explainable ai

From static analytics to continuous intelligence: toward a dynamic model of real-time organizational decision systems

Purpose Enterprise decision systems increasingly couple time-sensitive sensing, workflow-embedded recommendations or actions, bounded machine discretion and operational feedback. This paper aims to develop a dynamic model of Continuous Intelligence (CI) that explains the organizational consequences of alignment or imbalance among these properties. Design/methodology/approach The paper uses a problem-driven conceptual theory-building approach. It integrates and contrasts research on decision support, real-time and operational analytics, automation and delegation, hybrid intelligence, machine learning operations, AI capability and governance, information systems success and dynamic capabilities. The inferential pathway is documented through contrastive construct analysis, rejected alternative representations, mechanism mapping and conceptual boundary probing. Findings CI is a four-dimensional decision-system configuration enabled by technical architecture, a CI governance process and human judgment integration. Its distinctive explanatory value lies in interaction effects: temporal continuity can amplify error propagation, discretion can amplify both value and harm, embedded feedback can create adaptive learning or self-reinforcing lock-in, and control design can compress or prolong intervention. Contextual alignment, rather than maximal automation, determines expected outcomes. Research limitations/implications The framework is conceptual and does not establish causal effects. It offers five multilevel propositions, falsification conditions and a sequenced research agenda for formative profile construction, configurational testing and longitudinal evaluation. Practical implications The framework provides theory-informed governance heuristics for matching sensing cadence, decision authority, feedback and oversight to decision risk, reversibility and equivocality. It also identifies proportional implementation options and constraints for smaller organizations. Social implications The redistribution of decision authority from human agents toward AI systems is occurring without adequate governance frameworks in most enterprise deployments. At the industry level, correlated model behavior across organizations deploying similar CI architectures creates systemic risks including flash crashes in financial markets, cascade failures in logistics networks, and biased clinical triage at scale. The European Union Artificial Intelligence Act (EU AI Act) and equivalent regulatory frameworks are attempting to retrofit accountability structures onto CI deployments that outpaced governance development. This paper contributes the theoretical grounding needed to design accountability structures prospectively rather than reactively. Originality/value The contribution is configurational and explanatory rather than technological. The paper shows how familiar decision-system properties generate emergent organizational effects when recursively coupled within operational workflows and why comparable investments can produce divergent outcomes.

Don’t Take Their Word for It: Reply to Fan

Abstract In my (2024), I argued that there are principled reasons for thinking that AI cannot be an authority about one’s own mind. The idea is that it is only rational to treat another as an epistemic authority under certain conditions, and AI does not meet one of those conditions. Fan (2025) rejects the account of epistemic authority I favor and sketches an alternative. In this response, I explain why I am not convinced by the objection to the account and why the account fares better than the proposed alternative. I conclude by considering the bearing of the disagreement on issues in the philosophy of AI.

