows, decisions, business rules and exceptions.
3. Process analysis
Examine the current process to identify bottlenecks, unnecessary activities, duplicated work, delays, errors, risks and opportunities for improvement.
4. Process redesign
Develop an improved way of performing the work. This may involve removing unnecessary activities, simplifying approvals, clarifying responsibilities, changing business rules or redesigning the sequence of activities.
5. Process implementation
Put the redesigned process into operation. Depending on the requirements, this may involve organisational changes, new procedures, workflow systems, RPA, AI or a combination of technologies.
6. Process monitoring and continuous improvement
Measure how the process performs, identify deviations and emerging problems, and make further improvements as business requirements change.
These stages are interconnected. They should not be treated as isolated administrative exercises or a checklist completed simply to satisfy a project methodology.
The objective is to understand how work creates value, determine how it can be improved, and then select the most appropriate way to implement that improvement.
Process modelling, including Business Process Model and Notation (BPMN), can play an important role in this work. A well-constructed process model makes activities, decisions, responsibilities and interactions visible, providing a common understanding for business stakeholders, process analysts and technology teams.
However, drawing a process diagram is not the same as managing a process. BPM extends beyond modelling to include performance management, governance, execution and continuous improvement.
Redesign before you automate
One of the most valuable contributions of BPM is the opportunity to improve a process before investing in its automation.
Consider returning to the invoice-processing example.
Instead of immediately implementing an automation platform, the organisation first studies the existing process.
What the analysis reveals
- All invoices follow the same approval route, regardless of value or risk.
- Some approvals are duplicated because responsibilities overlap.
- Purchase order matching is performed manually even when the relevant information is already available electronically.
- There is no clearly defined route for resolving discrepancies.
- Employees regularly interrupt the normal workflow to request missing information.
- The organisation has limited visibility into where invoices are waiting and why.
Rather than automating the existing process as it stands, the organisation redesigns it.
Low-risk invoices with valid purchase order matches follow a simplified approval route. Higher-value transactions receive the appropriate additional approval. Duplicate checks are performed automatically. Discrepancies are directed to a clearly defined exception-handling process. Approval responsibilities are clarified, and process performance is monitored through measurable indicators.
Only after the redesigned process is understood does the organisation determine which activities should be automated.
The resulting solution is more than a faster version of the original process. It is a better process, supported by technology.
This is the difference between automating existing work and using automation as part of genuine process improvement.
It is also why process redesign should not be confused with technology implementation. The organisation must first decide what the improved process should achieve. Technology then becomes an enabler of that design.
Not every process needs the same type of automation
Once a process has been analysed and redesigned, the organisation can make a more informed decision about the technology required.
Automation is not a single technology, and different processes have different requirements. For example:
- Workflow automation is appropriate when activities follow relatively predictable sequences, with defined conditions, responsibilities and approval rules.
- Robotic Process Automation (RPA) can be useful when repetitive, rule-based tasks involve interacting with existing applications, particularly where direct system integration is unavailable or impractical.
- Intelligent Document Processing (IDP) can extract and classify information from documents such as invoices, contracts and application forms, reducing manual data entry.
- Artificial Intelligence can support activities involving classification, information extraction, summarisation, prediction and decision support, particularly where information is less structured.
- AI agents introduce another possibility: systems that can interpret objectives, use tools, interact with other systems and perform sequences of activities with varying degrees of autonomy.
These technologies are not necessarily alternatives to one another. A well-designed business process may combine several of them. (For a closer look at the platform categories, including BPM suites, see Low-Code Automation Platforms.)
An invoice-processing process, for example, might use document intelligence to extract invoice information, conventional workflow automation to manage approvals, an AI agent to investigate certain discrepancies, and an ERP system to record the final transaction.
The important decision is not which technology is currently attracting the most attention. It is which combination of capabilities best supports the redesigned process.
Without process understanding, organisations risk selecting technology based on its features rather than its contribution to the business outcome.
AI agents make BPM more important, not less
The emergence of AI agents introduces a significant change in the automation landscape.
Traditional automation generally follows predefined instructions. A workflow moves from one activity to another according to configured conditions. An RPA bot executes a sequence of programmed actions. If an unexpected situation occurs, the automation may stop, fail or require a predefined exception route.
AI agents introduce greater flexibility.
