{"id":"597b6042-c05a-48bf-899a-7106ffa00184","arxiv_id":"2508.00116","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper asserts that object-centric process mining, renamed Process Intelligence, is the missing grounding layer for AI applied to operational processes.","lead":"This paper argues that successful AI in business operations depends on process intelligence, a data grounding built from object-centric process mining. It is a position paper explaining why process-aware data representations should anchor generative, predictive, and prescriptive AI.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The abstract's 'missing link' claim is a necessity claim, but the paper offers no comparison against flat event representations, leaving the central thesis underdetermined; a benchmark would settle it.","rationale":"The reader's conditional verdict is appropriate. The central thesis is an empirical necessity claim dressed as a conceptual one. A position paper can make such claims as a research agenda, but the categorical 'No AI Without PI' title and the abstract's 'we show' wording go beyond agenda-setting. The concern is not that OCPM is useless; object-centric logs plausibly capture concurrency and many-to-many relationships that flat logs lose. Rather, 'missing link' requires counterfactual evidence. The concrete test directly targets this necessity: if a flat representation plus object identifiers performs equally well, then the missing link is not the representation but model capacity, feature engineering, or data quality. The definitional circularity makes this test essential; without it, the claim can be read as true by stipulation. Since the reader already conditioned the verdict on softening or providing evidence, my read does not change the verdict. I agree with the reader's identification of the weakest assumption as the decisiveness of process-structural grounding relative to alternative failure causes.","tokens_in":8967,"tokens_out":3588,"duration_ms":38422,"concrete_test":"Use a real OCEL dataset (e.g., order-to-cash with multiple interacting object types) from the public object-centric process mining benchmark repository. Hold out a temporal test split. Train two next-activity/remaining-time predictors with identical architecture and hyperparameters: one on the object-centric event log (OCPM representation) and one on a flat event table augmented with object IDs, timestamps, and case-level features. If the flat baseline matches or beats OCPM on held-out accuracy, the 'missing link' necessity claim is falsified in that setting; repeat on at least five independent datasets before generalizing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim, 'AI needs to be grounded using OCPM' and 'OCPM is the missing link connecting data and processes,' asserts that object-centric process structure is the decisive enabler for generative, predictive, and prescriptive AI in operational settings. For this necessity claim to be true, one must rule out alternative enablers: stronger base models, flat event tables enriched with object identifiers and timestamps, knowledge graphs, or workflow-level LLM orchestration. The paper does not provide a controlled comparison, a quantitative evaluation, or a formal argument that these alternatives are insufficient. There is also a definitional risk: PI is defined as the amalgamation of process-centric object/event techniques, so the conclusion that 'AI requires PI' can hold trivially for the restricted class of process AI without establishing that OCPM specifically is required. This is not an internal inconsistency, but it makes the central claim an unsupported causal hypothesis rather than a demonstrated result. The garbled full text prevents verification of any section-level evidence, so the abstract-level argument is the strongest available form of the claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a position paper arguing that effective AI for operational processes requires Object-Centric Process Mining (OCPM) as the grounding representation. The authors introduce the term Process Intelligence (PI) to denote the amalgamation of process-centric data-driven techniques, and they discuss how OCPM enables generative, predictive, and prescriptive AI. The paper surveys challenges such as process dynamics, organization-specific structure, and multiple object/event types, and it presents a series of claims and recommendations rather than experimental results.","tokens_in":9151,"tokens_out":3178,"duration_ms":30238,"significance":"If the central claim were established, this paper would provide a valuable framework for AI in process mining, shifting attention from flat event logs to object-centric models. The paper usefully identifies the gap between generic AI and structured process data, and it draws together three AI paradigms under one conceptual umbrella. Its strengths include a clear articulation of the representation problem and a concrete proposal (object-centric event data) as a candidate solution. However, the manuscript does not provide a controlled comparison, quantitative evidence, or a formal argument that OCPM is uniquely necessary, so the significance is currently that of a research manifesto rather than a demonstrated result.","major_comments":[{"comment":"The abstract asserts that 'AI needs to be grounded using OCPM' and that OCPM is 'the missing link connecting data and processes.' This is a necessity claim, but the paper offers no comparison against flat event logs augmented with object identifiers, knowledge-graph representations, or LLM orchestration over structured data. Without such a comparison or a formal argument ruling out alternatives, the claim is an