BPM phases and the enduring legacy of Re-engineering

Close-reading synthesis assembled 2026-04-22 from the wiki's BPM-foundations cluster (Slack 1995 → Dumas 2026). Answers three research questions:

RQ1. What are the phases of Business Process Management, and where is the field heading?
RQ2.1. How does Re-engineering (Hammer 1990/1993) live on today, and what has changed?
RQ2.2. What has not changed, and what can we learn from the 90s BPR experiences?

Framing assumption: Hammer & Champy's Reengineering the Corporation is not directly ingested; it is reconstructed here through the post-hoc synthesis in Reijers & Liman Mansar (2005), the pragmatic continuation in Sharp (2014), and the historical chapter in Dumas et al. (2018).

RQ1 — Seven phases of BPM

The wiki supports a seven-phase reconstruction from operational-management roots to today's agentic frontier. Phases overlap — each absorbs rather than replaces its predecessor.

Phase 0 · Operational-management roots (pre-1990)

The vocabulary BPM inherits — transformation view, five performance objectives (quality / speed / dependability / flexibility / cost), TQM, Lean / JIT, Six Sigma, Little's Law, statistical process control — is established in operations management long before BPM exists as a named discipline. Slack et al. (2007) positions TQM/Lean/Six Sigma as "the improvement canon of OM and upstream influences on BPR and BPM improvement methods". This matters for the BPR reading below: Hammer's radical-redesign rhetoric was a reaction against the incrementalism of TQM and Lean, not a rejection of their metrics.

Phase 1 · BPR / Reengineering (early-to-mid 1990s)

Hammer (1990, 1993) calls for radical, top-down, process-centric redesign — obliterate, don't automate. We see this era reconstructed through:

The posture is prescriptive, universalist, expert-driven. Quantitative evidence was scarce; trade-offs were implicit.

Phase 2 · BPM-lifecycle era (late 1990s – 2010s)

BPR is absorbed into a structured methodology: the six-phase BPM lifecycle (Identify → Discover → Analyse → Redesign → Implement → Monitor), canonicalised in Dumas et al. (2018). Two moves define the phase:

Heuristics are catalogued and evaluated. Reijers & Liman Mansar (2005) introduce the devil's quadrangle — cost, time, quality, flexibility — and evaluate each heuristic's qualitative impact on all four. Dumas 2018's Appendix A absorbs this catalogue.
Discovery becomes methodical. Dumas et al. Ch. 5 formalises three qualitative discovery methods (evidence-based, interview-based, workshop-based) with explicit stakeholder roles. Sharp (2014) adds TRAC scoping and the 90-minute facilitated session as practitioner tooling.

Phase 3 · Process Mining wave (2011–2020)

Van der Aalst (2011/2016) and the Process Mining Manifesto (2012) shift discovery from expert-elicited to data-driven. Three mining types — discovery, conformance, enhancement — and six guiding principles frame the movement. The manifesto's C8 "operational support" (detect / predict / recommend) seeds everything that follows.

Phase 4 · Prescriptive turn (2021–2023)

Dumas's 2021 keynote "Process Mining 2.0" states the thesis: "PM 1.0 stopped at insights; PM 2.0 must drive action." Chapela-Campa & Dumas (2023) operationalise this as a four-layer analytics pyramid: Descriptive → Predictive → Prescriptive → Augmented Process Execution, with each layer reducing human involvement and increasing autonomy. Kubrak et al. (2022) survey the PrPM field — 37 methods across six dimensions — confirming that case-level runtime intervention is now a recognisable sub-discipline.

Phase 5 · AI-augmentation era (2023–2024)

ABPMS Manifesto (Dumas et al. 2023) introduces four system-level characteristics: adaptable, proactive, explainable, context-sensitive. Kampik et al. (2024) Large Process Models propose a neuro-symbolic architecture fusing fine-tuned LLMs with symbolic BPM tooling and assert that "pure LLMs are insufficient for BPM". The AI is augmentative: the system is still the unit of analysis.

