Power Automate Process Mining: Event Logs, Licensing and a First Project Plan (2026)

Last updated: November 6, 2026

By ZapAI Team

Power Automate process mining reads the event history your systems already keep and draws the process people actually run, with every detour, loop and wait. It sits inside the platform our Power Automate consulting team builds on every week, and it’s licensed through Power Automate Premium. Task mining covers the desktop clicks that never reach a system log. As of November 2026, both are generally available, object-centric process mining is too, and Copilot can answer plain-English questions about a mined process.

Most automation programs pick their first target by gut feel. Somebody loud complains about invoices, so invoices get a flow. Sometimes that works. Sometimes it doesn’t, and six months later the real delay turns out to be three steps upstream, in a handoff nobody thought to measure.

Mining ends the guessing. This guide is for operations, finance and IT leads who want to know what the tool does, what it costs, and how much data work it really takes, with every product fact checked against Microsoft Learn and Microsoft’s pricing pages in November 2026. Our framework for automating business processes on Power Platform covers scoring and building. This post covers the step before. Seeing what’s real.

What Is Power Automate Process Mining?

It’s a Power Automate capability that extracts event data from your systems of record and visualizes the process behind it. Think X-ray, not org chart.

Microsoft’s process mining overview lists the core jobs. Visualize processes. Compare them. Find root causes. Monitor KPIs. Spot noncompliant steps.

In practice, a first analysis gives you five things.

• A process map with every activity and the paths between them
• Variants, meaning each distinct sequence a case followed
• Durations per activity and per handoff, which is where bottlenecks show up
• Rework and self-loops, like an invoice bouncing back to the same approver
• KPIs and filters by attributes such as vendor, region or amount

Two surfaces exist. The web experience in the Power Automate portal handles setup, ingestion and an embedded Power BI report. The Process Mining desktop app adds the deeper tooling, and Microsoft’s desktop app overview lists process animation, statistics, root cause analysis, variants, process compare, filtering, custom metrics and business rules. Full use needs Premium.

Conformance lives in process compare. You can layer two views of the same process, say this quarter against last, or import a properly formatted BPMN diagram and compare the mined map against it. Where your documented process and the data disagree, the data wins. Every time.

How Is Task Mining Different From Process Mining?

Process mining reads system logs. Task mining records what people do on their desktops, step by step, using Power Automate for desktop. Use the first to find the slow stage, and the second to see the clicks inside it.

Per Microsoft’s process mining and task mining overview, task mining captures recorded user actions, then produces a process map, an application analytics report showing where people spend time, and automation recommendations that suggest which connectors a flow could use. The recorder is current, not retired. Microsoft’s licensing page still names “process and task mining” as a Premium entitlement.

Recordings are small samples, though. They show how work gets done. Not how frequently. That’s the event log’s job.

Our rule of thumb. Mine the system first. Record desktops only for steps the map flags as slow and manual.

What Data Does Process Mining Need?

Three fields at minimum. A case ID for each instance of the process, an activity name for each step, and a start timestamp for each event. An end timestamp, resources and attributes are optional, and they make the analysis far more useful.

Microsoft’s data requirements page defines the attribute types.

Types matter. Case ID and activity name must be text, timestamps must be Date/Time, and a case needs at least two activities before the analysis will run. Microsoft’s troubleshooting guide also flags a classic trap where day-first dates fail to parse in a US-locale process, fixed by switching the Power Query locale.

Volume matters too. Rows past the ingestion limit get ignored. The transform and map data guide shows trimming to the latest 150,000 rows, but the pages we checked don’t print the limit. Test your extract.

Where Can the Event Log Come From?

Almost anywhere Power Query can reach, plus two direct lake routes. A CSV export is the fastest way to start. SQL Server, Dataverse through OData, Azure Data Lake Storage Gen2 and Fabric OneLake are the routes that keep a process refreshed.

Two details surprise people. Microsoft says Power Platform doesn’t support the Dataflow connector or the Dataverse connector as a process mining source, so Dynamics 365 data comes in through OData instead. And the lake routes skip Power Query entirely. Per Microsoft’s guide to ingesting from Fabric OneLake, the files or Delta tables must already be in final event log shape, timestamps in ISO 8601, header row under 1 MB, and incremental files named in date order.

