Pith. sign in

REVIEW 4 major objections 5 minor 59 references

Predicting Business Angel Early-Stage Decision Making Using AI

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper claims that an LLM grading startup pitches on eight CFA factors, plus a trained classifier on those grades, predicts business-angel deal/no-deal outcomes with 85.0% accuracy and tracks human expert grading at Spearman's rho =…

desk verdict Promising proof-of-concept that LLM-generated CFA scores can feed classifiers to beat metadata-only Shark Tank models, but the 85% headline is likely inflated by selection on the same holdout. read the letter →

arxiv 2507.03721 v1 pith:AP73X3K7 submitted 2025-07-04 cs.LG cs.AI

classification cs.LGcs.AI
keywords businessangelinvestmentCriticalFactorAssessmentlargelanguagemodelsstartuppitchevaluationdeal/no-dealpredictionmachinelearningclassificationSharkTank
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

The paper claims that an investment decision by a business angel can be predicted from a short startup pitch by combining two steps: a large language model that grades the pitch against the eight-factor Critical Factor Assessment (CFA) rubric, and a machine-learning classifier trained on those grades. Using 600 transcribed pitches with known deal/no-deal outcomes, the best model reaches 85.0% accuracy, and the LLM's CFA scores correlate strongly with human expert grading (Spearman's $\rho = 0.896$, $p < 0.001$). The authors argue this turns a validated but slow and expensive human evaluation process into a fast, cheap, and scalable one, and that a CFA-prompted model clearly beats an unprompted LLM on the same prediction task (77.4% vs. 58.1% accuracy on 31 pitches).

What carries the argument

The eight-factor Critical Factor Assessment (CFA) rubric is the carrier of the argument: a structured scoring scheme that grades a venture on Features & Benefits, Readiness, Barriers to Entry, Adoption, Supply Chain, Market Size, Entrepreneurial Experience, and Financial Expectations. The paper prompts LLMs to assign letter grades for each factor, converts them to a numeric scale, and feeds those values plus ask amount and equity into standard classifiers. The CFA's non-compensatory logic, where a deal is made only if no factor has a fatal flaw, is what connects the factor scores to the binary investment outcome.

What would settle it

Apply nested cross-validation to the existing 600 pitches, or run the same LLM grading and classifier pipeline on a newly collected set of pitches with known outcomes that played no role in any model or feature selection, and compare the resulting accuracy with 85.0%. If the unbiased estimate falls toward the roughly 66% base rate of deals, or if the CFA-prompted model no longer beats an unprompted LLM by a wide margin, the central predictive claim would be refuted.

Watch

Extended reading notes

Core claim

Using the CFA as the feature-extraction rubric, the paper finds that a prompted LLM reproduces the grading of trained human evaluators, and that a soft-voting ensemble of Naive Bayes, logistic regression, random forest, and gradient boosting, fed a subset of CFA factors plus the requested amount and equity, predicts deal/no-deal on a 60-pitch holdout with 85.0% accuracy, F1 of 0.83, specificity of 0.80, and ROC AUC of 0.82. Total CFA score is the strongest feature, followed by Financial Expectations, Ask Amount, Features & Benefits, Barrier to Entry, and Ask Equity. The paper also reports that the CFA-prompted model outperforms an unprompted LLM on 31 pitches, with the largest gap in specificity (0.92 vs. 0.39), and interprets this as evidence that the structured rubric, not the raw language model, carries the predictive power.

Load-bearing premise

The 85.0% accuracy is treated as an unbiased estimate of how the model will perform on new pitches; this requires that the held-out test set was never used to choose the model or the feature combination, and the paper does not describe nested cross-validation or repeated holdouts that would guarantee that.

Editorial extensions

If this is right

  • A CFA-prompted LLM plus a trained classifier can reproduce much of a human CFA evaluation in seconds and for under a dollar, replacing a process that takes three trained raters and several days at roughly $1,400 per venture.
  • Because the top model used only four factors plus the total score and the ask terms, future versions may need fewer features than the full eight-factor rubric to retain most of the predictive benefit.
  • The model can serve as a triage tool for angel groups, accelerators, and public grant programs, flagging weak ventures for closer human review rather than replacing the final investment decision.
  • The comparison with an unprompted LLM implies that embedding a validated evaluation rubric in the prompt is what improves prediction, a pattern that could extend to other structured frameworks.