PREVENTING WAR/BURNOUT: THE IRONIC PERSPECTIVES OF TRUTH, REALITY, AND WORKABILITY

PREVENTING WAR/BURNOUT: THE IRONIC PERSPECTIVES OF TRUTH, REALITY, AND WORKABILITY Gerhard Ris, author, is a pensioner, former DA magistrate, and lawyer · Preface This article has a one-page introduction as an excerpt on the topic: Preventing war/burnout: the ironic perspectives of truth, reality, and workability. It provides a tour of the horizons of the entire model, explained to an interested high school-level student. And, to any President of any court of law asked to take a timely interim measure in due order. My advice is to scan the index after having read the introduction. After that, scan the 17-page AI chat at the end of this paper. I’d give the chat an 8/10 score: understanding what I’m on about after AI was very rejecting at first, yet asked pertinent questions. In the end, AI deems what I state as consistent with all scientific online data, something AI is indeed good at, even though it understandably/admittedly makes mistakes in understanding what I am saying. Then scan the entire article, highlighting new insights since my last 19 DOI-publications on Zenodo: the CERN repository. The article proves that mainstream science can’t validly falsify the simple one-A4-block model since 2014 because it’s correct, with enormous consequences across the board as the introduction shows. The block model is a proven law of everything since 2024. A law that always applies, showing childishly easy 3D Euclidean geometry. Beware of going down the rabbit hole by getting stuck in details and losing oversight. Lots of terms are jargon that I’ve corrected on this elementary topic. I’ve redefined ‘religion'. Many people have lots of religious anchors reading certain terms that produce allergic reactions. This stems from not being able to do what AI does correctly: being both extremely open-minded and extremely precise in checking all online data for consistency extremely fast. The pertinent orderly question is: why can’t mainstream science pose valid falsification within twenty-four hours? The answer is simple: fear of losing peer-review power because my method, which is a slight improvement on Richard Feynman’s, is undeniably correct. My model and method meet the highest possible scientific standard, which ironically is the same as that for good engineers and good lawyers, who both are used to working with proven best practices given the lack of data in accordance with the laws that govern the topics at hand. Enjoy this scary testable fairy tale! PREVENTING WAR/BURNOUT: THE IRONIC PERSPECTIVES OF TRUTH, REALITY, AND WORKABILITY To save any decent democratic legal system, distinguish between: good narcissistic team members, mentally healthy criminal narcissists, and pathological narcissists. Narcissism is a linear function of every human; when irreversibly transformed, it produces a narcissist. This can only be kept in desirable balance by following the new law of human nature as an exact science written in a book that everyone can easily follow, ironically living in nature without the need to read it, unless you start using written language, arithmetic, algebra, and geometry. For then we will start to build “this time unsinkable” Titanics, subsequently, as repeating history, hitting the same iceberg. When you, in essence, do the same, the same happens. As all mentally healthy high school students learn to go by this book after twelve to eighteen years of Bildung, we won’t sink into the abyss. Bildung is neurologically internalised healthy hypnotic illusionism, defined as deep religion. As long as one follows ‘The Law’, one is guaranteed to have an optimal life. The 64 DNA personality types of the modelled factory of everything, such as producing justified content lives for all, can and thus must be learned by everyone. Deep religion, religion, slight religion, and the non/not yet religious unique self in a unique situation is the proven best practice method to assess any situation. Otherwise, mounting egoism will invariably slowly lead to a narcissistic authoritarian society ending in the thermodynamics of conflict and war as the law of the narcissist. Around 40% of humanity, including most professors of physics and mathematics, lack the talent to spot any new form of even the simplest ironic twist in arithmetic, algebra, and geometry. This can be easily tested by the 60% with the ironic talent in language by recording it on video. Only performing the proposed Oracle Senate Test and transforming the dysfunctional Trias Politica into a new Cinque Politica can save any decent parliamentary-democratic legal system. More billionaires, no poor, no war is easy because there is no money gene in our DNA. Freedom of Thought, Freedom of Relations, Security, and Respect are most critical in the respective distribution: 10%-10%-40%-40% of the collective human brains organised and trained consistently with ‘The Law’. Suicide is an ironic survival trait of the species. Judges by democratic law must indeed follow mainstream science. The President of any court is solely responsible for the order of proceedings and may even be honour-bound to ask preliminary questions, especially when science admits to being in transition because of serious structural errors of the highest order. Sick science can only be cured by staying in balance with the 10% artistic & 10% spiritual minority who have that DNA talent that majority rule religion ironically prevents. For the first time in our history on an infinite timeline, all the elementary science is in. An easily falsifiable claim within 24 hours by any behavioural scientist. They should be able to show any inconsistency very precisely. The reductio ad absurdum proof presented on one A4 shows that all brain waves are identical to all the waves in everything as the order function of everything. The more absurd the world, the more difficult irony becomes. Elementary scientists should, like good engineers, learn to treat evidence and proof like good lawyers and magistrates should, with incomplete data yet complete laws of everything. Mother Nature is a mass murderer with mass and not matter as the elementary murder weapon. The publication containing the first thousand elementary terms, descriptions, and definitions for use in elementary science, schools, and courts of law proves this extreme limitation of the Babylonian confusion is possible, even though never perfect; still, it is critically important. Elementary paradox should solely mean an inward contradiction, and any deeper insights a seeming contradiction best termed irony. A paradox is not an enigma or a seeming contradiction. Truth and truths are defined from one perspective; reality and realities are defined from more perspectives; workability (werkelijkheid in Dutch) is defined from all perspectives, which is at the beginning of absolute proof of everything existing proven beyond reasonable doubt infinite on circumstantial evidence of only data pro non con with only one remaining axiomatic assumption that everything is consistent, meaning no contradictions or incompleteness in ‘The Law’ also as the law of human nature. QED!

