Pith. sign in

REVIEW 3 major objections 5 minor 47 references

FinRobot: Generative Business Process AI Agents for Enterprise Resource Planning in Finance

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that replacing static, rule-based ERP workflows with a chain of LLM-driven sub-agents can cut wire-transfer processing time by 40% and reimbursement errors by 94%.

desk verdict Plausible architecture, unsupported performance claims; the evaluation's 'after' numbers are asserted, not measured. read the letter →

arxiv 2506.01423 v1 pith:RR5M5MYI submitted 2025-06-02 cs.AI cs.SEq-fin.GN

Hongyang Yang , Likun Lin , Yang She , Xinyu Liao , Jiaoyang Wang , Runjia Zhang , Yuquan Mo , Christina Dan Wang This is my paper · ORCID
classification cs.AIcs.SEq-fin.GN
keywords generativebusinessprocessAIagentsenterpriseresourceplanninglargelanguagemodelsmulti-agentorchestrationworkflowautomationwiretransferprocessingemployeereimbursementChain-of-Actions
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Enterprise resource planning systems in finance are still built on static, rule-based workflows, and this paper proposes replacing that design with generative AI agents that synthesize and execute workflows on the fly. The authors introduce a five-layer architecture—data modeling, business modeling, LLM reasoning, a Chain-of-Actions execution engine, and deployment—and validate it on two banking workflows: inter-bank wire transfers and employee reimbursements. They report that processing time fell by up to 40% for wires and 82% for reimbursements, and that the invoice error rate dropped by 94%, while new risk-control checkpoints improved regulatory coverage. The paper's goal is to show that a single agent-based framework can handle both structured, compliance-heavy processes and semi-structured, policy-driven ones, making AI-native ERP a realistic next step rather than a set of isolated tools.

What carries the argument

The load-bearing mechanism is the Chain-of-Actions (CoA) execution engine paired with the 5W3H1R event schema. The 5W3H1R schema (Who, When, Where, Why, What, How, How much, How long, Result) turns scattered business data into narrative, decision-centric events so that LLMs can reason causally over enterprise operations. CoA then parses user intent, generates a process plan, maps each step to a specialised sub-agent, executes the agents with contextual inputs, and triggers fallback or human review on failure; its task decomposition and graph-based scheduling let independent checks such as AML screening, account status, and payee verification run in parallel rather than serially. Together the two mechanisms convert a legacy serial workflow into a re-plannable execution graph, which is what produces the reported time and error gains.

What would settle it

Take the same institution and alternate weekly between the rule-based process and the GBPA process on matched wire-transfer and reimbursement workloads, recording end-to-end times and error counts for each run; the central claim predicts the GBPA weeks will show large, reproducible separations (wire times near 9 minutes versus 15, reimbursement errors near 0.8% versus 12.6%). A single audit that finds the 'Before' metrics were logged under non-equivalent conditions would also falsify the attribution, even without new experiments.

Watch

Extended reading notes

Core claim

The central claim is that ERP workflows do not need to be hardcoded; an LLM-driven controller can generate them from business intent and execute them through modular agents, and does so well enough to run production banking tasks. The framework, called Generative Business Process AI Agents (GBPAs), first converts fragmented logs and tables into narrative events using the 5W3H1R schema (Who, When, Where, Why, What, How, How much, How long, Result), then extracts business intent, produces a structured action specification, and hands it to the Chain-of-Actions engine, which decomposes goals, maps steps to specialised agents (document, retrieval, RAG, web search, authorization, data analyst, API, reasoning), and supports parallel branches, fallbacks, and human-in-the-loop review. On 607 historical wire-transfer records, the system re-planned a 13-node serial process into a 9-node process with two parallel clusters, cutting average processing from 15 to 9 minutes; on 250 reimbursement records, it reduced the process from 5 sequential steps to 3 stages with parallel pre-checks, cutting end-to-end time from 2.5 days to 4.25 hours and the error rate from 12.6% to 0.8%. The authors present this as evidence that intent-driven agent orchestration, not static BPMN rules, is a viable architecture for enterprise-grade financial workflows that must remain traceable and compliant.