Reading between the lines

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

  • If the 85.0% accuracy survives an honest re-estimate, the implicit conclusion is that a large share of early-stage screening can be automated without giving up much expert accuracy; the binding constraint would shift from grading cost to obtaining trustworthy outcome labels for training.
  • The reported number is likely an upper bound: the paper describes testing many feature combinations and classifiers and then selecting the best performers, with no nested cross-validation or repeated holdout, so the same data may have influenced model choice.
  • The most informative next test would run the same CFA-AI pipeline on full-length, unedited angel pitches whose outcomes are not shaped by television production; Shark Tank's aired pitches are curated and success-oriented, a limitation the paper itself acknowledges.
  • Because the LLM grader inherits patterns from its training text, any real deployment should be audited for bias across founder gender, race, and accent, which the paper raises as an open ethical issue rather than a demonstrated property.
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

4 major / 5 minor

Summary. The paper proposes an AI implementation of the Critical Factor Assessment (CFA) framework for predicting business angel deal/no-deal decisions. Using 600 transcribed Shark Tank pitches, the authors prompt several GPT models to assign CFA letter grades to eight factors, convert them to numeric scores, and then train machine-learning classifiers on these LLM-generated features together with ask amount and ask equity. The best model, a soft-voting ensemble using features (1, 3, 8, total, ask $, ask %), is reported to achieve 85.0% accuracy on a 60-pitch holdout. The paper also compares AI-generated CFA grades with human consensus on 31 transcripts (Spearman rho = 0.896) and reports that a CFA-prompted LLM outperforms an unprompted GPT-4 (77.4% vs 58.1% accuracy on N=31). The authors conclude that the integrated CFA-AI pipeline is scalable, reliable, and less-biased.

Significance. If the central claims hold, the paper would provide a useful practical contribution: a low-cost, fast, automated approximation of a validated expert evaluation framework, with potential applications in entrepreneurship education, pitch coaching, and early-stage investor triage. The idea of grounding an LLM feature extractor in a domain-specific rubric is sensible and worth testing. However, the current manuscript does not yet provide a trustworthy estimate of generalization performance, and the 'less-biased' claim is not supported by any direct evidence. The paper is significant in scope but needs stronger validation before its headline numbers can be accepted.

major comments (4)
  1. [Section 5.3 and Table 4] The reported 85.0% accuracy is not established as an unbiased estimate of generalization. The manuscript states that feature combinations ('all 8 factors, 8 choose 7, etc.'), ten classifier families, ensemble methods, and hyperparameter settings were all tested, and Table 4 then reports the best feature combination. Top-performing models were retrained on the full dataset and tested on a 10-20% holdout, but no nested cross-validation or repeated holdout is described. If the same holdout influenced feature/model selection, the accuracy is optimistically biased. With roughly 60 test pitches, the standard error of an accuracy estimate is about 4-6 percentage points, and selection among many configurations can inflate the apparent winner further. The authors should either report a nested cross-validation estimate, or test the final pipeline on a completely independent held-out sample that was never used in any selection step.
  2. [Section 6.1 and Table 3] The H1a result (Spearman rho = 0.896 for GPT-4.1-mini) is also affected by selection on the same test set. The paper compares four GPT models on the same 31 human-graded transcripts and selects GPT-4.1-mini as the best by three correlation metrics. This is a multiple-comparison selection on the same data used to report the winner's correlation, so the quoted rho is likely optimistic. A corrected estimate should be reported, for example by using a separate human-graded validation set or by applying a selection-aware adjustment.
  3. [Section 6.2 and Table 6] The H2a comparison is uneven. The CFA-prompted arm, as described in Sections 5.3 and 6.1, benefits from a supervised classifier trained on 600 labeled deal/no-deal outcomes, while the 'stock' LLM receives no such training. Even if Table 6 reports the direct deal/no-deal decision of the prompted GPT-4 rather than the trained classifier, the text does not make this clear, and the comparison as stated confounds 'framework awareness' with 'exposure to labeled outcomes'. To support the claim that a framework-aware AI outperforms a generalist AI, the two systems should be compared under the same conditions: both either untrained LLMs with only prompt differences, or both augmented with the same supervised classifier. The current H2a result cannot isolate the effect of the CFA framework.
  4. [Abstract and Section 8.5] The claim that the model is 'less-biased' is unsupported by any analysis in the manuscript. No measurements of gender, race, socioeconomic, or other demographic biases are reported; Section 8.5 lists such audits only as future research. The abstract and Section 1 should be revised to remove or substantially qualify the 'less-biased' claim, or the authors should add a direct bias analysis.
minor comments (5)
  1. [Section 7.3.2] The text says 'Features & Benefits (1), Barrier to Entry (3), Adoption (4), and Financial Expectations (8) were identified as the most predictive individual factors,' but Table 4 shows the top model using only factors 1, 3, and 8, while Adoption (4) appears only in the second CatBoost row. Please clarify which model(s) this statement refers to.
  2. [Section 5.1] The paper gathers 1,153 pitch records but matches subtitles for 600 pitches; the criteria for inclusion or exclusion should be stated, as this could affect representativeness.
  3. [Table 2] The 'Adjusted Score' column doubles the raw scale; please make explicit whether this transformation is used in the ML features or only for human-AI comparisons.
  4. [Section 7.5.1] The discussion of a rejected deal due to 'inflated valuation and investment ask' should acknowledge that Ask $ and Ask % are included as features in the model, so this information is partially captured.
  5. [Various] The paper would benefit from a reproducibility appendix listing the exact prompt templates, model versions and temperatures, API call structure, and the random seed / split procedure for the 60-pitch holdout.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the main predictions are benchmarked against external deal labels and human grades, and the CFA self-citations are provenance rather than derivation.

