Who owns the process?

1. The bill for an unowned process

Every business process worth the name, Procure-to-Pay, Order-to-Cash, plant maintenance, crosses functions, and no single function owns any of them.

Functions are owned. Processes are not. Geary Rummler and Alan Brache called the gaps between functions the “white space” on the organisation chart: unowned, and therefore unmanaged. Work that leaves one department is nobody’s responsibility until it arrives at the next. Fewer than half of large organisations have an end-to-end owner for Procure-to-Pay (APQC). That is the diagnosis, and it has been known for thirty years. What follows is the bill.

It costs cash every day, in a place no report shows

Impira’s Australian bands put the fully loaded median cost of a purchase order at A$60–100 and the top quartile under A$40. APQC’s Open Standards data reports a median of US$55 per purchase order (n=4,622). Much of the gap between the median and the top quartile is in the handoffs between procurement and accounts payable: the approval by email, the match done in a spreadsheet, the re-keying between two systems. The remainder is catalogue compliance and touchless matching, which are also easier to fix when one person owns the whole chain. Multiply the gap by annual volume and you have the bill. For an organisation processing 120,000 orders a year, a A$45 gap is A$5.4 million a year. It appears on nobody’s scorecard, because nobody owns the whole process. The extreme case is the debt that accumulates in the white space: an SAP-published case study of an Australian food manufacturer reported A$53 million of goods-received-not-invoiced value identified in one process segment. It had accumulated unnoticed. No system flagged it; someone had to go looking.

It traps working capital

In the processes we measure, only 5–10% of elapsed time is work, in line with the flow-efficiency figures Modig and Åhlström report. The rest is waiting: in an approval queue, in a batch run, in a handoff between two teams that report to different people. Each wait sits in a different function’s queue, so no function sees the total. On Order-to-Cash, the waiting that costs cash is the lag between delivery and invoice and the time a disputed order or credit re-issue sits unresolved. Each day of that lag releases the affected revenue ÷ 365, once; on A$1 billion, one day across the whole book is A$2.7 million, so even a fraction of a day matters.

It turns improvement programmes into reports

Process mining and process intelligence tools can now show an organisation exactly how its processes run. That evidence is worthless if nobody owns the process it describes. One practitioner in a recent study of process mining put it plainly: without owners, process mining “can end up just being a report.” The findings arrive, it is nobody’s job to act on them, and the report ages on a shelf. The improvements that do get made fade the same way: a system upgrade pushes work back into spreadsheets, and nobody is measured on noticing until the next annual review.

It carries the workarounds into your new ERP

Horváth’s 2025 survey of 200 S/4HANA programmes finds fewer than one in ten finish on schedule and about two thirds report significant budget overruns. The most-cited cause is process complexity discovered late. It is discovered late because nobody could say what the process was before the programme started. Without an owner, every gap between the process and the vendor’s standard is found during the build, at the most expensive moment, and the programme defaults to funding the extension.

It is why the AI investment returns nothing

BCG puts 70% of the AI problem in people and process. Gartner expects more than 40% of agentic AI projects to be cancelled by 2027 for unclear value and inadequate risk controls. An agent acts on the process it is given. Give it an unowned process and it inherits the workarounds and optimises the fragment it can reach. Australia’s Voluntary AI Safety Standard, ISO/IEC 42001 and the NIST AI Risk Management Framework each require named accountability for an AI system and its use. Where there is no owner, that accountability is unassigned, and the exposure is the organisation’s.

It lands on IT as fire, and on the auditor as findings

Problems in the white space reach IT only when they break. An email approval chain adds four days to a process for two years, and the first anyone hears of it is a complaint about the system. The same off-system approvals and spreadsheet matching are control weaknesses, and auditors find them eventually. Meanwhile the executive table argues from three versions of the truth: Finance talks budget, IT talks technical debt, operations talks headcount, and investment follows whoever argues best.

An unowned process is a line of cost no budget shows and a migration no one can scope.

Size your own bill in ten minutes

Ask three questions in your next executive meeting. Who owns Procure-to-Pay, end to end? The pause before the answer is the first finding. What does one purchase order cost us, fully loaded? The guess is the second. How much of that process actually runs in the system we paid for? Nobody will know, and that is the third. Then do the arithmetic: annual purchase-order count × (your fully loaded cost per order − A$40). If you cannot fill in your own cost, use A$85, the APQC median converted, and label the result indicative. The rest of this guide is about turning those three findings into an owner with a scorecard.