Phase 6 · Agentic era (2025–2026)

The centre of gravity shifts from the system to the agent. APM Manifesto (Calvanese et al. 2026) treats software and human agents as primary functional entities and requires four ordered capabilities: framed autonomy → explainability → conversational actionability → self-modification. Dumas, Milani & Chapela-Campa (2026) extend the 2023 pyramid by replacing the apex with Agentic BPM (see concepts/agentic-bpm-pyramid) and propose a triangular human / rule-based / agentic autonomy spectrum. Elyasaf et al. (2025) add a five-level autonomy roadmap for self-modifying ABPS.

Empirically, Vu et al. (2025) provide the only practitioner-grounded study in our corpus: "None reported prior hands-on experience with agentic AI — answers reflect expectations informed by existing RPA/automation experience." This marks the gap between vision and deployment.

Where is it heading?

Three vectors cross the current frontier:

Neuro-symbolic vs. agent-centric. LPM 2024 centres on the LLM with symbolic scaffolding; APM 2026 centres on the agent with LLMs as one reasoning substrate among many. Both reject pure-LLM approaches. The integration depth is unresolved.
Agentic extension of BPR heuristics. Dumas et al. (2026) explicitly call for an "agentic extension of the classic Reijers & Liman-Mansar redesign heuristics" (p. 45). The 29-heuristic catalogue was built for top-down redesign; it must be reformulated for agent-driven adaptation and evolution.
Notation inadequacy. BPMN's activity-centric lens is insufficient once agents become first-class performers. Future formalisms will need objective blocks, guard-rail annotations, and verification patterns — a looming research agenda.

RQ2.1 — How Re-engineering lives on, and what has changed

Continuity: Hammer's vocabulary is still in circulation

Every contemporary BPM textbook in our corpus cites Hammer & Champy, directly or transitively:

Reijers & Liman Mansar (2005) name Hammer & Champy as primary literature source.
Sharp (2014) simplifies Hammer & Champy's five-point "Case for Action".
Dumas et al. (2018) Ch. 1 & Ch. 8 cover BPR history and devote a chapter to heuristic redesign.
Van der Aalst (2011/2016) Ch. 2 positions BPR as one tradition among data mining, Lean Six Sigma, and others.
Dumas et al. (2026) — 33 years after Hammer — still acknowledges BPR-heuristic lineage and proposes agentic extensions.

BPR is not buried. It is canonicalised and embedded in Redesign as a lifecycle phase.

What has changed (Hammer 90s → BPM 2020s)

Dimension	Hammer-era BPR (early 90s)	Modern BPM (2018-2026)
Discovery	Expert interviews, workshops	Data-driven (process mining) + evidence-, interview-, workshop-based. Event-log evidence available for the first time.
Redesign target	Big-bang radical change	Continuous + runtime intervention. [[sources/2023-chapela-campa-augmented-process-execution
Evaluation	Implicit trade-offs, consultancy reports	Explicit — every heuristic evaluated against [[concepts/devils-quadrangle
Universalism	One-size-fits-all	**[[concepts/context-aware-bpm
Autonomy in redesign	Human-driven (C-suite, consultants)	Hybrid. APM agents can propose or execute adaptations autonomously within normative frames. Practitioners want configurable, risk-proportional autonomy ([[sources/2025-vu-practitioner-perspectives-agent-governance
Change management	Often neglected; high failure rates	Explicitness required. [[sources/2014-sharp-whats-wrong-with-this-process
Notation	Text, flowcharts, IDEF	BPMN / DMN / CMMN — already under pressure from agentic BPM.

Reijers & Liman Mansar 2005 — the hinge between eras

Their 29-heuristic catalogue is the pivotal bridge:

Preserves Hammer-era heuristics (task elimination, composition, parallelism, specialisation, centralisation, customer teams, task automation).
Grounds them in literature citations per heuristic — unlike Hammer's consultancy posture.
Introduces the concepts/devils-quadrangle (Brand & van der Kolk 1995) as explicit four-dimensional trade-off evaluation — making visible what Hammer left implicit.
Becomes the textbook standard: Dumas et al. (2018) Appendix A incorporates the catalogue.

