What is process mining?
Process mining is an analysis method that derives a process model from the event logs of information systems, compares that model with the target process and shows where to improve. Every record has a case (for example a sales order), an activity (approval, shipment, invoicing) and a timestamp; put side by side, these three rebuild the process step by step, with all of its variants.
The manifesto of the IEEE Task Force on Process Mining names three core uses: discovery (deriving the model from the logs), conformance checking (comparing the actual flow with the target model) and enhancement (enriching the model with real data). The XES standard is used to move event logs between tools.
What does an event log look like?
The raw material of process mining is the event log. The small example below shows the trail of two orders in ERP; when the same steps run in a different order and with different waits, across hundreds of cases the process map is built from these differences. The example is illustrative.
| Case | Activity | Time | Resource |
|---|---|---|---|
| Order A | Order entered | Monday morning | Sales |
| Order A | Credit approval | Monday afternoon | Finance |
| Order A | Shipment | Wednesday | Warehouse |
| Order B | Order entered | Monday morning | Sales |
| Order B | Price change | Tuesday | Sales |
| Order B | Credit approval | Thursday | Finance |
| Order B | Shipment | The following week | Warehouse |
Where does it create value in a company?
Process mining looks at business processes: flows that leave records, cross several departments and are costly when they stall. Typical starting points are:
In businesses that manufacture, machine-level signals, such as downtime, cycle time and sensor data, are the domain of MES and IIoT, not of process mining. The two complement each other: MES shows what happens on the shop floor, process mining shows where the business flow waits.
- order to cash: waits and loops between quote, order, shipment, invoice and collection,
- procure to pay: bottlenecks in request, approval, order, goods receipt and invoice matching,
- in manufacturing, the production order flow: material waits and replanning loops from opening an order to closing it,
- maintenance and quality: the path from a failure report to closure, from a nonconformity to corrective action,
- master data changes: approval rounds when a new material or customer is created.
MES and IIoT guide: how shop-floor data reaches the plan
Process mining use cases in manufacturing
In a manufacturing company, process mining is most useful on flows that leave records in the ERP and cross several departments. The table below summarises typical examples, what the analysis makes visible and which decision that view supports. Order to delivery, purchasing and invoice to cash are examined the same way in companies that do not manufacture; the production order and maintenance rows are specific to manufacturing.
The decisions in the table are examples. Which one makes sense depends on what the wait or deviation found by the analysis costs the company, and once a decision is applied it is checked again with the same measure.
| Process | What becomes visible? | Typical decision |
|---|---|---|
| Order to delivery | At which step an order waits between entry and shipment; where due-date changes and manual corrections concentrate | Changing the order of the credit and stock checks, reviewing the rule for promising delivery dates |
| Purchasing | Approval rounds from request to order; orders approved after the fact; mismatches between goods receipt and invoice | Simplifying approval limits, extending the use of framework agreements |
| Production order | The steps at which an order waits for material or capacity between opening and closing; replanned and split orders | Checking material availability before an order is released, changing how often planning runs |
| Maintenance | The path from a failure report to a work order and closure; waits for spare parts; recurring failure records on the same equipment | Redefining the stock of critical spare parts, bringing planned maintenance forward |
| Quality | The steps from a nonconformity record to corrective action; records left open | Making the owner and the closing step explicit |
| Invoice to cash | The path from invoice to collection; disputed and reissued invoices; at which step overdue receivables build up | Adding a check before invoicing, prioritising collection follow-up |
The difference between process mining, data mining, business intelligence and process modelling
These methods look close to one another but answer different questions. What sets process mining apart is that it looks not at individual numbers but at the order of events and the time between them.
| Method | Question it answers | Input | Output |
|---|---|---|---|
| Process mining | How does the process really flow, and where does it wait? | Event logs with case, activity and timestamp | The actual process map, variants and deviations |
| Data mining | What patterns and relationships are in the data? | Large data sets | Classification, clustering and prediction models |
| Business intelligence | Where do the indicators stand? | Summarised transaction data | Dashboards and reports |
| Process modelling | How should the process work? | Interviews and design | A target process model |
Aren’t ERP reports enough? How process mining differs from ERP reporting
ERP reports show the result of a moment or a period: open orders, late deliveries, overdue receivables. These results are needed to run the company, and process mining does not replace them. The difference lies in the question: a report says how much and in what state; process mining looks at the time order of the same records and shows by which path the result was reached.
An ERP report rarely shows a non-standard flow, because it lists a document in its final state: an order approved after the fact or a due date changed twice appears as a clean line. Process mining reads these intermediate steps from the document flow and the change logs. The two are used together: the report points to the problem, process mining finds its cause in the flow.
| Question | ERP report | Process mining |
|---|---|---|
| How late was it? | Lists the late documents and the overall status | Shows at which step and in which wait the delay built up |
| Which path did it take? | Shows the final state of the document | Shows every variant of the same process and its loops |
| Were the rules followed? | Needs a separate control report | Compares the actual flow with the target model; finds skipped steps and steps done after the fact |
| Did a change work? | Compares the results before and after | Shows whether the flow itself changed, or whether the wait merely moved to another step |
ERP guide: what ERP is and how to choose one
What data does it need?