Scheduled refresh runs once a day at most. Need it right after a nightly load? Microsoft documents a cloud flow that calls the Analyze bound action on the process, so a pipeline can trigger a refresh the moment data lands.

So at larger volumes, build the log in a Fabric lakehouse and point process mining at the finished table. Getting Dynamics 365 data there first? Our guide to Dataverse Link to Fabric and its alternatives covers getting it there.

Case-centric or object-centric?

Order-to-cash rarely fits one case ID. One order ships in three parts, two orders share an invoice, one payment clears four invoices. Force all that into “order number” and events get duplicated or lost.

Object-centric mining fixes that. Events can link to several objects at once. Microsoft’s object-centric process mining overview recommends starting case-centric when one case notion is clean, and moving to object-centric when the insight depends on how objects relate. It ingests from Azure Data Lake or OneLake. Status is murky. Microsoft’s message center announced general availability for May 1, 2026, a roadmap entry said June, and Process Intelligence Studio for object-centric models was due to reach general availability on October 31, 2026.

How Do You Build an Event Log From Dynamics 365 or Business Central?

You assemble it. Neither system keeps a ready-made event log for order-to-cash or accounts payable, so you stitch one together from document tables, posted documents, ledger entries and approval entries, each contributing one activity with its own timestamp.

That stitching is most of the project. Microsoft’s data page admits as much, noting that most application tables hold current state rather than history, and that the events you want may sit in several tables that have to be joined first.

Business Central

Here’s the shape we use for order-to-cash in Business Central. It’s ZapAI practice, not a Microsoft template, so validate every table against your own configuration and extensions.

Why SystemCreatedAt? Because posting dates in Business Central carry a day, not a time, and a process map built on dates alone can’t tell which of two same-day steps came first. Microsoft’s table system fields reference confirms every table carries SystemCreatedAt and SystemModifiedAt as date-time audit fields, populated by the platform. They record when the row was written. Backdated documents still show the write time.

Business Central has three gotchas. Each one bites.

• Fully posted orders get deleted from the open order table. Pull the order number from posted shipments and invoices, or turn on order archiving.
• Payments don’t point at orders. You walk from payment to invoice to order through ledger entry applications.
• Modified timestamps get overwritten on every edit. Only created timestamps are stable.

Need edits as events, like a changed ship date? The change log can record them. Microsoft warns it costs performance and database size, and advises against tracking ledger entries and posted documents, so log only the few fields you’ll actually mine.

Accounts payable follows the same pattern. Merge the vendor number and the vendor’s invoice number into one case ID in Power Query, then pull activities from the purchase invoice, approval entries, the posted purchase invoice and vendor ledger applications. PDF invoices? Our guide to intelligent document processing with Azure AI covers capture.

Dynamics 365 Sales, Customer Service and other Dataverse apps

Dataverse gives you more raw history, with a catch. Dataverse auditing logs creates, updates and deletes per column once an admin turns it on, but it only covers changes made after that moment, it consumes log capacity, and a few core tables such as ActivityPointer can’t be audited at all. Turn auditing on months before you plan to mine. Not the week before.

Supply Chain Management has the one ready-made route we found. Its warehouse material movement analysis template is a preview that needs version 10.0.35 or later and a linked Dataverse environment. Microsoft’s own docs mention no equivalent template for Business Central order-to-cash or AP.

What Does Power Automate Process Mining Cost in 2026?

Premium covers it, at $15 per user per month paid yearly, which adds 50 MB of process mining storage per license to a tenant pool capped at 100 GB. A Process Mining add-on adds another 100 GB for $5,000 per tenant per month.

Prices come from Microsoft’s Power Automate pricing page, which says they’re for marketing purposes and checkout may differ. The capacity rules come from Microsoft’s Power Automate license types page, whose comparison table marks process mining as included with Premium and not included with Process or Hosted Process. Some third-party guides say the Process plan includes mining, and Microsoft’s own table says it doesn’t.