2. Why the role fades

If the bill is that large, why has appointing an owner not settled it? Most organisations have tried. APQC has tracked the obstacles to process ownership for years. We group what they and we see into five design faults. Each has a design fix.

2.1 The role is part-time

The owner is a functional head with a day job. Process ownership is added on top. When the two conflict, the day job wins, because the day job is what the owner is measured and paid on.

2.2 The owner has no authority

Hammer’s process maturity model (PEMM) rates the process owner on identity, activities and authority at four levels. At the first, the owner is informal and can only lobby. At the second, the role is official, held by “a senior manager who has clout and credibility” who can convene a redesign and has some say over the technology budget. At the third, the owner controls the systems and projects that change the process and gives it priority over functional duties. At the fourth, the owner controls the budget and influences how people in the process are evaluated. Most organisations appoint at level one and expect level-three results, and a role created without authority fails to achieve its purpose.

2.3 The metrics belong to the functions

Procurement is measured on price variance. Accounts payable is measured on invoices per FTE. The plant is measured on wrench time. Each metric is rational inside its function and blind to the process. Hammer called these the sins of provincialism (letting functional boundaries set the metrics) and pettiness (measuring a small piece of what matters). A process owner handed functional metrics has nothing to own.

2.4 The scorecard is too big to carry

When organisations do build process metrics, they build too many. The SCOR supply-chain model has about 250 metrics. APQC’s measure library has thousands. Process intelligence tools expose dozens of configurable KPIs per process. A senior executive who is offered forty numbers will own none of them.

Notice what every mature standard does at the top. SCOR collapses its 250 metrics into its classic five level-one attributes: reliability, responsiveness, agility, cost, and assets. APQC distils its accounts payable library into four core KPIs. Lean shop-floor boards run on four letters: safety, quality, delivery, cost. The pattern is consistent. Executives act on a vital few. Everything else is diagnostic detail for the people who work in the process.

2.5 There is no forum

An owner with a scorecard and no meeting has a hobby. Ownership needs a place where the numbers are reviewed, decisions are taken and the owner is seen to be accountable. Annual reviews are too slow. A steering committee that meets quarterly to look at last quarter’s data is a history lesson.

Five faults, one root cause: the role is designed as a title, without the scorecard, authority and forum that make it a job.

3. What a working process owner needs

Strip the literature back and a process owner needs three things. None of them is a title.

Authority and forum are governance decisions. The executive sponsor grants them or does not. The scorecard is a design problem, and it is the one that has been solved badly for thirty years. The rest of this guide is about solving it, and about how a well-designed scorecard makes the other two easier to grant.

One more point before moving on. Process ownership is a part-time role with a real cost. Expect one to two days a month from the owner in the first quarter and about one day a month after that, plus a defined block of time from their team during the baseline (Section 8 gives the figures). A move to Hammer’s third level, where the process takes priority over the owner’s function, is a separate decision in quarter two, with its own role statement, and the owner can decline it. The design goal is a role a busy executive will accept with that commitment stated up front, not a role that needs a new department and not one sold as a few hours a month.

4. Portfolio Process Mining, from the beginning

This section introduces the method. Readers who know process mining can skip to 4.2.

4.1 Process mining in one paragraph

Every transaction in an enterprise system leaves a timestamped record: a purchase order created, approved, changed, received, invoiced, paid. Process mining reads those records and reconstructs how the process actually ran, case by case, rather than how the process manual says it runs. It shows every path, every loop, every wait and every rework step, with volumes and timings attached. Tools such as SAP Signavio and Celonis do this well and at scale. The result is a true picture of the process as the system saw it.

The limitation is in the last phrase. Process mining reads what the system records. Most process waste never reaches a system: the approval by email, the spreadsheet used to match invoices, the phone call to chase a delivery, the re-keying between two applications. A system-only view does not just miss this work. It reports the process as healthy. Impira’s reference Procure-to-Pay data set, a three-region manufacturer with about 120,000 purchase orders a year, modelled on client engagements, shows the effect. All three regions read close to 100% digital on the system extract. After the off-system work was documented and added back, one region fell to 40% digital. The activity view of cost showed why: system touches account for A$9–13 of the A$140 fully loaded cost per order in that region. The system-only cost counts only the cost of on-system activities; the enriched cost adds the off-system effort documented in the sessions. The reconciled cost to serve, the process functions’ cost pool divided by orders and reconciled to the general ledger, does not move with enrichment. The activity view shows where that cost is consumed, including in cost centres outside the process functions.