Time Is One-Dimensional Because Prediction Demands It, Not Because a Law Says So ── The Same Procedure, With Only the Signature Changed, Gives Amplifications of 0.012 and 3.5 × 10^32 ── [Paper 253]

Why is time one-dimensional? The answer offered here is that no law decrees it; the requirement that one be able to predict allows nothing else. The equations of a universe with two times can be written, and their solutions exist. What breaks is prediction. No new mathematical theorem and no new law is claimed. Scope of this paper (scope note): no new mathematical theorem and no new law is claimed. Hadamard's three conditions for well-posedness, the ill-posedness of the Cauchy problem for Laplace's equation, the ill-posedness of the backward heat equation, the Cauchy problem for ultrahyperbolic equations, and the classification of second order operators by signature are all standard. It is not proved that spacetime must be 3+1 ── the discipline is that of Paper 69, and what is done here is to add one entry to its ledger. No anthropic argument is made ── the phrase because there are observers is never used. No measured value is cited ── every number is computed from a definition. No theory of partial differential equations is built ── only exact solutions and a finite Fourier representation are used. Prediction is not defined ── what is treated is the single point of continuous dependence on the data. It is not claimed that ultrahyperbolic equations have no solutions ── solutions exist; what fails is uniqueness and continuous dependence. The arrow of time is not solved ── what Paper 98 recorded as open is left untouched. Quantum theory is not treated. The relation to earlier papers. Paper 69 wrote why 3+1 as an overdetermination and assembled five independent roots selecting four dimensions; not one of them selects which of the four is time, and this paper fills that empty place. Paper 159 showed that ill-posed is not one word and placed the separator in the decay of the singular values; this paper applies that same separator to the number of time dimensions. Paper 154 showed that a limit without its order is not a quantity; that concerns the order of limits and this the signature. Papers 118, 119 and 122 showed that the one word time covers four logical types; what is asked here is not the type but the number. Paper 251 separated special in two dimensions; this paper stands beside it, on the side of special in one. First, being solvable and being predictable are different demands. Hadamard wrote what it is for a problem to be well-posed as three conditions: that a solution exists, that it is unique, and that it depends continuously on the data. The three are independent; the first two are about whether it can be solved, and what corresponds to prediction is the third alone. Why the third is prediction: initial data is measured and then entered, and measurement always carries error. If the error changes the answer, then holding a formula for the solution one still cannot state tomorrow's value. Second, with no time dimension the initial value problem explodes. Take Laplace's equation and, treating the vertical coordinate as the time, solve it as an initial value problem (Hadamard's example). With data of size one over n, the amplification at unit height is 1.101 times ten to the third at n=10, 1.213 times ten to the