Now the SMB math. Ten Premium users give the tenant a 500 MB pool. Fifty give it about 2.5 GB. By our arithmetic at list price, each extra gigabyte through Premium seats costs about 20 licenses, or $300 a month. The add-on is cheaper per gigabyte, but it comes as one $5,000 monthly block, so for most smaller firms it only makes sense with very large logs or a tenant already near the 100 GB cap. Size your extract first.

Other costs hide in the corners.

• Dataverse file capacity. Process data uses mostly file storage, per Microsoft.
• A Power BI Premium workspace if you want custom reports
• Fabric capacity, if the event log gets built there
• Analyst time. The biggest line by far.

Our Power Platform licensing guide covers the wider plan structure, and our post on what your Microsoft 365 licenses already cover explains when Premium becomes necessary anyway. Already hold Premium seats? Then trying it costs nothing extra.

What Does Copilot Do in Process Mining?

Copilot answers natural-language questions about a mined process, surfaces top insights and suggests automation candidates. It works in the web experience and the desktop app, in English, and it sees only the most frequent activities and variants.

Copilot’s view is spelled out on Microsoft’s Copilot in process analytics page. Frequencies, durations, case counts and rework statistics for the top activities, edges and variants. Good prompts sound like “which activity has the highest rework percentage?” or “what’s the most common variant?”

The limits are worth reading before a demo.

• No statistics on custom attributes like vendor, financial or resource fields
• Only the most common activities and variants on large processes
• Not available in task mining
• English only
• Admins can turn it off in the Power Platform admin center.

Outside the United States, Australia and the United Kingdom, an admin may need to allow cross-geo calls before Copilot appears at all, which is worth checking before you book a demo with the leadership team. In our view, Copilot is a good first-hour tour guide for business users. The root cause work still happens in filters and comparisons.

How Does Power BI Fit In?

Every process gets an embedded Power BI report you can’t edit. To customize it, connect the process to your own Power BI Premium or Fabric workspace, then build on the published semantic model in Power BI Desktop.

Setup is admin work. Microsoft’s guide to using your own Power BI workspace has you register the Process Insights Prod service principal in Entra ID, turn on two developer settings in the Power BI admin portal, and add that principal as a workspace admin. Microsoft recommends a premium per capacity workspace. Skip the admin role and nothing works.

Worth it? Usually. Dataverse-managed workspaces cap each environment at 1,000 reports, your own workspace removes that cap, and custom reports let you show results in a form finance already recognizes, like aging buckets by vendor, without teaching anyone to read a process map. Our Power BI consultants do this step when a process review becomes a standing dashboard.

What About Privacy and Security?

Access runs on Dataverse security roles. Environment makers create and share processes, the Process Mining User role views reports only, and admins see every process. Task mining recordings can include screenshots and personal data, so plan redaction before anyone records.

Per Microsoft’s process mining security and privacy page, recorders can pause and resume to skip sensitive screens, and delete steps or change input data. After import, owners can rename steps and delete screenshots. Conditional Access policies must include both Power Apps and Power Automate.

Event logs carry risk too. A log with user names as resources is a performance record of individual employees. HR may want a say. Our practice is to mine by role or team unless a named person truly needs to be in the data, and to keep the environment’s maker list short.

How Does It Compare With Celonis and UiPath Process Mining?

For Microsoft shops, Power Automate process mining is the cheaper entry point, since Premium includes it. UiPath and Celonis aim at large ERP estates, especially SAP and Oracle.

Sticking to public documentation, UiPath’s process mining app templates cover purchase-to-pay and order-to-cash for SAP, Oracle Cloud, Oracle E-Business Suite, JDE, NetSuite and Coupa, each with input specs, dashboards and KPIs. Celonis calls its platform system-agnostic, promotes object-centric process mining, and offers task mining. Microsoft has both of those too. Its prebuilt Dynamics content is thin.

For a company running Business Central or Dynamics 365 with a few hundred thousand events a year, our view is that Power Automate is the sensible start. Heavy SAP shops with dedicated process excellence teams should price the others. We covered the RPA side of this choice in Power Automate vs UiPath.

What Should Your First Process Mining Project Look Like?

Pick one process with clean system data and a business owner who wants answers. Order-to-cash and accounts payable are our usual picks. Budget most of the effort for the event log, keep the scope to one entity and twelve months, and end with a short list of fixes.