4.2 What Portfolio Process Mining adds

Portfolio Process Mining is Impira’s method for turning process mining into a management system. It runs on the process intelligence platform an organisation already has. It adds four things.

• Enrichment. Your process analysts and ours run short, structured sessions with the people who do the work and capture each off-system step, with its evidence, as a separate, tagged event source in the data model (system, observed or reported). Nothing is written to the system log, so every metric can be read with and without the enriched events. The picture becomes complete, and it stays auditable. Appendix C states the lineage.

• Four metrics. Every process, every variant and every business unit is measured on the same four numbers, described below.

• A diagnostic. Three of the four numbers, read together, place each process in one of eight patterns; the fourth flags quality risk across all of them. Each pattern calls for a different intervention.

• Roll-up and continuity. The same four numbers are read at every level from variant to process to business to enterprise, aggregated as volume-weighted ratios. System-recorded events update as transactions flow; enriched off-system events are dated to the session and re-validated each quarter. Improvement and regression are visible in weeks, not at the next annual review.

4.3 The four metrics

The process owner carries all four numbers for their process. The executives named here read one number across every process, sponsor it at the executive table, and arbitrate when it conflicts with a functional KPI.

Three of the four are established measures with decades of use behind them. Digital Work Ratio is Impira’s addition, and it exists because the off-system blind spot is the largest source of hidden cost in most processes we see. It is not automation rate or touchless rate. Those measure how much on-system work runs without a human. DWR measures how much of the work reaches the system at all, and a process can be 90% touchless on the 40% the system sees. It is also the first gate for AI: an agent acts on the events it can reach, so DWR tells you how much of a process leaves a record an agent can be governed and measured against. Appendix C gives the formal definition of each metric, its known weaknesses and the control for each.

Read together, the four numbers describe a process the way a doctor’s four vital signs describe a patient. High DWR with low CTS and high CTE is a healthy process: protect it and copy it. Low DWR with high CTS and low CTE is a process that has left the system, costs too much and waits too long: rebuild it before you automate it. High DWR with high CTS is a process that is used but over-engineered: simplify it. The eight patterns and their interventions are in Appendix A.

Four numbers for every process, at every level of the organisation. That is the whole scorecard, and it is small enough for an executive to carry.

5. How the four metrics make ownership work

Return to the three things an owner needs. Portfolio Process Mining solves the first directly and makes the other two much easier to grant. It does not replace them, and this section is honest about that.

5.1 A scorecard a cross-functional executive can own

The owner of Procure-to-Pay does not need to run the matching rules or the approval workflow; the people in the process do that. The owner needs to know that DWR is 58%, CTS is A$85 per order against a benchmark of A$40, CTE is 6% and FTR is 71%. And they need to know which way each is moving. That is a scorecard an executive will accept. It is four numbers. Each answers a question the CFO, CIO or COO gets asked about the process. None can be improved by one function at another’s expense. The C-suite reads one number across all processes; the owner carries all four for one process.

5.2 Metrics that are hard to game, and where gaming shows up

Three of the four are end-to-end by construction. Procurement cannot improve cost to serve by pushing work onto accounts payable; the cost moves but the total does not. The plant cannot improve cycle time efficiency by closing work orders early; first time right falls. This is Hammer’s provincialism and pettiness designed out, and it is what makes the owner’s position tenable in a room of functional heads: the owner reads the process number that sits above everyone’s functional numbers, and second-guesses none of them.

Hard to game is not impossible to game, and a CFO will ask. Cost to serve moves if you change the cost pool, split transactions, or move spend out of scope onto purchasing cards. Cycle time efficiency moves if you change where the clock starts. First time right cannot count rework that happens off-system, so a process with low DWR will over-report its quality until it is enriched. And Digital Work Ratio is the exception to end-to-end: its denominator is built from the enrichment sessions, so a thinner enrichment pass raises it without anything changing. The controls are simple and they are written into the charter: definitions and scope are frozen at baseline and changes are restated; every metric is published twice, system-only and enriched, so the size of the off-system layer is always visible; the enrichment protocol is fixed and repeated the same way each quarter; and DWR is a diagnostic, never an incentive. Appendix C sets out each metric’s definition, failure mode and control.

5.3 A common language for the executive table

Picture the conversation. The CFO asks why cost to serve is A$85 per order against a benchmark of A$40. The CIO answers that DWR is 58% and a third of the work is in Excel. The COO adds that CTE is 6% and approvals are the bottleneck. That is one problem seen from three chairs, not three arguments. The same four numbers roll up to the view the CFO uses to allocate capital, and drill down to the variant a team leader uses to fix a step. They roll up as ratios, volume-weighted, never as averages of averages. Cost to serve rolls up within a process; across processes with different transaction units it becomes a dollar gap to benchmark, not a rate. Everyone argues from one page.

