---
title: "Business Intelligence and Business Process Management in the Era of Generative AI: A Review of Big Data Analytics, Process Mining, and Decision Support Systems"
type: source
tags: [bi, bpm, big-data, process-mining, genai, llm, dss, integrative-framework, narrative-review, layered-architecture]
authors: [Theodorakopoulos, Leonidas; Theodoropoulou, Alexandra]
year: 2026
venue: "Applied Sciences (MDPI) 16: 4603. DOI: 10.3390/app16104603. Published 7 May 2026."
kind: paper
raw_path: "raw/Process Frameworks & BPM/2026-theodorakopoulos-bi-bpm-genai-review.pdf"
doi: "10.3390/app16104603"
sources: []
key_claims:
  - "Narrative + conceptual review (explicitly NOT PRISMA-style), search window 2015 onwards across Google Scholar, Scopus, Web of Science. Methodological justification anchored in Cooper's taxonomy of knowledge syntheses."
  - "Literature does NOT support the claim that BI, BPM, big data analytics, process mining, GenAI, and DSS have converged into a stable, unified field — meaningful but uneven integration."
  - "Proposes a 5-layer integrative framework with cross-layer feedback loops: Data & Computation (big data analytics) → Organisational Insight (BI/business analytics) → Process Intelligence (BPM/process mining) → Augmentation (GenAI) → Decision (DSS). GenAI is positioned explicitly as an augmentation layer, NOT the new core — 'derivative in the crucial sense that augmentation is only as strong as the evidentiary, process, and governance foundations it augments.'"
  - "GenAI in BPM = complementarity, not substitution. Conventional ML remains central for tightly-specified predictive/optimisation tasks; GenAI's distinctive role is interaction + knowledge mediation. Strongest evidence so far concerns process-model drafting and document-level synthesis — not autonomous process authorship."
  - "5 limitation categories for GenAI in process-critical environments (Table 3): hallucination & factual instability, semantic imprecision in domain-specific tasks, weak real-world evaluation, privacy/security exposure, limited accountability/auditability."
  - "Implementation tensions accumulate across layers: weak data quality propagates upward; explainability ≠ effective oversight (simple explanations weaker than cognitive forcing); trust must be calibrated rather than maximised; governance is a design constraint, not an afterthought."
  - "5 future-work directions strongly overlap APM gaps: cross-layer evaluation frameworks; human-centered + explainability-aware GenAI for BPM; privacy-preserving/federated process analytics; process-aware multimodal LLM systems; real-time adaptive DSS under governance constraints."
created: 2026-05-11
updated: 2026-05-11
---

# Theodorakopoulos & Theodoropoulou 2026 — Business Intelligence and Business Process Management in the Era of Generative AI

55-page narrative-conceptual review published in *MDPI Applied Sciences* (May 2026). Authors are based at the Department of Management Science and Technology, University of Patras (Greece). The review's headline contribution is a **layered integrative framework** for connecting BI, BPM, big data analytics, process mining, GenAI, and DSS — explicitly tempered against techno-optimism.

This is the wiki's first ingested source on integrative BI ↔ BPM ↔ GenAI ↔ DSS framing. Complementary to (but lighter than) the APM Manifesto's pyramid framing.

## What the paper does

A conceptual (not bibliometric, not PRISMA) review that:

- **Searches** Google Scholar, Scopus, Web of Science for peer-reviewed work from 2015 onwards on BI, BPM, big data analytics, process mining, GenAI, DSS, and their intersections.
- **Thematically clusters** the literature into conceptual foundations, data & computational infrastructure, process intelligence, GenAI in business processing, integrative decision-support architectures, application domains, and implementation tensions.
- **Synthesises** the clusters into a layered architecture rather than a bibliometric aggregation.

Three research questions: (RQ1) how do BI/BPM/BDA/PM/GenAI/DSS relate within contemporary business-processing environments; (RQ2) what conceptual roles do they play as parts of an integrated decision-support architecture; (RQ3) what implementation tensions limit practical integration.