seventh at n=20, 2.942 times ten to the fifteenth at n=40, and 3.463 times ten to the thirty-second at n=80. The data tends to zero and the solution diverges. Neither existence nor uniqueness has failed ── there is a solution and it is unique. What has failed is the third condition alone. Third, with one time dimension the same procedure stays bounded. Change the equation to the wave equation; one sign has been changed and nothing else. With the same data the amplification is 0.0544 at n=10, 0.0457 at n=20, 0.0186 at n=40 and 0.0124 at n=80. At the same n=80 that is 3.463 times ten to the thirty-second against 0.0124 ── thirty-four orders of magnitude. What changed is one sign in the equation, and not the data, not the method, not the precision. What makes the difference is the signature. Fourth, this is the core. One and the same heat equation exchanges well-posedness for ill-posedness when only its direction is reversed. Mode n is multiplied by the exponential of minus n squared t. At n=80 that is 1.604 times ten to the minus twenty-eighth forwards against 6.235 times ten to the twenty-seventh backwards. Running it on a grid of 512 points, the maximum going forwards stays below one at 0.9759, 0.8944 and 0.8165, while backwards it grows to 2.073 times ten to the eleventh, 2.417 times ten to the hundred and twenty-third, and 1.304 times ten to the two hundred and sixty-fourth, overflowing double precision before reaching t=0.05. To measure what this means for prediction, relative noise of ten to the minus tenth ── standing for observational error ── is added to the data and that component alone is sent both ways: forwards it decays to 9.398, 5.403 and 3.835 times ten to the minus eleventh, while backwards it has grown to 4.135 times ten to the seventeenth already at t=0.001. The signal is buried; one holds the same formula for the solution and cannot state a value. Here is the core: there is no asymmetry on the side of the law. The heat equation is one equation, and reversing time does not turn it into another. The asymmetry is on the side of well-posedness. This does not explain the arrow of time ── as Paper 98 recorded honestly, why there is a low entropy past is unsolved. What can be said here is one step short of that: the phenomenon of being able to predict one way and not the other does not itself require an asymmetric law. Fifth, with two time dimensions what happens next is not determined. Giving time two dimensions, a plane wave gives the dispersion relation that the sum of the squares of the two frequencies equals the square of the wave number. With one time the same procedure returns two values, plus and minus the wave number; with two it returns a whole circle in the frequency plane, a continuum. So the data does not determine what happens next. What has failed this time is uniqueness ── in the third section it was continuous dependence. One word, ill-posed, is naming two different failures (Paper 159). Sixth, one measure separates them: the signature of the principal symbol. Counting the signs of the eigenvalues, (4,0) is elliptic and ill-posed, (3,1) is hyperbolic and well-posed, and (2,2) and (1,3) are ultrahyperbolic and ill-posed. Only one time dimension is well-posed. Since (1,3) is (3,1) with the overall sign reversed and means the same physics, what is to be counted is the size of the smaller sign