5.4 Targets and a forum come with the method

Cost to serve is benchmarked externally and the other three have maturity bands. So every owner starts with a target for each metric and, on cost to serve, a gap in dollars. Because the system-recorded events update continuously, the forum has something new to read every week; the enriched off-system events are dated and re-validated each quarter. A weekly trend review with process owners, a monthly intervention review, and a quarterly portfolio review, both chaired by the executive sponsor, is enough. Each intervention is registered before it starts with the metric it should move, the size of move expected, and the period needed to see it at that process’s volume. The forum judges it against that, not against last week’s number. Reacting to ordinary week-to-week variation is how measurement programmes make processes worse.

5.5 What the metrics do not do

They do not confer authority. In Hammer’s model a scorecard fixes the Metrics enabler; the Owner enabler moves only when the sponsor grants identity and authority. An official appointment by the executive, with authority over the process rules within the delegations framework, takes the owner to the second level on identity and activities. A say over the systems and projects that change the process takes their authority there too, and toward the third. Those are decisions the sponsor makes, and Section 6 asks for them explicitly. Measurement without accountability produces what practitioners call watermelon metrics: green on the outside, red inside. The four numbers make ownership tractable. The sponsor makes it real.

6. How to stand up process ownership in 90 days

This is the sequence Impira uses. It is designed so that the organisation names owners and quantifies recoverable value in the same quarter, and so that ownership matures on evidence rather than on a restructure. It gives the owner authority over the process rules, within the delegation of authority, on day one, and asks them to earn authority over budget.

6.1 The owner charter

Each appointment should fit on one page. Impira uses the template below. The form matters less than the content: the four numbers, the decision rights, the delegation of authority, the counterparts, the conflicts and the forum are written down and signed by the sponsor, the owner and each counterpart before the baseline starts.

6.2 For the person being appointed

If you are the executive being asked to own a process that runs through functions you do not control, ask for these before you accept.

1. Counterparts named and signed by the sponsor: one person in each function, carrying the numbers for the segment their function runs.

2. Authority over the process rules within the delegation of authority: exception routing, step sequence and rework rules from day one, and a standing right to propose threshold changes to the CFO, answered within the monthly review. Most of cycle time lives there.

3. You review the baseline, and how it was measured, before it goes to the executive.

4. The KPI-conflict statement, and the precedence of the four numbers for the term, agreed with your line manager before you accept.

5. Your time commitment in writing, including after the first quarter: one to two days a month in the first quarter, about one day a month after.

6. Worklist recovery effort sized from the week-3 list and released to your teams by Finance. It is additional to the baseline effort.

7. The analyst run cost in the sponsor’s budget, or the process CoE’s, before the sprint starts.

8. A two-way review at day 90. The sponsor can extend; you can step back.

An owner who accepts without these is accepting the title, and Section 2 describes how that ends.

6.3 If you are about to migrate your ERP

A core-system migration is the strongest forcing function for process ownership an organisation will ever have, and the most expensive place to discover it is missing. Modern programmes run fit-to-standard: each process is compared with the vendor’s standard, and every gap is either closed by changing the process or funded as an extension. Without an owner, nobody is accountable for that choice, and the programme defaults to funding the gap. Horváth’s 2025 survey of 200 S/4HANA programmes finds fewer than one in ten finish on schedule and about two thirds report significant budget overruns, with process complexity discovered late as the most-cited cause.

The four numbers do not replace the programme’s own scoping method. They tell it where to look. A process with low DWR and high cost to serve is where fit-to-standard workshops should start, and where an owner should be in the room. A process at the top of the bands is a candidate to adopt standard with minimal debate; the others are re-engineered or retired, with the evidence on the table. When an owner exists, the programme has someone who can decline an extension request with evidence. During a migration, the monthly intervention review reports into the programme’s design authority. It is not a second steering committee. Baseline before scoping locks. Re-baseline after cutover, because the events the new system records are not the same events, and only a fresh baseline makes before-and-after comparable.

If you are already in Realize, the baseline serves the next wave and the cutover: take it on the processes still to be scoped, and on the first wave’s processes as the pre-cutover reference for hypercare.