full rationale

The derivation chain is: (i) LLMs assign the eight CFA grades from pitch transcripts; (ii) those grades are compared with human consensus grades on 31 pitches; (iii) the grades plus Ask Amount/Equity are used as features to train classifiers against independently known Shark Tank deal/no-deal outcomes; (iv) the best classifier is tested on a held-out set. Each target quantity (human grade, deal outcome) is external to the model's feature-generation step, so the reported 85.0% accuracy and Spearman r=0.896 are not defined into existence by the inputs. The paper's reliance on Maxwell's earlier CFA work [6] is a genuine self-citation, and the claim that CFA is 'validated' does lean on that prior publication; however, the present paper's own H1b test against external deal labels and its H2 comparison with an unprompted LLM give the central claim independent empirical content. The manuscript also flags its own limitations (Sections 7.5.1-7.5.2: Shark Tank entertainment bias and class imbalance), and the ambiguous holdout/model-selection description in Section 5.3 is a statistical-generalization concern, not a circular reduction of the prediction to the fitted inputs. No equation or construction in the paper makes the predicted outcome equal to the input features by definition.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on the CFA framework and the Shark Tank proxy as imported assumptions, plus several unstated model-selection choices. The most important hidden cost is feature and model selection without a nested validation scheme, which can inflate accuracy. No new entities are introduced.

free parameters (4)
  • Feature subset for best model = Features & Benefits (1), Barrier to Entry (3), Financial Expectations (8), total CFA score, Ask $, Ask %
    Chosen after testing all eight factors, all 7-factor subsets, etc. (Section 5.3); if the selection used the holdout set, the reported 85.0% is optimistic.
  • Ensemble composition = Soft voting of Naive Bayes, Logistic Regression, Random Forest, Gradient Boosting with GPT-4 CFA features
    Top model selected from a large set of classifiers; model selection on the test set would inflate accuracy (Section 5.3, Table 4).
  • Deal classification threshold = 50% confidence
    Table 9 defines deal predictions as AI confidence greater than 50%; the threshold is a modeling choice not justified independently.
  • LLM hyperparameters = Not reported
    The paper mentions temperature and context windows but provides no values, so these fit parameters are unreported (Section 8.1).
assumptions (3)
  • domain assumption The CFA eight-factor framework is a valid predictor of business angel funding decisions.
    Adopted from Maxwell et al. (2009) [6]; the paper does not re-derive or independently validate the framework, yet uses it as ground truth for model design and interpretation (Sections 4, 5.2).
  • domain assumption Shark Tank pitches, transcripts, and Kaggle deal outcomes are a valid proxy for real business angel decisions.
    Justified only by analogy to Maxwell's Dragons' Den research (Section 5.1); this is load-bearing for external validity and acknowledged as a limitation in Section 7.5.1.
  • domain assumption Averaged grades from three trained students on 31 transcripts provide reliable human ground truth for the CFA.
    The human benchmark used to validate LLM grading is small and uses student raters rather than experienced angel evaluators (Section 5.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Predicting Business Angel Early-Stage Decision Making Using AI." pith.science (2026). https://pith.science/paper/AP73X3K7