An agent may interpret information, decide which tool to use, retrieve data from different systems, communicate with people or other agents, and determine the next action based on the context.
This flexibility creates opportunities, but it also introduces new risks.
An agent operating within a business process must understand more than the immediate task it has been assigned. It needs to understand the objective of the process, the boundaries within which it can operate, the information it is allowed to access, the decisions it may make and the situations that require human intervention.
Consider a supplier onboarding process.
An AI agent might collect supplier information, review submitted documents, check required fields, identify missing information, communicate with the supplier and prepare the case for approval.
But what happens when a document is inconsistent? Can the agent reject the application? Can it change supplier details? Can it approve a supplier that fails a compliance check? Is it permitted to contact an external organisation? When must it stop and request human review?
These are not merely technical questions about what an AI model can do. They are process governance questions.
The organisation must establish the rules, responsibilities, decision boundaries and escalation paths before granting the agent operational authority.
This is where BPM provides a necessary foundation.
A process-aware AI agent should operate within a defined business context, understand the objective it is helping to achieve, respect organisational constraints and recognise when a situation falls outside its permitted authority.
In this environment, process management is not a restriction that prevents AI from being useful. It is what makes greater autonomy manageable.
The emerging concept of agentic BPM
This relationship between AI agents and business processes is now receiving serious attention in academic research.
A 2026 paper published in the journal Information Systems, titled Agentic Business Process Management: A Research Manifesto, proposes a framework for managing processes in which human and software agents work together.
Four capabilities for agentic BPM
- Framed autonomy: Agents should have the freedom to act within clearly established process boundaries.
- Explainability: Agents should be able to explain the reasoning behind their actions.
- Conversational actionability: Agents should be able to interact with people, other agents and business systems to execute work.
- Self-modification: Agents should support adaptation and improvement while remaining subject to appropriate governance.
The concept is particularly interesting because it distinguishes automation from autonomy. Automation executes predefined instructions, while autonomy allows an agent to make context-sensitive decisions within an established process framework.
A separate 2026 study, A Modular Large Language Model Agent Architecture for Adaptive and Autonomous Process-Aware Execution, explores this idea through a modular architecture involving a Frame Agent and an Operational Agent. The researchers demonstrate its application using real-world meter-to-cash process data and compare its execution with an RPA bot.
These studies suggest an important direction for the future of BPM: rather than treating AI agents as independent tools that perform isolated tasks, organisations can increasingly design them as participants in managed business processes.
This does not mean that every organisation needs autonomous agents or that traditional workflow automation is becoming obsolete. Many business activities remain better suited to deterministic, rule-based automation.
The point is that greater technological autonomy requires greater clarity about the processes in which that autonomy is exercised.
The process must define the boundaries of automation
One of the biggest mistakes organisations can make when adopting AI agents is to confuse the ability to perform an action with the authority to perform it.
An AI agent might technically be capable of approving a transaction, changing customer information, issuing a refund or modifying a supplier record.
But technical capability does not automatically establish business authority.
The process must define what the agent is allowed to do.
For example, an organisation might permit an agent to
- Collect and validate information.
- Identify missing documents.
- Recommend an appropriate next action.
- Prepare an approval request.
- Execute low-risk actions within established rules.
- Escalate unusual or high-risk cases to a human employee.
However, the organisation may require human approval before the agent can make certain financial commitments, change sensitive records or take actions with legal or regulatory implications.
These boundaries should not exist only as informal instructions in an AI prompt. They should be supported by organisational policies, process rules, access controls, system permissions, monitoring and appropriate escalation mechanisms.
This is particularly important when multiple agents interact with different business systems.
A process may involve an AI agent collecting information, another checking compliance, a workflow engine managing approvals and an employee making the final decision. Without a clear process framework, each component may function independently while the overall business process becomes difficult to control.
BPM provides the structure needed to coordinate these activities and maintain accountability.
Measuring automation success: faster is not always better
Another important reason to begin with BPM is that it helps organisations define what successful automation actually means.
Many automation projects measure success through technical indicators:
- Number of automated tasks.
- Number of bots deployed.
- Hours of manual work eliminated.
- Processing speed.
- Number of transactions handled automatically.
These measurements are useful, but they do not necessarily demonstrate that the business process has improved.
An organisation might automate a process and reduce its processing time by 30%, yet still experience a high number of errors, customer complaints or unresolved exceptions.