6.4 If you are investing in AI

The larger forcing function is the AI programme most large organisations are now funding. The money is going in and the value is not coming out. MIT NANDA’s 2025 GenAI Divide report, from about 150 executive surveys and 50 interviews, found 95% of organisations reporting no measurable return, and named the cause: tools that “integrate poorly, or match workflows” badly, rather than weak models. BCG puts 70% of the AI challenge in people and process and 10% in algorithms. McKinsey found that redesigning the workflow is the single attribute most associated with profit impact from generative AI, and that only 21% of organisations have done it. Gartner expects more than 40% of agentic AI projects to be cancelled by 2027 for unclear business value and inadequate risk controls.

An AI agent needs three things from the process it is put into, and they are the same three things a process owner needs. First, the process must be visible. An agent acts on the events it can reach. At 40% on-system, 40% of the work leaves a record you can govern, audit and measure. An agent in the mailbox may touch the rest, but nobody can show what it did there, and no governance framework will accept that. Work that lives in email and spreadsheets has no system record for the agent to act on or for you to audit. Digital Work Ratio is the first gate, and almost nobody measures it. Second, the process must have an owner. The frameworks Australian boards are asked about each require named accountability for an AI system and its use: Australia’s Voluntary AI Safety Standard (guardrail 1), ISO/IEC 42001 (clause 5.3), the NIST AI RMF (GOVERN 2.1) and, where it applies, human oversight under EU AI Act Article 14. None of them says ‘process owner’. In our view, for an agent inside a business process, all of them need one. Bain’s advice on agentic transformation is blunter: charge general managers with the targets, not the CIO. Third, the process must have a baseline. Both the Australian and UK government Copilot trials measured self-reported time savings and stated plainly that they collected no data on organisational outcomes. Adoption is not value. The four numbers, taken before the agent arrives, are how an executive shows what it did to cost, cycle time and rework afterwards.

The practical sequence is the one in this guide, with one addition. When the owner’s scorecard is set, mark each process by pattern. Processes in Protect & Scale or Streamline Flow have the on-system share an agent needs; the AI programme’s own risk and integration checks still apply. Processes in Re-engineer or Adopt & Train do not; bring the work on-system first, or the agent inherits the workarounds. This is also the argument for the platform investment already made: the process intelligence vendors position their platforms as the context layer for agents, Celonis with AgentC and SAP with Signavio as the process context for Joule. The platform is already there. What is missing is the owner and the baseline.

7. Objections we hear, and what we say

Process ownership fails when it is a title. It works when it is four numbers, a signature and a meeting.

8. Where to start

The fastest way to test this is a Baseline Sprint: two or three processes, four weeks, on the process intelligence platform you already run. Owners are named and charters issued with the scope note, and signed at kick-off. The sprint baselines the four metrics, and recoverable worklists reach the owners in week three. It closes with targets proposed for the owner’s signature at the first monthly review, a quantified roadmap, and the metric definitions written down so that your team can re-measure without us. Value is banked when the Finance sponsor posts it to the benefits register, typically weeks 8–12.

The sprint needs Signavio Process Intelligence (or Celonis) with custom event log entitlement. If you hold Process Insights only, we confirm the uplift with your SAP account team before the scope note, and it is your cost, not ours.

It is priced as a fixed fee, paid on accepted deliverables, and it asks the following of you. The figures are typical; the scope note fixes them.

If you would like to talk through how this would apply to your processes, contact David van den Berg at david@impira.com.au.

Appendix A. The eight patterns

High and low states of DWR, CTS and CTE place a process in one of eight patterns within four quadrants. The cut points are 60% for DWR, the APQC or Hackett median for the process for CTS, and 10% for CTE; a process near a cut point is read on its trend, not its position. FTR is deliberately held outside the eight patterns: it flags quality risk across all of them rather than multiplying the patterns to sixteen. The patterns also read as an AI-readiness map: a process in Protect & Scale or Streamline Flow has the on-system share an agent needs, and the AI programme’s own risk and integration checks still apply; a process in Re-engineer or Adopt & Train does not, and an agent deployed on it will optimise the fragment it can reach.

Appendix B. Hammer’s process ownership ladder

Adapted from Michael Hammer, “The Process Audit”, Harvard Business Review, April 2007. Use it to be honest about where each owner sits, and to plan the next grant of authority.

Appendix C. Metric definitions and controls

This appendix is for the CFO, the CIO and the architecture review. It states each metric precisely, names the way it can be moved without the process improving, and the control that stops that. Definitions are frozen at baseline and any change is restated.