@misc{pith2026250703721,
  author       = {Pith},
  title        = {Pith review of: Predicting Business Angel Early-Stage Decision Making Using AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AP73X3K7}},
  note         = {Machine review of arXiv:2507.03721}
}
read the original abstract

External funding is crucial for early-stage ventures, particularly technology startups that require significant R&D investment. Business angels offer a critical source of funding, but their decision-making is often subjective and resource-intensive for both investor and entrepreneur. Much research has investigated this investment process to find the critical factors angels consider. One such tool, the Critical Factor Assessment (CFA), deployed more than 20,000 times by the Canadian Innovation Centre, has been evaluated post-decision and found to be significantly more accurate than investors' own decisions. However, a single CFA analysis requires three trained individuals and several days, limiting its adoption. This study builds on previous work validating the CFA to investigate whether the constraints inhibiting its adoption can be overcome using a trained AI model. In this research, we prompted multiple large language models (LLMs) to assign the eight CFA factors to a dataset of 600 transcribed, unstructured startup pitches seeking business angel funding with known investment outcomes. We then trained and evaluated machine learning classification models using the LLM-generated CFA scores as input features. Our best-performing model demonstrated high predictive accuracy (85.0% for predicting BA deal/no-deal outcomes) and exhibited significant correlation (Spearman's r = 0.896, p-value < 0.001) with conventional human-graded evaluations. The integration of AI-based feature extraction with a structured and validated decision-making framework yielded a scalable, reliable, and less-biased model for evaluating startup pitches, removing the constraints that previously limited adoption.

Figures

Figures reproduced from arXiv: 2507.03721 by the authors.

Figure 1
Figure 1. Soft Voting Ensemble – Naive Bayes, Logistic Regression, Random Forest, Gradient Boosting (GPT-4) (1, 3, 8, Total, Ask $, Ask %) Confusion Matrix [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. CFA-Prompted LLM vs. Unprompted LLM Confusion Matrices (N=31) 6.3 Error Analysis For the 31 pitches and 8 factors evaluated by the CFA-prompted GPT4 model, Factors 1, 3, 5 exhibited the largest mean difference when comparing to human evaluations ( [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 54 canonical work pages

  1. [1]

    Mrkajic, S

    B. Mrkajic, S. Murtinu, V.G. Scalera, Is green the new gold? Venture capital and green entrepreneurship, Small. Bus. Econ. 52(4) (2017) 929–950

  2. [2]

    Ramadani, Business angels: who they really are, Strateg

    V. Ramadani, Business angels: who they really are, Strateg. Change. 18(7–8) (2009) 249–258. https://doi.org/10.1002/jsc.852

  3. [3]

    Freear, J.E

    J. Freear, J.E. Sohl, W.E. Wetzel, Angels: personal investors in the venture capital market. Entrep. Reg. Dev. 7(1) (1995) 85–94. https://doi.org/10.1080/08985629500000005

  4. [4]

    Cohen, P

    L. Cohen, P. Wirtz, Characteristics of entrepreneurs, entrepreneurial finance, and growth paths, 2nd Emerging Trends in Entrepreneurial Finance Conference, SSRN Electronic J (2018). https://doi.org/10.2139/ssrn.3151908

  5. [5]

    Tenca, A

    F. Tenca, A. Croce, E. Ughetto, Business angels research in entrepreneurial finance: a literature review and a research agenda, J. Econ. Surv. 32(5) (2018) 1384–1413. https://doi.org/10.1111/joes.12224

  6. [6]

    Maxwell, S.A

    A.L. Maxwell, S.A. Jeffrey, M. Lévesque, Business angel early stage decision making, J. Bus. Ventur. 26(2) (2009) 212–225. https://doi.org/10.1016/j.jbusvent.2009.09.002

  7. [7]

    Kahneman, Thinking, fast and slow, Penguin UK, 2011

    D. Kahneman, Thinking, fast and slow, Penguin UK, 2011

  8. [8]

    Åstebro, Assessing the commercial viability of seed- and early-stage ventures, J

    T. Åstebro, Assessing the commercial viability of seed- and early-stage ventures, J. Priv. Equity. 6(1) (2002) 9–12. https://doi.org/10.3905/jpe.2002.320029