unsupported causal hypothesis. Please either add empirical or formal support or reframe the claim as a conjecture or research agenda (e.g., 'we argue that OCPM is a strong candidate').","section":"Abstract"},{"comment":"The paper defines Process Intelligence (PI) as 'the amalgamation of process-centric data-driven techniques' and then concludes that 'AI requires PI.' If PI is defined as the collection of all process-centric AI techniques, then the conclusion can become true by definition for any process-AI system, without establishing that OCPM specifically is required. Please clarify the logical status of the claim: is it a definitional statement, a design principle, or an empirical claim? Locate this risk in the abstract and the introduction and address it explicitly.","section":"Section 1 (introductory paragraphs)"},{"comment":"The full text received for review is heavily corrupted by an encoding error, so I could not verify any section-level evidence. Based on the readable abstract, the manuscript does not appear to include experiments, benchmarks, or case studies. For a journal, the central 'enabler' claim needs at least one detailed worked example or a small empirical demonstration to show how OCPM changes AI outcomes. Please add such evidence or explicitly state that the paper is a perspective piece and move the necessity claim to a hypothesis.","section":"Full text (unreadable) and Abstract"}],"minor_comments":[{"comment":"The title 'No AI Without PI!' is catchy but overstates the argument; consider a title like 'Process Intelligence as a Grounding for AI in Operational Processes.'","section":"Title"},{"comment":"The abstract uses 'we show' where 'we argue' would be more accurate given the absence of proof; please align the wording with the paper's evidentiary level.","section":"Abstract"},{"comment":"The term PI is later used as an abbreviation for Process Intelligence, but the acronym is also commonly used for other concepts in AI; please define it at first use and avoid ambiguity.","section":"Throughout"},{"comment":"The text I reviewed contains many garbled characters; please ensure the published PDF uses correct encoding so figures and tables are legible.","section":"PDF/encoding"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim is strongly aligned with the author's own OCPM research program, and the paper does not engage with alternative representations in a balanced way. While this is not a reason for rejection, it strengthens the need for an external benchmark or neutral comparison in revision. The corrupted full text prevented a thorough check for missing references or internal consistency issues."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a position paper, and judged on that scale it's a good one. Van der Aalst's new contribution is not OCPM itself—that's an existing program he helped build—but the umbrella term Process Intelligence and the systematic argument that object-centric event data and process models are what let generative, predictive, and prescriptive AI work on operational processes. The paper does well at explaining why generic, text-centric AI is a poor fit for structured, organization-specific, dynamic process data. That's a real and underappreciated point.\n\nThe soft spot is the force of the claim. 'AI needs to be grounded using OCPM' and 'OCPM is the missing link' are necessity claims. The paper gives no comparison against flat event tables with object identifiers, knowledge graphs, or workflow-level LLM orchestration. So the central thesis is underdetermined. There is also a definitional risk: PI is defined as the amalgamation of process-centric techniques, which makes 'AI requires PI' close to 'process AI requires process techniques.' That's not an internal contradiction, but it means the headline conclusion is a research hypothesis rather than a demonstrated result. The stress-test note lands on this correctly.\n\nI could not check section-level detail because the full text in my copy is badly garbled; my assessment rests on the abstract and figures. That limits how hard I can push, but the abstract is the strongest form of the claim anyway.\n\nThe lack of a benchmark is not fatal for a position paper, but the title overclaims. A few small changes would fix it: soften 'needs to' to 'benefits from,' or add one small empirical illustration. The citation pattern looks honest; the paper draws openly on prior OCPM work and the self-citations there are legitimate given the author's role. No code, data, or formal proofs, so there's no formal evidence to inspect—again, normal for the genre.\n\nWho benefits: BPM and process-mining researchers, and applied AI teams deciding whether to invest in process-aware data infrastructure. They'll get a clear agenda and a useful vocabulary. I'd send it to peer review, but with reviewers instructed to push on the necessity claim—conditional accept at most, not reject.","headline":"A credible programmatic agenda from the OCPM camp; the PI umbrella is useful, but the title's necessity claim outruns the evidence and should be softened or supported.","tokens_in":9655,"tokens_out":3356,"would_cite":true,"duration_ms":35120,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that AI for operational processes requires object-centric process mining as its grounding, and that process models and event data are the missing link.","keywords":["Object-Centric Process Mining","Process Intelligence","Generative AI","Predictive AI","Prescriptive AI","operational processes","event data","process grounding"],"falsifier":"A controlled comparison on the same operational event logs—one version represented as an object-centric event log with mined process models, the other