class. One thing follows: it is not that time is special. The sign class with only one member is what we call time. The direction looks reversed because the definition comes first and time second. Seventh, one independent entry is added to the census of Paper 69. The roots it assembled ── conformal invariance of the Maxwell action, graviton degrees of freedom, exotic four-space, Bertrand and Ehrenfest stability, the maximum of the ball volume ── all select how many, and not one of them selects which of them is time. That is the empty place this paper fills. And this root cannot be derived from the other five: conformal invariance, graviton degrees of freedom and exotic four-space are all statements made after a signature has been assumed, and well-posedness stands on the side of that assumption. They are distinct roots (Paper 58). Even so, 3+1 is not proved here. What can be said reaches no further than that prediction is not an available activity unless there is exactly one time dimension, and adds nothing about why space has three. Closing. Time is one-dimensional, and not because time is special. The equations of a two-time universe can be written and their solutions exist. What breaks is prediction ── add a dimension and the answer stops being unique; remove one and measurement error buries it. There is exactly one signature in which the word tomorrow means anything. And the fifth section showed that even inside that one, one direction permits prediction and the other does not ── without once invoking an asymmetric law. On the making of this work: The ideas and content of this work stem from the author's own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and checking the algebra. Any remaining errors or misinterpretations are solely the author's. Feedback and corrections are sincerely appreciated. ----- 時間はなぜ一本なのか。本稿の答は、そう決めた法則があるからではなく、「予言できる」という要求がそれしか許さないから、である。時間が二本ある宇宙の方程式は書ける。解も存在する。壊れるのは予言のほうである。新しい数学定理も新しい法則も主張しない。 本稿の射程(射程注記):新しい数学定理も新しい法則も主張しない。アダマールの適切性の三条件、ラプラス方程式のコーシー問題の不適切性、後ろ向き熱方程式の不適切性、超双曲型方程式のコーシー問題、二階作用素の符号数による分類は、いずれも標準的である。時空が 3+1 でなければならないとは証明しない──論文69 と同じ規律であり、本稿がするのはその台帳に一本足すことだけである。人間原理を立てない──「観測者がいるから」とは一度も言わない。測定値を引かない──本稿の数はすべて定義から計算したものである。偏微分方程式の理論を作らない──使うのは厳密解と有限次元のフーリエ表示だけである。「予言」を定義しない──扱うのは初期条件への連続依存という一点であって、認識論には立ち入らない。超双曲型方程式に解が無いとは言わない──解は存在する。壊れるのは一意性と連続依存である。時間の矢を解かない──論文98 が未解決と書いたことに手をつけない。量子論を扱わない。 既刊との関係。論文69 は「なぜ 3+1 次元か」を過剰決定として書き、D=4 を選ぶ五つの独立な根を並べた。だがそのどれも、四つのうちどれが時間かを選んでいない。本稿はその空いた位置に一本足す。論文159 は「不適切」が一語でないことを示し、分離子を特異値の落ち方に置いた。本稿は同じ分離子を時間の本数に当てる。論文154 は順序を書かない極限は量ではないと示した。あちらは極限の順序、こちらは符号数であり、別の軸である。論文118・119・122 は「時間」という一語が四つの論理型を覆うことを示した。本稿が問うのは型ではなく本数である。論文251 は「二次元だけ特別」を分けた。本稿はその隣、一次元だけ特別の側に立つ。 第一に、「解ける」と「予言できる」は別の要求である。アダマールは、問題が適切であることを三つの条件で書いた──解が存在すること、解が一つに決まること、初期条件に連続に依存すること。三つは独立であり、前二つは解けるかどうかの話で、予言に対応するのは三つ目だけである。なぜ三つ目が予言なのか。初期条件は測って入れるものであり、測定には必ず誤差がある。誤差が答を変えてしまうなら、解の式を持っていても、明日の値を言うことができない。 第二に、時間が 0 本だと初期値問題が爆発する。ラプラス方程式を取り、縦の座標を時間だと思って初期値問題として解く(アダマールの例)。初期値の大きさが 1/n のとき、y=1 での増幅率は n=10 で 1.101×10^3、n=20 で 1.213×10^7、n=40 で 2.942×10^15、n=80 で 3.463×10^32 であった。初期値は 0 に向かっているのに、解は発散する。しかも存在も一意性も壊れていない──解はあり、一つに決まる。壊れているのは三つ目だけである。 第三に、時間が 1 本だと同じ手順が有界に収まる。方程式を波動方程式に替える。符号を一つ変えただけである。同じ初期値を入れると、増幅率は n=10 で 0.0544、n=20 で 0.0457、n=40 で 0.0186、n=80 で 0.0124 になった。同じ n=80 で 3.463×10^32 と 0.0124 ──34 桁の差である。替えたのは方程式の一つの符号であって、初期条件でも解き方でも精度でもな