Here’s how we run it. ZapAI practice, adjust to taste.

1. Pick the process and write one question it must answer, like “why do invoices over $10,000 take longer to pay?”
2. Agree on the case ID and 6 to 12 activities with the process owner.
3. Extract a sample to CSV and load it into a trial or Premium environment.
4. Walk the map with the people who do the work. They’ll spot bad data fast.
5. Fix the extract and rerun. Expect two or three rounds.
6. Move the extract to a scheduled source once the shape is stable.
7. Filter, compare and drill into the top three bottlenecks.
8. Write up findings with an owner and a next step for each.

Set expectations early. The first map looks messy. That’s the point. In our experience, a good share of what looks like process variation in a first pass is really data quality, like missing timestamps, duplicate activity names or test records, so the earliest findings are about the data itself.

A realistic outcome is a ranked list of three to five problems with evidence attached. Not a dashboard. Wondering whether your data and team can support AI work beyond this? Our AI readiness assessment checklist covers it.

How Do Findings Turn Into Flows, Bots or Agents?

Match the fix to the cause. Rules-based handoffs between systems become cloud flows. Repetitive screen work in apps without APIs becomes desktop flows. Judgment-heavy exceptions, like matching a messy remittance, are where AI agents fit.

A few patterns we see repeatedly.

• Approvals stuck waiting on one person. Add escalation and delegation rules in a cloud flow.
• Rekeying between Business Central and another app. Use a connector first, then a desktop flow if no API exists.
• Invoices bouncing for missing PO numbers. Fix capture upstream with document processing.
• Exceptions needing a human read. Consider an agent with human review.
• Policy gaps, like orders shipped before credit checks. Sometimes the fix is training. Not software.

Then measure. Rerun the same process mining analysis after go-live and compare it against the baseline with process compare. That before-and-after map is your best ROI evidence. Our list of Power Automate use cases shows ten fixes. Desktop automation belongs with our RPA services team, and agent work with our AI automation services.

What Belongs on Your Process Mining Checklist?

Licenses, environment, data, people and a question. Check these before the first extract and you’ll save weeks.

• Power Automate Premium for whoever builds and analyzes, or a 90-day trial for the pilot
• A Power Platform environment with a Dataverse database and spare file capacity
• Environment Maker role for builders, Process Mining User for viewers
• One process, one question, one owner
• Case ID agreed and tested for uniqueness
• Activity names mapped to business language
• Real date-time stamps, not posting dates alone
• Twelve months of history, or a documented reason for less
• Test records and cancelled cases filtered out
• Dataverse auditing or the Business Central change log turned on for future history
• Privacy review if user names or recordings are included
• A Power BI Premium or Fabric workspace if custom reports are planned
• Baseline metrics saved for after go-live.

What Could Change After Ignite 2026?

Possibly pricing or packaging. Ignite runs November 17 to 20, 2026, at Moscone Center in San Francisco. Treat the licensing table as a snapshot.

We’ll check whether Copilot gets custom attributes, whether a Business Central template appears, and whether mining capacity moves toward Copilot Credits. Expect a refresh. Late November, probably.

Wrapping Up

Power Automate process mining is the cheapest way for a Microsoft shop to stop guessing about where its processes break. Premium already includes it, extra seats usually beat the add-on below enterprise scale, and Copilot makes first answers quick. The real work is the event log. Build it carefully from created timestamps, agree on the case ID, and walk the first map with the people who do the job.

ZapAI is a Microsoft Dynamics 365, Power Platform, Fabric and AI consulting firm in Irving, Texas, working with businesses across Dallas-Fort Worth and nationwide. Our business process automation team builds event logs from Dynamics 365 and Business Central, runs the first analysis, and turns the findings into flows, bots and agents. Want to find out which process to fix first? Talk to us about a process mining pilot, and we’ll scope the data, the licenses and the question before anyone builds anything.

Questions Teams Ask About Process Mining

Is process mining included in Power Automate Premium?

Do we need the $5,000 Process Mining add-on?

What is the minimum event log for process mining?

Can Power Automate process mining connect directly to Business Central?

Is task mining still available in Power Automate?

How frequently can a process refresh?

Does Copilot in process mining see all of our data?