The automation may be technically successful while the business outcome remains disappointing. It is the same gap between task-level gains and process-level results explored in last month's essay, The AI Productivity Paradox.
BPM encourages organisations to evaluate performance at the process level. Depending on the nature of the process, relevant indicators may include:
| Performance area | Example measurement |
|---|---|
| Efficiency | Average end-to-end processing time |
| Cost | Cost per completed transaction |
| Quality | Error and rework rates |
| Compliance | Percentage of transactions meeting policy requirements |
| Customer experience | Customer satisfaction or resolution time |
| Reliability | Percentage of cases completed without manual intervention |
| Exceptions | Percentage of transactions requiring escalation |
The organisation should establish meaningful baseline measurements before implementing automation and compare them with actual results afterward.
It should also monitor performance continuously. Business processes change, regulations evolve, transaction volumes fluctuate and organisational responsibilities are modified.
An automation solution that performs well today may become inefficient tomorrow if the underlying process changes and the technology is not updated accordingly.
This is why automation should be treated as part of an ongoing process management discipline rather than a one-time technology project.
A practical framework before starting any automation initiative
Before selecting an automation platform, an organisation should be able to answer several fundamental questions.
1. What business outcome are we trying to achieve?
Start with the objective, not the technology. Is the goal to reduce processing time, improve accuracy, lower operating costs, improve customer experience or strengthen compliance?
2. Do we understand the existing process?
Identify the activities, participants, systems, business rules, decisions, inputs, outputs and exceptions. Understand how work is actually performed, not simply how management assumes it is performed.
3. What is wrong with the current process?
Analyse bottlenecks, unnecessary approvals, duplicated work, delays, errors, manual intervention and control weaknesses.
4. Can the process be simplified or redesigned?
Remove activities that do not add value, clarify responsibilities, improve decision rules and establish appropriate exception paths before automating.
5. Which activities should be automated, and which should remain under human control?
Not every activity benefits from automation. Some decisions require judgement, accountability or specialised expertise. Determine where conventional automation, AI or human intervention is most appropriate.
6. What technology best supports the redesigned process?
Select tools based on process requirements, integration capabilities, security, scalability, governance and expected business value.
7. How will performance be measured and controlled?
Define process-level KPIs, establish monitoring arrangements and determine how deviations, exceptions and changing business requirements will be managed.
These questions provide a practical starting point for organisations that want automation to deliver lasting improvements rather than isolated technical successes.
The future belongs to organisations that understand their processes
Automation technology will continue to evolve. RPA platforms will become more intelligent, workflow systems will incorporate more AI capabilities, and autonomous agents will take on increasingly complex activities.
But technology alone will not determine which organisations succeed.
The organisations that achieve sustainable results will be those that understand how their businesses operate, recognise where improvement is needed, redesign processes around meaningful outcomes and introduce automation with a clear purpose.
BPM is not an administrative exercise that organisations must complete before they are allowed to use technology. It is the discipline that connects business objectives, people, processes and technology.
And as AI agents become more capable of taking action, this discipline becomes even more important.
An organisation cannot responsibly delegate work to an autonomous system if it has not first established what that work involves, what outcome it should produce, what constraints apply and who remains accountable.
The real opportunity is not to automate every possible task. It is to build better business processes and use technology to execute them more effectively.
Before asking what you can automate, ask whether you understand the process well enough to know what should be automated.
Because in the end, successful automation does not begin with a bot, a workflow or an AI agent.
It begins with understanding the business process.
References and further reading
- Calvanese, D., Casciani, A., De Giacomo, G., Dumas, M., Fournier, F., et al. (2026). Agentic Business Process Management: A Research Manifesto. Information Systems, 102738. https://doi.org/10.1016/j.is.2026.102738
- Skolik, A., Kaltenpoth, S., Beverungen, D., & Müller, O. (2026). A Modular Large Language Model Agent Architecture for Adaptive and Autonomous Process-Aware Execution. Information Systems, 102748. https://doi.org/10.1016/j.is.2026.102748
- Dumitriu, F., Greavu-Șerban, V., Necula, S.-C., & Fătu, V.-C. (2026). Integrating LLM in Business Process Management: A Conceptual Framework for Augmenting the Process Lifecycle. Systems, 14(9), 1076. https://doi.org/10.3390/systems14091076