Load-bearing premise

The before-and-after comparison assumes the baseline metrics (15 minutes per wire, 2.5 days per reimbursement, 12.6% error rate) were measured under the same staff, systems, and workload as the GBPA runs; the paper does not say how these baselines were collected, so the gains could be overstated if the historical records reflect different staffing, backlog, or tooling.

Editorial extensions

If this is right

  • Mid-sized financial institutions could deploy GBPAs against existing core banking and ERP back-ends without replacing the underlying systems, since agents are exposed as stateless microservices and workflows run on standard orchestration engines.
  • Serial workflows with long idle waits can be re-planned into parallel clusters: the wire-transfer case cut inter-node wait time by 57%.
  • Regulatory oversight can be strengthened by inserting extra risk-control checkpoints (e.g., transfer-purpose review) into the generated workflow, rather than bolting them on after the fact.
  • Because the same framework handled both structured wire data and semi-structured receipt data, the architecture is claimed to scale across other ERP domains such as procurement, invoicing, and order-to-cash.
  • Generated workflows remain interpretable and evolvable, so human roles shift from executing steps to reviewing agent plans—a precondition for AI literacy as a concrete enterprise capability.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported improvements come from a before/after reconstruction on historical records rather than a controlled experiment; a sharper test would alternate GBPA and rules-based processing on matched workloads and compare distributions, not just averages.
  • If the 5W3H1R schema generalizes, it could be tested on other event-log-rich ERP processes (for example, order-to-cash or procurement) to see whether LLM-written plans reproduce the 40%-class gains outside the two chosen workflows.
  • The architecture implies that audit logs can be regenerated from the CoA task graph, which may give regulators a way to inspect not just the final output but the decision path—an extension the paper leaves implicit.
  • The '94% error drop' combines both detection (automated tax-system checks at submission) and prevention (parallel pre-check), so the gain may depend as much on added validation steps as on the agent architecture itself; separating those contributions would clarify what is attributable to orchestration.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes Generative Business Process AI Agents (GBPAs), a five-layer architecture for ERP automation in finance that combines a 5W3H1R event-centric data schema, an LLM integration layer, a Chain-of-Actions (CoA) execution engine, and a microservice-based deployment layer. The authors present two case studies, bank wire transfers and employee reimbursements, with before/after tables claiming up to 40% processing-time reduction, 82% reimbursement-time reduction, and 94% error-rate reduction.

Significance. If the claimed performance were supported by rigorous empirical measurement, this would be a useful contribution to LLM-driven enterprise workflow automation, particularly for compliance-heavy financial processes. The architectural description is systematic, the choice of two contrasting workflows is well motivated, and the paper identifies relevant prior work. However, the manuscript's central quantitative claims are not backed by a described measurement protocol, deployment details, or raw data; as submitted, Section 4 supports only a design proposal and an estimate, not a validated system. A substantially stronger empirical section, or an explicit reframing of the before/after figures as simulation-based projections, is needed before the results can be assessed.