Sharp 2014 — BPR corrected, not rejected

Sharp's two 2014 contributions explicitly preserve Hammer's instincts while fixing BPR's practitioner-blindness:

Simplifies the Case for Action into a blame-free three-question frame.
Elevates performers — "Performers are usually ignored but are most impacted by change. 'If they don't see what's in it for me' there is a real chance that change efforts will be undercut."
Phases redesign — scope first, then discovery, then analysis — explicitly rejecting big-bang.
Grounds final assessment in a six-enablers framework (workflow · IS · motivation · HR · policies · facilities), echoing the systematic-evaluation principle of the devil's quadrangle.

Does the agentic era still cite BPR?

Yes, explicitly. Dumas et al. (2026) calls for agentic extensions of the Reijers & Liman-Mansar catalogue. The APM Manifesto does not name Hammer directly but positions itself as successor to the Process Mining Manifesto (2012) and the ABPMS Manifesto (2023) — both of which sit downstream of BPR. The lineage is intact.

RQ2.2 — What has not changed, and what we should learn

Enduring principles — still sound after 30 years

Principle	Hammer-era origin	Modern endurance
Customer-centric focus	End-to-end need over internal hand-offs	Dumas 2018 discovery foregrounds stakeholder identification; Sharp 2014 devotes a customer question pack; ABPMS names "context-sensitive" as a core characteristic.
End-to-end process view	Cross-functional over functional silos	Sharp 2014 explicitly surfaces when "what looked like many processes is actually one end-to-end process"; agents in APM are required to be process-aware, not task-aware.
Process-as-first-class-object	Process is the unit of analysis	Baked into every BPM textbook; APM makes process-awareness a required agent capability.
Executive sponsorship + governance	CEO-mandated change	Six success factors (Rosemann & vom Brocke) include strategic alignment and governance. Vu 2025 finds governance structures the most-cited requirement (8 of 22 practitioners) for agent adoption.
Culture change is non-negotiable	Acknowledged by Hammer; mishandled in practice	Still lists as a BPM-maturity factor. Vu 2025: "cultural acceptance of autonomy-to-agent transfer is a first-order adoption barrier — not just a technical issue."

Enduring failure modes — all five diagnosed in modern sources

Over-ambitious scope. BPR promised 30–50% efficiency gains; many projects delivered 5–10% or were abandoned. Correction embedded in Sharp (2014): "Get on TRAC by scoping before analysing." Vu 2025 practitioners cite clear business goals as an adoption prerequisite.
Performer neglect. BPR was C-suite-driven and worker-blind. Sharp 2014 explicitly corrects this. Vu 2025 finds job displacement (4 of 22) and the need for human-agent collaboration with defined roles and intervention points as first-order concerns.
Over-reliance on technology. Automation ≠ redesign. Reijers 2005 explicitly restricts scope to the technical challenge, acknowledging socio-cultural change as outside. Kampik 2024 asserts "classical BPM tooling stays; LLMs augment but do not replace". APM Manifesto repeats: "LLMs alone are insufficient".
Underestimating change management. BPR routinely treated implementation as an afterthought. Dumas 2018 Ch. 12 lists five success factors beyond the redesign itself (governance, methods, IT, people, culture). Elyasaf 2025's distinction between adaptation (instance-level, fast) and evolution (model-level, slow) formalises a distinction BPR-era thinking conflated.
One-size-fits-all universalism. BPR's generic prescription was its signature. The entire context-aware BPM movement (vom Brocke 2016, CAMAS 2021) is a standing correction, grounded in contingency theory: "no single best way — optimal management is contingent on external and internal context." Four dimensions × 14 factors structure the context classification. Kampik 2024 proposes per-context LLM fine-tuning for the same reason.

The devil's quadrangle as timeless frame

Time / cost / quality / flexibility (Brand & van der Kolk 1995, contemporaneous with Hammer) has survived where BPR's universalism did not. It reappears across three decades:

Reijers 2005: evaluation scaffold for each of 29 heuristics.
Dumas 2018 Ch. 8: central to redesign pedagogy.
Dumas 2026: KPIs feeding adaptation triggers in agentic BPM.

Enduring lesson: every redesign is a trade-off conversation, not a win-win conversation. Every heuristic improves some dimensions while degrading others. This remains true whether the redesigner is a 1993 consultant or a 2026 LLM agent.