## The 5-layer integrative framework (Section 7, Figure 4)

| Layer | Domain | Role |
|---|---|---|
| **Data & Computation** | Big Data Analytics | Captures, stores, integrates, processes heterogeneous data streams (event logs, transactions, customer interactions, IoT, multimodal sources). *Enabling base*, not interpretation. |
| **Organisational Insight** | BI / business analytics | Transforms dispersed operational/transactional/contextual inputs into managerial interpretation. *Distributed interpretive capability*, not just dashboards. |
| **Process Intelligence** | BPM / process mining | Reconstructs how work unfolds across cases, activities, handoffs, delays, deviations. Goes *beyond control-flow* into multi-perspective + predictive + explanatory. |
| **Augmentation** | Generative AI | Mediates between layers — translates data, models, and analytical outputs into forms humans can query, summarise, recombine. Explicitly **NOT** the new core. |
| **Decision** | Decision Support Systems | Converts insight, process signals, generated explanations into alternatives, priorities, interventions, accountable choices. Action-conversion layer. |

Cross-layer **feedback loops**: decisions trigger interventions → interventions generate new traces → new traces reshape data foundation. Framework is recursive, not linear.

## Position on GenAI

A central thesis: GenAI is positioned as **complementarity, not substitution**. Conventional ML remains central for tightly-specified predictive and optimisation tasks ("classification, prediction, anomaly detection, resource allocation, narrow-task decision support"); GenAI's distinctive role is "summarisation, process knowledge extraction, conversational querying, draft model generation, explanation" (Table 1). The clearest evidence so far is in process-model drafting (Nivon & Salaün, Hörner et al., Klievtsova et al.) and document-level synthesis — *not* autonomous process authorship.

Section 6.5 catalogues five limit categories for GenAI in process-critical environments (Table 3):

| Limitation | Why it matters | Typical consequence |
|---|---|---|
| Hallucination & factual instability | Fluent but unfaithful generation | Fabricated justifications, false confidence |
| Semantic imprecision in domain tasks | Near-correct outputs may fail sequence/rule logic | Process errors, compliance failures |
| Weak real-world evaluation | Benchmark success ≠ workflow success | Overestimation of deployment readiness |
| Privacy/security exposure | Data leaked/inferred/mishandled in prompts | Disclosure risk, regulatory exposure |
| Limited accountability/auditability | Outputs hard to trace to verifiable sources | Reduced contestability, weaker oversight |

## Implementation tensions (Section 9)

- **Data quality and interoperability accumulate across layers** — weak source-data alignment propagates upward and makes integration appear performative rather than real.
- **Explainability ≠ effective oversight** — Cecil et al. and Buçinca et al. cited: explainable AI advice does not significantly reduce people's tendency to follow incorrect advice; cognitive-forcing mechanisms outperform simple explanations.
- **Trust must be calibrated, not maximised** — Afroogh et al. distinguish trustworthiness (system property) from trust (user disposition); over-trust is as problematic as resistance.
- **Governance is a design constraint** — Papagiannidis et al. argue responsible-AI governance is operationalisation through structural/relational/procedural practices, not abstract principles.
- **Organisational readiness ≠ willingness to experiment** — includes governance literacy, data stewardship, training, process-redesign capacity.

## Application domains (Section 8, Table 4)

Seven business domains illustrate the framework: operational process optimization · customer-facing/service processes · supply chain & logistics · finance, compliance & risk · knowledge-intensive admin workflows · procurement, sourcing & contracts · human–AI managerial decision-making. Each domain highlights which layers are most involved, what practical value the framework adds, and the main implementation concern.

## Future-work agenda (Section 10.2)

1. **Cross-layer evaluation frameworks** — move beyond isolated component benchmarks toward integrated business-processing evaluation.
2. **Human-centered & explainability-aware GenAI for BPM** — focus on oversight conditions, cognitive burden, contestability.
3. **Privacy-preserving and federated process analytics** — for cross-organisational, regulated environments.
4. **Process-aware multimodal LLM systems** — reason across documents, event logs, interfaces, images while staying grounded.
5. **Real-time adaptive DSS under governance constraints** — adaptive recommendations remaining auditable and accountable.