major comments (3)
  1. [§4.4–4.5, Tables 2–3] The 'After GBPAs' metrics (9 minutes, 4.25 hours, 0.8% error rate) are presented as results, but the manuscript never states how they were obtained. The only data collection described is of historical records before the intervention (607 wire-transfer records and 250 reimbursement records), and the 'Experiment Execution' paragraphs describe workflow redesigns proposed by the system, not measurements of a deployed system. There is no sample size, measurement interval, instrumentation, or transaction logs for the after condition. As written, these figures could be predictions derived from the task-graph redesign, which would make the abstract's 'Results show' claims unsupported. The authors should either provide genuine deployment measurements with a stated protocol or explicitly relabel all after-condition figures as estimates or simulations.
  2. [§4.4 'Data Description'] The baseline is described only as 'collected and modeled 607 historical transaction records' with no information about how end-to-end time and error rates were measured in the before condition, what staffing or tooling was in place, or whether the workload during the baseline period was comparable to the post-GBPA period. Without this information, the 40% time reduction and 94% error reduction cannot be attributed to GBPAs; the comparison may reflect differences in backlog, staffing, or measurement methodology rather than the intervention.
  3. [§4.4–4.5] The GBPA framework's automatic reconstruction of the as-is workflow (13 serial nodes for wire transfers, 9 subprocesses for reimbursements) is used as the basis for optimization, but its accuracy is never validated. There is no comparison against the institution's authoritative process definitions, no expert review, and no precision/recall or other accuracy numbers. If the LLM-based reconstruction misses or adds nodes, the downstream claims about node reduction and parallelization are not trustworthy. The authors should validate the reconstructed workflows against a ground-truth process model or human expert annotation.
minor comments (5)
  1. [Figure 1 caption] The caption contains typos: 'Generarative' should be 'Generative' and 'Al Agents' should be 'AI Agents'; Figure 3 also uses 'GPBAs' while the rest of the paper uses 'GBPAs'.
  2. [§3.6] Section 3.6 calls the engine 'Chain-of-Agents (CoA)', which is inconsistent with Section 3.5's 'Chain-of-Actions (CoA)'; please harmonize the terminology.
  3. [References [2] and [3]] References [2] and [3] appear to be the same paper (Aguirre and Rodriguez, WEA 2017); the duplicate should be removed.
  4. [Algorithm 1] Algorithm 1 is informal; since the CoA engine is a core contribution, specifying the task-graph representation, dependency handling, and fallback semantics more precisely would make the contribution easier to evaluate and reproduce.
  5. [§4.5] The 'error rate' reported in the reimbursement case study is not defined; the denominator should be stated explicitly (per invoice, per record, per workflow, or per transaction).

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the claimed reductions are empirical before/after estimates, not derivations from fitted parameters; the self-citations are not load-bearing.

full rationale

The paper does not present a derivation chain of the kind the circularity check targets. The central claims in Section 4.4 and Section 4.5—40% time reduction and 94% error-rate reduction—are before/after comparisons built from 607 reconstructed wire-transfer records and 250 reimbursement records, followed by a proposed redesigned workflow. No equation, fitted parameter, or defining relation makes the 'after' metrics equal to the 'before' metrics by construction; the paper simply asserts the redesigned process achieves 9 minutes and 4.25 hours. That is an evidentiary weakness about how the 'after' values were measured, not a circular reduction. The authors cite their own prior work (FinGPT [43], FinRobot [44], [47], Mo and Yu [24]), but these citations appear only as related work or background and are not used to justify the quantitative results or to exclude alternatives. There is no imported uniqueness theorem, no ansatz smuggled in via self-citation, and no renaming of a known result that would make the framework's output equivalent to its input. The most serious concern—that Tables 2 and 3 may report modeled estimates rather than measured deployment outcomes—is a correctness and reproducibility risk, not circularity, because there is no demonstrated identity between inputs and outputs. Overall, no specific circular step can be quoted under the required standard, so the circularity score is low.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper rests on three domain assumptions: LLM reliability in decomposing workflows, the adequacy of the 5W3H1R schema, and the comparability of the before/after metrics. None are tested with control conditions or released artifacts. No free parameters or invented entities are present; the quantitative claims are empirical measurements rather than fits.

assumptions (4)
  • domain assumption LLMs can reliably decompose business goals into executable sub-tasks and generate valid structured action specifications.
    The whole CoA execution engine depends on this; Section 3.4 and Algorithm 1 assume it without evaluation.
  • domain assumption The 5W3H1R event schema preserves enough causal and contextual information for LLMs to reason over enterprise workflows.
    Section 3.2 asserts this bridges the semantic gap, but no comparative experiment is run with and without the schema.
  • domain assumption The baseline 'Before GBPAs' metrics accurately reflect the prior process performance.
    Sections 4.4 and 4.5 use historical records to define baseline time and error rate, but measurement conditions are unspecified.
  • domain assumption Fine-tuned models (e.g., GPT-4, DeepSeek, Qwen) are available and can be adapted for these tasks.
    Section 3.4 lists them as the reasoning core; no exact model versions or fine-tuning details are given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FinRobot: Generative Business Process AI Agents for Enterprise Resource Planning in Finance." pith.science (2026). https://pith.science/paper/RR5M5MYI