flattened into single-case tables—using the same predictive and prescriptive AI methods would settle the claim: if flattened representations match or beat the object-centric ones, the 'missing link' argument loses its empirical support.","tokens_in":8766,"feed_emoji":"🧩","tokens_out":5575,"duration_ms":53379,"temperature":0.7,"pith_summary":"Organizations struggle to make AI work on end-to-end operational processes, even as generative, predictive, and prescriptive AI advance elsewhere. This paper argues that the reason is a missing link: AI models are not grounded in the structure of the process they are supposed to improve. Process data are structured and organization-specific, and processes are dynamic, so the paper proposes Object-Centric Process Mining (OCPM) as the necessary grounding. OCPM records events as they relate to multiple object types and extracts process models from those events, and the paper calls the resulting combination of process-centric techniques Process Intelligence (PI). If the argument holds, organizations should build AI on object-centric event data and process models rather than on generic text or single-event tables.","feed_headline":"AI needs object-centric process mining to fix operations","feed_subtitle":"Grounded event data, not raw text, is what lets generative and predictive AI improve end-to-end processes.","key_machinery":"The central mechanism is the object-centric event log: a collection of events in which each event can refer to multiple objects of different types, such as an order, a delivery, and an invoice. Object-Centric Process Mining (OCPM) turns these logs into process models and diagnostics that preserve object interactions. This mechanism carries the argument by supplying the structural grounding that the paper claims generic AI lacks; it lets generative, predictive, and prescriptive AI connect their outputs to the real process rather than to a flattened, case-id-based approximation.","core_discovery":"The paper claims that AI needs to be grounded using Object-Centric Process Mining (OCPM). Process-related data are structured and organization-specific, and processes are highly dynamic, so generic AI that treats data as text or as flat tables cannot reliably diagnose or improve them. OCPM connects the event data to the actual process by recording events in relation to multiple object types and mining process models from those events. The paper introduces the term Process Intelligence (PI) for the amalgamation of process-centric, data-driven techniques that handle many object and event types, and argues that PI is what enables generative, predictive, and prescriptive AI in organizational settings.","pith_inferences":["If the paper is right, then adding an object-centric event log and a mined process model as context should improve the behavior of a fixed generative or predictive model on process tasks; that is a direct experimental test the paper does not run.","The argument suggests a data-engineering priority: before scaling models, organizations should build event data that records multiple object types, because the claimed bottleneck is structural rather than computational.","A public benchmark that compares OCPM-grounded models with flattened single-case baselines on the same event logs would turn the paper's central metaphor into an empirical question."],"forward_implications":["Predictive and prescriptive models can exploit the relations among objects in an event log instead of forcing each event into a single case.","Generative AI suggestions can be constrained by a mined process model, grounding recommendations in the organization's own process structure.","The same object-centric event data can support diagnosis, prediction, and improvement, making Process Intelligence one shared substrate for all three forms of AI.","AI projects in process-heavy organizations would start with object-centric event-data capture rather than with model selection."],"supporting_citations":[],"fun_headline_variants":["Object-Centric Process Mining: AI's missing link in operations","Without Process Intelligence, AI fails on operational processes","Process Intelligence: the key to AI in dynamic operational settings","AI can't fix processes without PI"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the main reason AI fails in operational settings is the absence of process-structural grounding, so that object-centric process data are the decisive enabler rather than model capability, data quality, or organizational incentives.","fun_headline_variants_meta":{"raw":{"variants":["Object-Centric Process Mining: AI's missing link in operations","Without Process Intelligence, AI fails on operational processes","Process Intelligence: the key to AI in dynamic operational settings","AI can't fix processes without PI"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000442,"raw_usage":{"total_tokens":2188,"prompt_tokens":842,"completion_tokens":1346,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":458,"completion_tokens_details":{"reasoning_tokens":1284}},"tokens_in":458,"tokens_out":1346,"duration_ms":10093,"temperature":1.0,"reasoning_tokens":1284,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:20:52.495291+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A controlled comparison on the same operational event logs—one version represented as an object-centric event log with mined process models, the other flattened into single-case tables—using the same predictive and prescriptive AI methods would settle the claim: if flattened representations match or beat the object-centric ones, the 'missing link' argument loses its empirical support.","supporting_citations":[],"review_version":1}