Show all 59 references
  1. [9]

    Blank, B

    S. Blank, B. Dorf, The startup owner’s manual: the step-by-step guide for building a great company. John Wiley & Sons, 2020

  2. [10]

    Cantamessa, V

    M. Cantamessa, V. Gatteschi, G. Perboli, M. Rosano, Startups’ roads to failure, Sustainability. 10(7) (2017) 2346. https://doi.org/10.3390/su10072346

  3. [11]

    Kamps, How venture capital works, Apress eBooks, 2020, pp

    H.J. Kamps, How venture capital works, Apress eBooks, 2020, pp. 9–14. https://doi.org/10.1007/978-1-4842-6065-4_2

  4. [12]

    Vohora, M

    A. Vohora, M. Wright, A. Lockett, Critical junctures in the development of university high-tech spinout companies, Res. Policy. 33(1) (2003) 147–175. https://doi.org/10.1016/s0048-7333(03)00107- 0

  5. [13]

    Markham, S.J

    S.K. Markham, S.J. Ward, L. Aiman-Smith, A.I. Kingon, The Valley of Death as context for role theory in product innovation, J. Prod. Innov. Manag. 27(3) (2010) 402–417. https://doi.org/10.1111/j.1540-5885.2010.00724.x

  6. [14]

    Osnabrugge, A comparison of business angel and venture capitalist investment procedures: an agency theory-based analysis, Ventur

    M.V. Osnabrugge, A comparison of business angel and venture capitalist investment procedures: an agency theory-based analysis, Ventur. Cap. 2 (2000) 91. https://doi.org/10.1080/136910600295729

  7. [15]

    Gorman, W.A

    M. Gorman, W.A. Sahlman, What do venture capitalists do?, J. Bus. Ventur. 4(4) (1989) 231–

  8. [16]

    W.R. Kerr, R. Nanda, J. McQuade, Entrepreneurship reading: financing entrepreneurial ventures, Harvard Business Publishing, 2014. https://hbsp.harvard.edu/product/8072-PDF-ENG (accessed 15 Preprint: July 4, 2025 23 February 2025)

  9. [17]

    Macht, J

    S.A. Macht, J. Robinson, Do business angels benefit their investee companies?, Int. J. Entrep. Behav. Res. 15(2) (2009) 187–208. https://doi.org/10.1108/13552550910944575

  10. [18]

    Hoyos-Iruarrizaga, A

    J. Hoyos-Iruarrizaga, A. Fernández-Sainz, M. Saiz-Santos, High value-added business angels at post-investment stages: key predictors, Int. Small Bus. J. 35(8), 949–968. https://doi.org/10.1177/0266242616686401

  11. [19]

    Croce, M

    A. Croce, M. Guerini, E. Ughetto, Angel financing and the performance of high-tech start-ups, J. Small. Bus. Manag. 56(2) (2016) 208–228. https://doi.org/10.1111/jsbm.12250

  12. [20]

    Festel, S.H

    G.W. Festel, S.H. De Cleyn, Founding angels as an emerging subtype of the angel investment model in high-tech businesses, Ventur. Cap. 15(3) (2013) 261–282. https://doi.org/10.1080/13691066.2013.807059

  13. [21]

    Sohl, The angel market in 2023: an inflection point for women angels? Center for Venture Research 42, 2023

    J.E. Sohl, The angel market in 2023: an inflection point for women angels? Center for Venture Research 42, 2023. https://scholars.unh.edu/cvr/42

  14. [22]

    https://www.cbinsights.com/research/report/venture-trends-2023/ (accessed 16 February 2025)

    CB Insights, State of Venture 2023 Report, 2024. https://www.cbinsights.com/research/report/venture-trends-2023/ (accessed 16 February 2025)

  15. [23]

    Mason, R.T

    C.M. Mason, R.T. Harrison, Is it worth it? The rates of return from informal venture capital investments. J. Bus. Ventur. 17(3) (2002) 211–236. https://doi.org/10.1016/s0883-9026(00)00060-4

  16. [24]

    Kakati, Success criteria in high-tech new ventures, Technovation

    M. Kakati, Success criteria in high-tech new ventures, Technovation. 23(5) (2009) 447–457. https://doi.org/10.1016/s0166-4972(02)00014-7

  17. [25]