@misc{pith2026250601423,
  author       = {Pith},
  title        = {Pith review of: FinRobot: Generative Business Process AI Agents for Enterprise Resource Planning in Finance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RR5M5MYI}},
  note         = {Machine review of arXiv:2506.01423}
}
read the original abstract

Enterprise Resource Planning (ERP) systems serve as the digital backbone of modern financial institutions, yet they continue to rely on static, rule-based workflows that limit adaptability, scalability, and intelligence. As business operations grow more complex and data-rich, conventional ERP platforms struggle to integrate structured and unstructured data in real time and to accommodate dynamic, cross-functional workflows. In this paper, we present the first AI-native, agent-based framework for ERP systems, introducing a novel architecture of Generative Business Process AI Agents (GBPAs) that bring autonomy, reasoning, and dynamic optimization to enterprise workflows. The proposed system integrates generative AI with business process modeling and multi-agent orchestration, enabling end-to-end automation of complex tasks such as budget planning, financial reporting, and wire transfer processing. Unlike traditional workflow engines, GBPAs interpret user intent, synthesize workflows in real time, and coordinate specialized sub-agents for modular task execution. We validate the framework through case studies in bank wire transfers and employee reimbursements, two representative financial workflows with distinct complexity and data modalities. Results show that GBPAs achieve up to 40% reduction in processing time, 94% drop in error rate, and improved regulatory compliance by enabling parallelism, risk control insertion, and semantic reasoning. These findings highlight the potential of GBPAs to bridge the gap between generative AI capabilities and enterprise-grade automation, laying the groundwork for the next generation of intelligent ERP systems.

Figures

Figures reproduced from arXiv: 2506.01423 by the authors.

Figure 2
Figure 2. Transforming Business Data into LLM-Readable Event Semantics via 5W3H1R 3.2 Data Modeling Layer: Event-Centric Representation via 5W3H1R A major challenge in applying LLMs to enterprise systems is that business data is often fragmented and non-narrative, making it diffi￾cult for models to perform causal or contextual reasoning. To address this, we adopt a structured schema—5W3H1R—which reformats logs and transaction… view at source ↗
Figure 1
Figure 1. Architecture of Generarative Business Process Al Agents (GBPAs) task-relevant knowledge, and generates structured action speci￾fications grounded in organizational context. 4. Chain-of-Actions (CoA) Execution Engine: CoA constructs an executable plan from high-level goals. Each action is mapped to a callable sub-agent—such as RAG agents, web search agents, au￾thorization agents, data analysts, or compliance validato… view at source ↗
Figure 4
Figure 4. AI-Native Bank: Wire Transfer Process • Reasoning LLMs: Interpret user intents and enforce constraint rules (e.g., AML, compliance). • Role-Based Agents: Pre-Transaction Assessment Agent, Trans￾fer Execution Agent, and Transaction Query Agent cooperate to complete the workflow end-to-end. • Modular Business Components: Underlying services such as current account, payment execution, and correspondent bank in￾terfaces… view at source ↗
Figures from the paper (2 more)
Figure 3
Figure 3. Figure 3: GBPAs Integrated with Core Banking Systems 4.1 System-Level Integration: GBPAs in Core Banking Architecture [PITH_FULL_IMAGE:figures/full_fig_p005_3.png]
Figure 5
Figure 5. Figure 5: Bank Wire Transfer Process: Traditional vs. GBPAs-Optimized reasoning. Using these events, the GBPA framework automatically reconstructed the end-to-end workflow, identifying 13 serially exe￾cuted process nodes including balance checks, identity verification, transacti…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 32 canonical work pages