Explainable AI (XAI) via Causal Inference and Counterfactual Analysis

Current Explainable AI (XAI) methods frequently deliver post-hoc explanations that lack a fundamental understanding of the causal relationships underpinning AI decision-making. This paper proposes a novel approach to XAI that integrates causal inference and counterfactual analysis, aiming to generate more insightful and actionable explanations. Our methodology leverages causal discovery techniques to identify the key causal factors driving a model's decisions, moving beyond merely highlighting correlations. Furthermore, we employ counterfactual analysis to simulate "what-if" scenarios, allowing us to assess the potential impact of altering specific input features and understand the sensitivity of the model. This approach provides a deeper understanding of the AI system's behavior, ultimately leading to more robust and trustworthy AI models. We demonstrate the potential of this integrated framework through conceptual arguments and a detailed outline of the proposed methodology.

Auditable Credit Risk Intelligence for the U.S. Financial System: A Scalable Explainable-AI Framework Reconciling Predictive Performance with ECOA, FCRA, and Model Risk Governance

Machine-learning credit underwriting in the United States operates under three families of obligation that are not reducible to one another: a statutory prohibition on discrimination, a statutory duty to disclose the specific principal reasons for an adverse action, and prudential expectations for model risk management. Prevailing practice reconciles these demands by sacrificing model capacity, deploying low-capacity scorecards whose interpretability is structural rather than earned. This study argues that the United States regulatory realignment of 2025 and 2026 makes that settlement less defensible, not more. The Consumer Financial Protection Bureau withdrew the interpretive circulars governing algorithmic adverse-action practice and amended Regulation B to disclaim disparate-impact liability under the Equal Credit Opportunity Act, while leaving the statutory disclosure duty untouched and preserving liability for the intentional use of facially neutral proxies. The federal banking agencies simultaneously replaced the prescriptive 2011 interagency model risk guidance with a principles-based instrument that expressly disclaims enforceable standards. Fewer obligations are now externally specified, and more must be self-specified, self-justified, and self-evidenced. Because effects-based exposure persists under the Fair Housing Act and state analogues, and private litigation is unaffected by federal guidance withdrawal, the value of verifiable self-generated evidence rises as external specification recedes. The paper develops ACRIS, the Auditable Credit Risk Intelligence Stack, a five-layer architecture in which admissibility, explainability, disparity control, and governance evidence are enforced during training and serving rather than audited afterwards. Its layers comprise a provenance-gated feature-admissibility screen bounding residual proxy information through conditional mutual information; a shape-constrained predictive core; a constrained-optimization layer recording an entire searched alternative-model frontier, including rejected candidates and rejection rationales; an explanation engine deriving principal reasons from a counterfactual approval baseline and gating disclosure on a per-decision stability margin; and an append-only, tamper-evident governance ledger. A finite-sample sufficient condition for top-k reason-set preservation under parameter resampling is proved, with its assumptions and failure modes stated explicitly. The framework is not empirically validated. A pre-registered protocol of falsifiable propositions, corpora, temporal validation design, baselines, metrics, statistical plan, and pre-committed disconfirmation criteria is specified so that every central claim can be tested and, if wrong, refuted.

Advancing Decoding Methods for Enhanced AI Model Performance and Interpretability

The rapid advancement of artificial intelligence, particularly in large language models, has brought significant challenges in decoding methods that balance performance with interpretability. This paper presents a comprehensive analysis of advanced decoding techniques that enhance both model capabilities and explainability. We explore novel approaches including adaptive temperature sampling, nucleus sampling with dynamic thresholds, and hybrid decoding methods that combine multiple strategies. Our research demonstrates that these advanced techniques can significantly improve model performance metrics while maintaining or enhancing interpretability. Through extensive experimentation across diverse datasets, we show that our proposed hybrid decoding method achieves a 12.3% improvement in perplexity scores while maintaining competitive computational efficiency. The findings contribute to the growing body of research on transparent AI systems and provide practical insights for practitioners aiming to deploy more reliable and interpretable AI models in high-stakes applications.

aiDIVA – hybrid AI for rare disease diagnostics using evidence-based, machine learning and language models

Genome sequencing enables accurate detection of genetic variants and is transforming rare disease diagnostics. While data generation is scalable, prioritization and clinical interpretation remain challenging, often requiring expert manual classification. AI-driven decision support systems are therefore needed to assist in causal variant identification or to fully automate large-scale re-analysis of unsolved cases. Existing tools often estimate variant impact on protein function, but few integrate genomic, phenotypic, and clinical annotation data for diagnosis. We present aiDIVA, an ensemble-AI combining statistical and machine learning models trained on genomic and phenotypic data to identify causal variants among tens of thousands per patient. aiDIVA applies a random forest model to classify pathogenicity and generates evidence-based scores for dominant and recessive diseases. These predictions are integrated with clinical metadata to prioritize the most likely causal variants. Large language models further refine and explain results. The aiDIVA-meta model consolidates all scores into a ranked list. aiDIVA-meta reported the causal variant among the top-3 candidates in 97.4% of a pre-training collected cohort with prior evidence in ClinVar or HGMD, and in 93.3% of a post-training collected cohort of previously unreported variants.