At minimum three fields: a case ID, an activity name and a timestamp. Further fields such as resource, cost or customer deepen the analysis but are not required. These records usually already exist in the ERP’s document flow, change logs and approval tables; no new system has to be installed.
The hard part is not finding the data but defining the case correctly. Whether you follow an order line, the whole order or, in manufacturing, the production order depends on the question asked; a wrong case definition draws a wrong map from the right data.
Check how ready your data is for process mining: Process Mining Readiness Test
What to look for when choosing process mining software
Which software fits depends on the question you want answered and on where your data sits. When comparing, these points decide:
Tool classes also answer different questions. In the table below, process mining is the only class that measures the actual flow; the others complement it.
- data intake: how the event log is taken from your ERP, which file formats are supported, incremental loading,
- working with modelling: whether the discovered flow can be compared with a BPMN target model,
- simulation: whether an improvement can be tested before it goes live,
- data location and access: in which region the data is hosted, authentication and role-based permissions,
- ease of use: whether business analysts can use it without specialist support, and the interface languages,
- integration: APIs and ways to connect to other systems.
| Tool class | What it does | Strength | Watch out for |
|---|---|---|---|
| Process modelling tool | Draws and documents the process in a shared language such as BPMN | Shared language, approval and governance | Does not measure the actual flow; the drawn and the lived process can diverge |
| Process mining platform | Derives the actual flow, variants and bottlenecks from event logs | Data instead of opinion; conformance and deviation measured | Needs a sound case definition and a clean event log |
| Business intelligence and reporting | Tracks results in indicators and tables | Widely used, flexible dashboards | Shows what happened, not at which step the process waits |
| Task mining | Records the steps a user takes on screen | Makes manual, repetitive screen work visible | Needs desktop recording; employee privacy must be assessed separately |
Why do process mining projects fail?
Process mining projects usually fall short not because of the software but because of missing preparation and ownership. The typical causes, how to spot them and how to prevent them:
What they have in common: the analysis should start with a question and end with a decision. A map is not a result on its own; the result comes from applying a change the map points to and measuring again with the same measure.
| Cause | How to spot it | Prevention |
|---|---|---|
| Event log quality | Missing or inconsistent timestamps, the same activity recorded under different names, document numbers that do not match across systems | Data preparation is planned as a separate piece of work before the analysis; activity names and the case definition are settled together |
| Wrong case definition | The map shows loops or broken flows that do not exist in reality | The case is chosen for the question asked and confirmed with the unit that runs the process |
| A process without an owner | Findings are presented, but nobody takes on the change | For every process examined, an owner who can decide is named from the start |
| Analysis that never becomes a decision | Maps and dashboards are produced, nothing changes in how work is done | Every finding closes with a decision, an owner and a new measurement |
| Scope too wide | All processes are tackled at once, none in depth | Start with one process that leaves records and is costly when it stalls; widen the scope after the first result |
| Field knowledge left out | Findings are misread because of steps that leave no record in the system | Findings are read together with the teams that run the process |
AI and process mining
AI contributes to process mining in three places. Anomaly detection highlights cases that depart from the usual flow, for example orders that move in an unexpected order or wait unusually long. Prediction estimates the risk of delay or the likely next step of an open case from the trail of past cases. Natural-language queries let an analyst ask a question such as «which orders waited longest for credit approval?» without writing a query language.
Its limits should be known just as clearly. AI cannot see a step that is not in the event log, and a prediction drawn from incomplete or inconsistent records is only as reliable as those records. A prediction is a probability, not a certain result: it should be possible to explain why an alert was raised, and the decision stays with a person. Answers given in natural language are also checked against the underlying data. Sending records that contain personal data to an AI service is handled separately, as part of the personal data assessment.
Agentic AI guide: AI agents, risks and controls
The limits of process mining, and personal data
Process mining sees only the steps that have a record in the system. An approval given by phone or a check carried out on paper does not appear on the map; that is why the findings are interpreted together with the teams on the ground.
Event logs often also contain information about the user who carried out the transaction. The purpose of the analysis is therefore defined as process improvement, personal data is anonymised or aggregated unless it is needed, and the assessment under the Turkish Personal Data Protection Law (KVKK) is made before the analysis.
ProcessMind: mining, modelling and simulation on one platform
ProcessMind is a process intelligence solution that brings process mining, BPMN modelling, process architecture, simulation and operational dashboards together on one cloud platform. It derives the actual flow, variants and bottlenecks from event logs, and models the target process and tests it by simulation before it goes live.
The vendor positions it as a self-service process mining platform for small and mid-sized companies. The interface and documentation are available in ten languages, Turkish included; data is hosted in the AWS Frankfurt region. It also draws its scope clearly: it offers process analysis and operational process simulation, not machine-level or PLC analytics.
Where to start
With the question, not the software: knowing in which process which delay or deviation is costly is the first step. Then the case definition that will follow that process is set, and where the data sits is established.
MindDX does this work in Assess. Processes and the records of the existing systems are reviewed; in which process and with which data process mining starts is written into the Validated Scope document.