  1. [1]

    Addo-Tenkorang and P

    R. Addo-Tenkorang and P. Helo. Enterprise resource planning (erp): A review literature report. InProceedings of the World Congress on Engineering and Computer Science, volume 2, pages 19–21. sn, 2011

    work page 2011

  2. [2]

    Aguirre and A

    S. Aguirre and A. Rodriguez. Automation of a business process using robotic process automation (rpa): A case study. InApplied Computer Sciences in Engineering: 4th Workshop on Engineering Applications, WEA 2017, Cartagena, Colombia, September 27-29, 2017, Proceedings 4, pages 65–71. Springer, 2017

    work page 2017

  3. [3]

    Aguirre and A

    S. Aguirre and A. Rodriguez. Automation of a business process using robotic process automation (rpa): A case study. InApplied Computer Sciences in Engineering, pages 65–71. Springer, 2017

    work page 2017

  4. [4]

    Beheshti, J

    A. Beheshti, J. Yang, Q. Z. Sheng, B. Benatallah, F. Casati, S. Dustdar, H. R. M. Nezhad, X. Zhang, and S. Xue. Processgpt: transforming business process management with generative artificial intelligence. In 2023 IEEE International Conference on Web Services (ICWS), pages 731–739. IEEE, 2023

    work page 2023

  5. [5]

    Bingi, M

    P. Bingi, M. K. Sharma, J. K. Godla, et al. Critical issues affecting an erp implementation.Inf. Syst. Manag., 16(3):7–14, 1999

    work page 1999

  6. [6]

    Genai can revolutionize erp trans- formations, 2025

    Boston Consulting Group. Genai can revolutionize erp trans- formations, 2025. URL https://www.bcg.com/publications/2025/ gen-ai-can-revolutionize-erp-transformations. Accessed: 2025-05-06

    work page 2025

  7. [7]

    Brown, B

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, et al. Language mod- els are few-shot learners.Advances in Neural Information Processing Systems, 33:1877–1901, 2020

    work page 1901

  8. [8]

    Cetindamar, K

    D. Cetindamar, K. Kitto, M. Wu, Y . Zhang, B. Abedin, and S. Knight. Explicating ai literacy of employees at digital workplaces.IEEE trans- actions on engineering management, 71:810–823, 2022

    work page 2022

Show all 47 references
  1. [9]

    Dalsaniya and K

    A. Dalsaniya and K. Patel. Enhancing process automation with ai: The role of intelligent automation in business efficiency.International Journal of Science and Research Archive, 5:322–337, 04 2022. doi: 10.30574/ijsra.2022.5.2.0083

    2022 doi

  2. [10]

    C. Gao, X. Lan, N. Li, Y . Yuan, J. Ding, Z. Zhou, F. Xu, and Y . Li. Large language models empowered agent-based modeling and simulation: A survey and perspectives.arXiv preprint arXiv:2312.11970, 2023

    2023 arXiv

  3. [11]

    Goundar, A

    S. Goundar, A. Nayyar, M. Maharaj, K. Ratnam, and S. Prasad. How ar- tificial intelligence is transforming the erp systems.Enterprise systems and technological convergence: Research and practice, 85, 2021

    2021

  4. [12]

    Gravitas

    S. Gravitas. Autogpt. https://github.com/Significant-Gravitas/ AutoGPT, 2023

    2023

  5. [13]

    V . Gujar. The role of ai agents in crm and erp integration: An analysis. International Journal of Science and Research (IJSR), 14:1032–1037, 02 2025. doi: 10.21275/SR25217112600

    2025 doi

  6. [14]

    D. Guo, D. Yang, H. Zhang, J. Song, and et al. Deepseek-r1: Incentiviz- ing reasoning capability in llms via reinforcement learning. Jan. 2025. doi: 10.48550/arXiv.2501.12948. arXiv:2501.12948

    2025 doi arXiv

  7. [15]

    L. Haider. Artificial intelligence in erp. 2021

    2021

  8. [16]