AI as an Aviation Safety Decision Maker

Artificial intelligence is rapidly evolving from an analytical support technology into an active participant in aviation safety decision-making. This paper examines how Safety Management Systems (SMS) must evolve as AI increasingly recommends or executes safety-critical operational decisions within aircraft operations, air traffic management, maintenance, dispatch, and organizational safety processes. Particular attention is given to automation bias, explainability, human override authority, accountability, AI-generated hazards, operational design domains, and continuous safety assurance. The study examines developments involving the FAA, EASA, Airbus, the United States military, and DARPA, while also addressing legal and civil implications arising from shared human–AI decision authority. It argues that future SMS frameworks must treat AI simultaneously as a safety tool, decision-making participant, potential hazard source, and safety control. Successful integration will require explicit authority boundaries, explainable and reconstructable decisions, continuous operational monitoring, enforceable safety constraints, and preservation of human and organizational accountability.

Advanced multi-modal Threat Detection: Explainable AI for Early Warning from Human Behavioral Cues Advanced multi-modal Threat Detection Preprint

Autonomous Enterprise Platforms: A Framework for AI-Guided Decision Loops, Predictive Intelligence, and Continuous Organizational Adaptation

Enterprise platforms are evolving from systems that primarily record and analyze business operations into intelligent environments capable of predicting outcomes, recommending interventions, executing decisions, and learning from their consequences. This article proposes a conceptual Autonomous Enterprise Platform (AEP) based on continuous AI-guided decision loops integrating enterprise sensing, contextual intelligence, predictive analytics, decision intelligence, prescriptive policies, autonomous execution, learning, and governance. The proposed framework extends the classical Monitor, Analyze, Plan, and Execute model of autonomic computing by incorporating continuous prediction, intervention, evaluation, and adaptation. The study synthesizes research published between 2000 and 2022 on autonomous agents, autonomic computing, self-adaptive systems, predictive process monitoring, reinforcement learning, and prescriptive analytics. Three key studies provide the conceptual foundation: Kephart and Chess on autonomic computing, Metzger et al. on proactive process adaptation using deep learning, and Kubrak et al. on prescriptive process monitoring. The framework distinguishes operational, learning, and governance loops to support continuous enterprise adaptation. It emphasizes the transition from predicting business outcomes to selecting and executing appropriate interventions. The study also examines challenges involving causal reasoning, intervention timing, resource constraints, model drift, explainability, and human oversight. Overall, AI-guided decision loops provide a foundation for adaptive, intelligent, and governed enterprise platforms capable of continuous decision making, organizational learning, and operational optimization.

Exploring the Relationship Between Visit Frequency and Customer Retention in Study Cafes Using AI-Based Predictive Modeling

This study empirically investigates actual customer repurchase behavior by analyzing behavioral log and payment data collected from study cafe users in an attendance-based learning service environment. Unlike prior studies that primarily relied on survey-based measures, this study integrates attendance records, payment data, and explainable artificial intelligence (XAI) techniques to provide a data-driven understanding of customer retention behavior. It compares the predictive performance of traditional regression models with machine learning approaches. Specifically, logistic regression, random forest, XGBoost, and neural networks were employed, with SHAP analysis applied as an XAI technique. The results indicate that visit frequency has a significant positive effect, supporting H1, while stay-duration variables show only limited and inconsistent effects, providing partial support for H2 and H3. Total payment in May negatively affects subsequent repurchase, suggesting possible saturation or substitution effects, thereby supporting H4. Age demonstrates a negative effect (supporting H5), and regional differences are captured by the XGBoost model (supporting H6). In terms of predictive performance, machine learning models outperformed logistic regression, with XGBoost achieving the strongest overall results among the evaluated models (ROC-AUC = 0.676; PR-AUC = 0.642). Overall, this study contributes to the literature by presenting empirical evidence based on behavioral data and by highlighting the practical interpretability of integrating XAI techniques. From a managerial perspective, the findings provide actionable insights for designing customer retention strategies based on visit frequency, spending behavior, and regional characteristics, thereby supporting AI-driven decision-making in attendance-based service environments.