Inherited method discipline

Modern BPM discovery still rests on BPR-era instincts, operationalised:

Structured interviews, not free-form conversations. Sharp 2014 prescribes a 45-minute structured + 15-minute free-form balance, with rainy-day questions to surface exceptions.
Stakeholder buy-in as ongoing, not terminal. Dumas 2018 Ch. 5 and Sharp 2014 both front-load stakeholder involvement in discovery and assessment — not at sign-off.
Exception-first elicitation. Dumas 2018 Ch. 5 derives rainy-day questions from an exception taxonomy (internal/external business exceptions, technology exceptions, timeouts).

These are not new inventions; they are BPR-era lessons compressed into repeatable practice.

Synthesis — three observations across the RQs

1. Every phase inherits rather than replaces. The seven-phase arc is not a sequence of paradigm shifts. Process Mining absorbed BPR. PrPM absorbed PM. ABPMS absorbed PrPM. APM absorbs ABPMS. The devil's quadrangle and the BPR heuristics survive in every subsequent layer; BPMN's activity-centric lens, by contrast, is now under strain at the agentic frontier.

2. The dominant correction is anti-universalism. The through-line from vom Brocke (2016) to Vu (2025) is the same: no universal prescription works. BPR's failure was not its heuristics but its universalism. Context-aware BPM encoded this lesson; configurable-autonomy-per-risk-profile (Vu's practitioner finding) extends it to the agentic era.

3. Culture is still the largest failure-mode surface. BPR failed on culture. BPM maturity frameworks name culture as a success factor. Vu 2025 names it as a "first-order adoption barrier — not just a technical issue" for agentic BPM. Three decades of vocabulary evolution have not made culture easier. If agentic BPM fails, it will most likely fail the way BPR failed.

Interlude — Does PrPM repeat or learn from BPR?

A close re-read of Kubrak, Milani, Nolte & Dumas (2022) lets us test the lineage claim directly. The 37-method SLR catalogues prescriptive process monitoring along six dimensions (objective, metric, intervention, modeling technique, input data, policy). The reading produces a split verdict.

Repeat signals — where PrPM rhymes with BPR's failure modes

Universalism inside each method. Kubrak's framework captures what a method prescribes but not for which context it is appropriate. No method in the corpus is context-aware in the vom Brocke (2016) sense; contingency-theoretic context classification is absent. The BPR-heuristics catalogue had the same blind spot (Reijers & Liman Mansar 2005).
Technology-first framing, change-management-absent. The PrPM SLR treats prescription as a purely technical pipeline (predictor → policy → intervention). There is no discussion of organisational adoption, performer buy-in, cultural fit, or governance — precisely the layer where BPR collapsed in the 1990s. Kubrak's one gesture toward implementation reality is the remark that second-order effects "[require] human judgment and iterative policy validation (e.g., via A/B testing)" (Kubrak 2022) — an aside, not a research programme.
Hand-specified interventions, expert-driven. Gap #2 in Kubrak's "Quo vadis?" list: no method systematically discovers interventions from logs; analysts hand-pick them. Hammer-era BPR did the same — consultants hand-picked redesigns. The PrPM field has, to date, automated the triggering and selection of interventions but not their discovery.
In-vivo validation absence. 36 of 37 methods are back-tested on historical or synthetic logs; only Dees et al. (2019) attempted a live deployment, and it "did not lead to desired outcomes" (Kubrak 2022). BPR consultancies claimed 30–50% gains based on similarly thin empirical ground.
No engagement with the design-time predecessor. The Kubrak paper contains zero mentions of Hammer, Reijers, BPR, the devil's quadrangle, or the BPM lifecycle. The prescriptive turn is framed as an outgrowth of process mining, not a runtime reinterpretation of BPR's redesign catalogue. This is historiographically surprising: Reijers & Liman Mansar's 29 heuristics include task reassignment, resource allocation, and control-flow reordering — the same intervention classes PrPM prescribes per case.