    C. E. Helfat and M. A. Peteraf. The dynamic resource-based view: Capability lifecycles.Strategic management journal, 24(10):997–1010, 2003

    2003

  9. [17]

    Huang, W

    X. Huang, W. Liu, X. Chen, X. Wang, H. Wang, D. Lian, Y . Wang, R. Tang, and E. Chen. Understanding the planning of llm agents: A survey.arXiv preprint arXiv:2402.02716, 2024

    2024 arXiv

  10. [18]

    F. R. Jacobs et al. Enterprise resource planning (erp)—a brief history. Journal of operations management, 25(2):357–363, 2007

    2007

  11. [19]

    Kampik, C

    T. Kampik, C. Warmuth, A. Rebmann, R. Agam, L. N. Egger, A. Gerber, J. Hoffart, J. Kolk, P. Herzig, G. Decker, et al. Large process models: A vision for business process management in the age of generative ai. KI-Künstliche Intelligenz, pages 1–15, 2024

    2024

  12. [20]

    LangGenius. Dify. https://github.com/langgenius/dify, 2024

    2024

  13. [21]

    A. Liu, B. Feng, B. Xue, B. Wang, and et al. Deepseek-v3 technical report. Dec. 2024. doi: 10.48550/arXiv.2412.19437. arXiv:2412.19437

    2024 doi arXiv

  14. [22]

    Long and B

    D. Long and B. Magerko. What is ai literacy? competencies and design considerations. InProceedings of the 2020 CHI conference on human factors in computing systems, pages 1–16, 2020

    2020

  15. [23]

    S. Mhaskey. Integration of artificial intelligence (ai) in enterprise re- source planning (erp) systems: Opportunities, challenges, and impli- cations.International Journal of Computer Engineering in Research Trends, 11, 12 2024. doi: 10.22362/ijcert/2024/v11/i12/v11i1201

    2024 doi

  16. [24]

    Innovation-Driven and Technology- Enabled Digital Economy

    Y . Mo and S. Yu.Methods and Practices of Bank Digital Transfor- mation: Transformation Framework, Domain Modeling, and Business Model Innovation. China Machine Press, Beijing, China, 2023. ISBN 978-7-111-72979-2. Recommended by AEA China and AZone Archi- tect Community. Part ...

    2023

  17. [25]

    Y . B. Moon. Enterprise resource planning (erp): a review of the litera- ture.International journal of management and enterprise development, 4(3):235–264, 2007

    2007

  18. [26]

    Y .-B. Moon. Enterprise resource planning (erp): a review of the litera- ture.International journal of management and enterprise development, 4(3):235–264, 2007

    2007

  19. [27]

    Niederwieser, D

    E. Niederwieser, D. Siegele, and D. Matt. Ai-driven erp systems. Zeitschrift für wirtschaftlichen Fabrikbetrieb, 120:112–117, 03 2025. doi: 10.1515/zwf-2025-0007

    2025 doi

  20. [28]

    Achiam, S

    OpenAI, J. Achiam, S. Adler, S. Agarwal, L. Ahmad, and et al. Gpt- 4 technical report. Mar. 2023. doi: 10.48550/arXiv.2303.08774. arXiv:2303.08774

    2023 doi arXiv

  21. [29]

    Z. Pan, H. Luo, M. Li, and H. Liu. Chain-of-action: Faithful and multimodal question answering through large language models.arXiv preprint arXiv:2403.17359, 2024

    2024 arXiv

  22. [30]

    Manus ai: China’s revolutionary ai agent set to dis- rupt global industries, 2024

    Pandayoo. Manus ai: China’s revolutionary ai agent set to dis- rupt global industries, 2024. URL https://pandayoo.com/post/ manus-ai-chinas-revolutionary-ai-agent-set-to-disrupt-global-industries/. Accessed: 2024-05-08

    2024

  23. [31]

    Radford, K

    A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, et al. Improving language understanding by generative pre-training. 2018