Aggregation. All four are ratios. They aggregate only as the ratio of summed numerators to summed denominators, weighted by volume, and an aggregate can move against every one of its components when the mix shifts. Cost to serve aggregates within a process family only; across processes with different transaction units, report total process cost and the dollar gap to benchmark.

Hierarchy. The owner sits at the end-to-end value stream as defined in the client’s process architecture (Procure-to-Pay, Order-to-Cash). Each counterpart owns the sub-process their function runs. Metrics are diagnosed at the sub-process level (PCF level 3 or the client’s equivalent) and interventions at variants. Each metric, its definition and its owner are attached to the repository element by its identifier; the enriched as-is variant is recorded against the same element; the CoE owns the definition change log.

Enrichment protocol. Nothing is written to SAP or to the system event log. Off-system steps are captured per variant as step definitions (actor, frequency, duration range, evidence) and loaded as a separate event source, tagged system / observed / reported, joined to cases by variant in the process data model and excluded from conformance checks. Enriched events are dated to the session, held as a custom event source on your tenant, written by named analysts, versioned by quarter; retention and re-validation are stated in the specification. Impira holds no copy after handover. Every metric is computed on the system source alone and on both together. Reported durations carry a range. A sample of reported steps is corroborated against aggregate, de-identified counts (shared-mailbox volumes, ticket counts, file timestamps) agreed with HR and Privacy before the sessions. Two analysts enriching the same variant should land inside an agreed band. The protocol is repeated the same way each quarter. Enrichment data is not used to assess the people who do the work or to build a headcount case, and the charter says so.

Verifying interventions. Each intervention is registered before it starts with the metric it should move, the expected size of the move and the observation period needed to detect it at that process’s volume and variation. It is judged when the period elapses or the signal clearly exceeds the baseline range, whichever comes first. Week-to-week noise is not a verdict.

Sources

• Rummler, G. and Brache, A., Improving Performance: How to Manage the White Space on the Organization Chart (Jossey-Bass, 1990; 3rd ed. 2012).

• Hammer, M. and Stanton, S., “How Process Enterprises Really Work”, Harvard Business Review, November 1999.

• Hammer, M., “The Process Audit”, Harvard Business Review, April 2007 (the PEMM maturity model).

• Hammer, M., “The 7 Deadly Sins of Performance Measurement and How to Avoid Them”, MIT Sloan Management Review, Spring 2007.

• APQC, “Why Procure-to-Pay Needs End-to-End Ownership”; “Answering the Challenges of Process Ownership”; “4 KPIs That Set Good Accounts Payable Organizations Apart”; Open Standards Benchmarking, cost per purchase order (median US$55, n=4,622); Process Classification Framework.

• Supply Chain Council / ASCM, SCOR model: level-1 performance attributes.

• Mamudu, A. et al., “Process Mining Success Factors and Their Interrelationships”, Business & Information Systems Engineering (2024).

• Horváth & Partners, “SAP S/4HANA transformations rarely go as planned” (April 2025, n=200).

• SAP, customer story: “GWF: Unlocking working capital through new levels of transparency, accountability, and insight into a critical financial process” (August 2025), on goods-received-not-invoiced value identified with SAP Signavio: www.sap.com/asset/dynamic/2025/08/d8f5cd33-1a7f-0010-bca6-c68f7e60039b.html

• Modig, N. and Åhlström, P., This is Lean (Rheologica, 2012), on flow efficiency. ASQ Quality Glossary, first-pass yield.

• MIT NANDA, The GenAI Divide: State of AI in Business 2025 (July 2025). BCG, Where’s the Value in AI? (October 2024). McKinsey, The State of AI (March 2025). Bain & Company, State of the Art of Agentic AI Transformation (September 2025).

• Gartner, press release on agentic AI project cancellations (June 2025). Celonis, AgentC announcement (October 2024). SAP, “Welcome to the (Process) Conversation: Joule with SAP Signavio Solutions Now Generally Available”, SAP News, 10 February 2026, on SAP Signavio as the process context for Joule.

• Australian Government Digital Transformation Agency, Evaluation of the whole-of-government trial of Microsoft 365 Copilot (October 2024). UK Government, Microsoft 365 Copilot cross-government experiment (2025).

• Australian Government, Voluntary AI Safety Standard (August 2024). NIST AI Risk Management Framework 1.0, GOVERN 2.1. ISO/IEC 42001:2023, clause 5.3. EU AI Act, Article 14.

• Impira Pty Ltd, Portfolio Process Mining: a data-driven framework for application and portfolio management (2026).