Learn signals — where PrPM genuinely diverges

Evidence-basis is in the DNA. Every method in Kubrak's corpus is grounded in an event log (real-life in 19 cases, synthetic in 9, both in 3). BPR had no analogue of this; its heuristics were consultant-distilled.
Per-case incrementalism, not big-bang. PrPM prescribes one intervention at a time for one running case. There is no "obliterate-don't-automate" rhetoric. The Sharp (2014) scoping correction — scope first, redesign later — is fully absorbed in practice (even if not cited).
Framed within the BPM lifecycle. PrPM sits inside the Monitor phase and feeds back into Redesign — both canonical lifecycle stages. The four-layer analytics pyramid (Chapela-Campa & Dumas 2023) formalises this as a recurring loop, unlike BPR's one-shot redesign ceremony.
Emerging causality + trade-off awareness. The Kubrak paper explicitly argues causality-based policy design (Shoush & Dumas 2021; Bozorgi et al. 2021; Fahrenkrog-Petersen et al. 2022) is the most promising direction — an epistemic posture Hammer-era BPR never adopted. Causal estimation of CATE per intervention is a structurally better-specified analogue of the devil's quadrangle trade-off evaluation (even if the DQ is not named).

Genuinely new territory — not comparable to BPR at all

Runtime per-case intervention. BPR had no conceptual slot for acting during case execution; it was design-time only. PrPM is fundamentally a runtime discipline.
Reinforcement-learning policies (Metzger, Kley & Palm 2020) and LSTM-based next-action recommenders (Weinzierl et al. 2020a/b; Park & Song 2019) have no 1990s counterpart.
Alarm-based prescription (Teinemaa et al. 2018; Fahrenkrog-Petersen et al. 2022) — probability-above-threshold triggering — is a new policy primitive that BPR heuristics could not express.

The verdict

PrPM has learned on evidence-basis, incrementalism, and causal-reasoning sophistication, and it has extended BPM into runtime territory BPR never occupied. But it risks repeating BPR's twin mistakes — universalism (context-blind prescription) and technology-first framing (implementation-layer neglect). The Kubrak SLR's silence on BPR is symptomatic: the field has engineered a runtime prescriptive stack without ingesting the 1990s lessons about why top-down, expert-driven process prescription fails in practice. The fix is not technical. It is to reconnect PrPM explicitly to the Reijers & Liman Mansar catalogue and the devil's quadrangle, to make each prescription's trade-off profile explicit, and to treat in-vivo validation (Gap #1 in Kubrak's list) as the field's top priority — exactly the correction vom Brocke (2016) and Sharp (2014) made to BPR a generation ago.

Gaps in this wiki for strengthening the argument

No direct Hammer 1990/1993 ingest. The BPR reconstruction is mediated by Reijers 2005, Sharp 2014, and Dumas 2018. A deep read of Reengineering the Corporation would let us verify claims about Hammer's actual prescriptions rather than trusting post-hoc synthesis.
No ingested 1990s critics of BPR. Davenport & Stoddard ("Reengineering: Business Change of Mythic Proportions?", 1994), Willcocks & Smith, Harmon — contemporary reactions that diagnosed BPR failures in real time.
PrPM as repeat or learn-from-BPR. Now integrated — see §Interlude above. A deep re-read of Kubrak 2022 (2026-04-22) yielded a split verdict: PrPM learns on evidence-basis, per-case incrementalism and causal-reasoning sophistication, but risks repeating BPR's universalism and technology-first framing, and is historiographically disconnected from the Reijers & Liman Mansar heuristics catalogue.
Deployed agentic-BPM failure data. The agentic corpus is visionary. Case studies of real deployments (2026 onward) will ground prospective failure modes empirically rather than speculatively.

syntheses/abpms-to-apm-evolution — narrower focus on the ABPMS → APM shift.
syntheses/apm-manifesto-core-messages — close-reading of Calvanese et al. 2026.
syntheses/llm-bpm-reading-list — LLM-in-BPM reading list.
syntheses/ppm-landscape — PPM/PrPM corpus synthesis.
concepts/devils-quadrangle, concepts/bpr-heuristics, concepts/context-aware-bpm, concepts/bpm-lifecycle, concepts/bpm-maturity, concepts/agentic-bpm.