    2018

  24. [32]

    Shen and Q

    M. Shen and Q. Yang. From mind to machine: The rise of manus ai as a fully autonomous digital agent.arXiv preprint arXiv:2505.02024, 2025

    2025 arXiv

  25. [33]

    G. Team, R. Anil, S. Borgeaud, J.-B. Alayrac, J. Yu, R. Soricut, J. Schalkwyk, A. M. Dai, A. Hauth, K. Millican, et al. Gem- ini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    2023 arXiv

  26. [34]

    Automating repetitive tasks: Ai for invoicing, inventory and more, 2024

    The Economic Times. Automating repetitive tasks: Ai for invoicing, inventory and more, 2024. URL https://economictimes.indiatimes.com/small-biz/sme-sector/ automating-repetitive-tasks-ai-for-invoicing-inventory-and-more/ articleshow/111306183.cms?from=mdr. Accessed: 2025-05-06

    2024

  27. [35]

    Themistocleous, Z

    M. Themistocleous, Z. Irani, R. M. O’Keefe, and R. Paul. Erp problems and application integration issues: An empirical survey. InProceed- ings of the 34th Annual Hawaii International Conference on System Sciences, pages 10–pp. IEEE, 2001

    2001

  28. [36]

    W. M. Van der Aalst, M. Bichler, and A. Heinzl. Robotic process au- tomation, 2018

    2018

  29. [37]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, volume 30. Cur- ran Associates, Inc., 2017. URL https://proceedings.neurips.cc/paper_ files...

    2017

  30. [38]

    L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, Z. Chen, J. Tang, X. Chen, Y . Lin, et al. A survey on large language model based autonomous agents.arXiv preprint arXiv:2308.11432, 2023

    2023 arXiv

  31. [39]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models.Advances in neural information processing systems, 35:24824–24837, 2022

    2022

  32. [40]

    Q. Wu, G. Bansal, J. Zhang, Y . Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liu, A. H. Awadallah, R. W. White, D. Burger, and C. Wang. Autogen: Enabling next-gen llm applications via multi-agent conversation framework. InarXiv, 2023

    2023

  33. [41]

    S. Wu, O. Irsoy, S. Lu, V . Dabravolski, M. Dredze, S. Gehrmann, P. Kambadur, D. Rosenberg, and G. Mann. BloombergGPT: A large language model for finance.arXiv preprint arXiv:2303.17564, 2023

    2023 arXiv

  34. [42]

    Z. Xi, W. Chen, X. Guo, W. He, Y . Ding, B. Hong, M. Zhang, J. Wang, S. Jin, E. Zhou, et al. The rise and potential of large language model based agents: A survey.arXiv preprint arXiv:2309.07864, 2023

    2023 arXiv

  35. [43]

    Yang, X.-Y

    H. Yang, X.-Y . Liu, and C. D. Wang. Fingpt: Open-source financial large language models.FinLLM Symposium at IJCAI 2023, 2023

    2023

  36. [44]

    H. Yang, B. Zhang, N. Wang, C. Guo, X. Zhang, L. Lin, J. Wang, T. Zhou, M. Guan, R. Zhang, et al. Finrobot: An open-source ai agent platform for financial applications using large language models.arXiv preprint arXiv:2405.14767, 2024

    2024 arXiv

  37. [45]

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y . Cao. React: Synergizing reasoning and acting in language models. InInter- national Conference on Learning Representations (ICLR), 2023

    2023

  38. [46]

    Zhang, X

    Z. Zhang, X. Bo, C. Ma, R. Li, X. Chen, Q. Dai, J. Zhu, Z. Dong, and J.- R. Wen. A survey on the memory mechanism of large language model based agents.arXiv preprint arXiv:2404.13501, 2024

    2024 arXiv

  39. [47]

    T. Zhou, P. Wang, Y . Wu, and H. Yang. Finrobot: AI agent for equity research and valuation with large language models. InICAIF 2024: The 1st Workshop on Large Language Models and Generative AI for Finance, 2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.