Substantial overlap with the open gaps catalogued in [[syntheses/llm-bpm-reading-list]] §E.2.

## Limitations

Acknowledged by the authors (Section 10.1):

- **Narrative review, not PRISMA** — coverage is selective and conceptual rather than exhaustive/statistical.
- **Framework not empirically validated** — should be read as conceptual architecture, not tested implementation model.
- **Evidence base is uneven** — BI/BPM/PM literature is comparatively mature; GenAI-in-process-critical evidence is newer, less stable, often prototype-level.

Wiki-maintainer additional caveats:

- *MDPI Applied Sciences* is a high-volume open-access journal; not a core BPM venue. The authors are from a management-science department, not the established BPM research community.
- The "BI / BPM / PM / BDA / GenAI / DSS as layered architecture" framing is useful as vocabulary but overlaps substantially with prior layered framings (Process Mining 2.0 from Dumas 2021; the APM Manifesto's framed-autonomy → explainability → conversational-actionability → self-modification capability stack). **Treat as integrative/teaching reference**, not as a primary theoretical contribution.

## Connections

**Concepts:**
- [[concepts/agentic-bpm]] — 5-layer framing as a complementary lens on the same problem APM addresses; useful contrast.
- [[concepts/explainability-apm]] — calibrated-trust + cognitive-forcing argument; relevant to APM's explainability capability.
- [[concepts/predictive-process-monitoring]] — situates PPM within Process Intelligence layer.
- [[concepts/prescriptive-process-monitoring]] — situates PrPM in the Decision layer.
- [[concepts/llm-assisted-process-modelling]] — strongest documented GenAI-BPM evidence cluster.
- [[concepts/llm-based-ppm]] — relevant to the Augmentation layer's process-aware future direction.

**Sources cited that exist in the wiki:**
- [5] Weinzierl ML in BPM systematic review (*not yet ingested*).
- [6] [[sources/2024-kampik-large-process-models]] — LPM vision.
- [45] Ceravolo PPM Challenges (*not yet ingested*).
- [47] Teinemaa outcome-PPM (*not yet ingested* — flagged as priority in [[syntheses/ppm-landscape]] §6).
- [54] Carmona conformance checking textbook (*not yet ingested*).
- [76] Bernardi RAG + process-aware DSS (*not yet ingested*).
- [77] Rebmann LLM-on-semantics-aware-PM (*not yet ingested* — also priority in [[syntheses/llm-bpm-reading-list]] §B.L8).
- [78] [[sources/2024-kampik-large-process-models]] — Kourani LLM in BPM (*partial overlap with Licardo et al. 2026 already ingested*).
- [79] Grohs LLMs in BPM (*not yet ingested* — flagged in reading list §B.L6).
- [81] Berti PM-LLM-Benchmark (*not yet ingested* — flagged in reading list §B.L8 vicinity).
- [101] [[sources/2023-chapela-campa-augmented-process-execution]] — augmented process execution (currently a stub).
- [122] van der Aalst beyond control-flow PM (*not yet ingested*).
- [126] van der Aalst beyond workflow-like representations (*not yet ingested*).
- [129] Rizzi explainable PPM user evaluation (*not yet ingested*).
- [148] Graafmans DMAIC + PM (*not yet ingested*).
- [154] Middelhuis learning-based resource allocation (*not yet ingested*).
- [217] [[sources/2020-rama-maneiro-deep-learning-ppm-review]] — DL-PPM benchmark.

**Syntheses:**
- [[syntheses/llm-bpm-reading-list]] — new entry in §A.7 (Integrative reviews); future-work directions feed §E.2 gaps.
- [[syntheses/abpms-to-apm-evolution]] — alternative integrative framing alongside ABPMS → APM evolution.