    Vinig, M

    T.G. Vinig, M. De Haan, How do venture capitalists screen business plans? Evidence from The Netherlands and the US, SSRN Electronic J. (2002). https://doi.org/10.2139/ssrn.321860

  18. [26]

    Mason, M

    C. Mason, M. Stark, What do investors look for in a business plan? Int. Small Bus. J. 22(3) (2004) 227–248. https://doi.org/10.1177/0266242604042377

  19. [27]

    Hall, C.W

    J. Hall, C.W. Hofer, Venture capitalists’ decision criteria in new venture evaluation. J. Bus. Ventur. 8(1) (1993) 25–42. https://doi.org/10.1016/0883-9026(93)90009-t

  20. [28]

    Adomdza, Why do inventors continue when experts say stop? The effects of overconfidence, optimism and illusion of control, Master’s thesis, University of Waterloo, Canada, 2004

    G. Adomdza, Why do inventors continue when experts say stop? The effects of overconfidence, optimism and illusion of control, Master’s thesis, University of Waterloo, Canada, 2004. http://etd.uwaterloo.ca/etd/gkadomdz2004.pdf

  21. [29]

    Astebro, Key success factors for technological entrepreneurs’ R&D projects, IEEE Trans

    T. Astebro, Key success factors for technological entrepreneurs’ R&D projects, IEEE Trans. Eng. Manag. 51(3) (2004) 314–321. https://doi.org/10.1109/tem.2004.830863

  22. [30]

    Arundale, C

    K. Arundale, C. Mason, Business angel groups as collective action: an examination of the due diligence process, Int. Entrep. Manag. J. 21, 39 (2024). https://doi.org/10.1007/s11365-024-01052-7

  23. [31]

    Koziol, M

    K. Koziol, M. Schmitz, S. Bort, Gender differences in entrepreneurial equity financing—a systematic literature review, Small Bus. Econ. (2025). https://doi.org/10.1007/s11187-024-00989-x

  24. [32]

    Gonzales M., O

    M.A. Gonzales M., O. Terzidis, P. Lütz, B. Heblich, Critical decisions at the early stage of start- ups: a systematic literature review, Innov. Entrep. 13 (2024) 83. https://doi.org/10.1186/s13731-024- 00438-9 Preprint: July 4, 2025 24

  25. [33]

    https://innovationcentre.ca/product- details/cfa-snapshot/ (accessed 16 February 2025)

    Canadian Innovation Centre, CFA Snapshot, 2015. https://innovationcentre.ca/product- details/cfa-snapshot/ (accessed 16 February 2025)

  26. [34]

    Ries, The Lean Startup: How Today's Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses, Crown Business, New York, 2011

    E. Ries, The Lean Startup: How Today's Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses, Crown Business, New York, 2011

  27. [35]

    Sohl, The early-stage equity market in the USA, Ventur

    J.E. Sohl, The early-stage equity market in the USA, Ventur. Cap. 1(2) (1999) 101–120

  28. [36]

    Cassar, The financing of business start-ups, J

    G. Cassar, The financing of business start-ups, J. Bus. Ventur. 19(2) (2009) 261–283. https://doi.org/10.1016/s0883-9026(03)00029-6

  29. [37]

    C. M. Mason, R.T. Harrison, Auditioning for money, J. Priv. Equity. 6(2) (2003) 29–42. https://doi.org/10.3905/jpe.2003.320037

  30. [38]

    Barbaroux, From market failures to market opportunities: managing innovation under asymmetric information, J

    P. Barbaroux, From market failures to market opportunities: managing innovation under asymmetric information, J. Innov. Entrep. 3(1) (2014) 5. https://doi.org/10.1186/2192-5372-3-5

  31. [39]

    Baker, V

    H.K. Baker, V. Ricciardi, How biases affect investor behaviour, Eur. Financial. Rev. (2014) 7–

  32. [40]

    Kanaan, Psychology and financial decisions: a literature assessment, Manag

    G.K. Kanaan, Psychology and financial decisions: a literature assessment, Manag. Financ. 19(5) (1993) 1–10. https://doi.org/10.1108/eb013720

  33. [41]

    https://ssrn.com/abstract=2457425

  34. [42]

    Kliger, A

    D. Kliger, A. Kudryavtsev, The availability heuristic and investors’ reaction to company-specific events, J. Behav. Finance. 11(1) (2010) 50–65. https://doi.org/10.1080/15427561003591116

  35. [43]

    Altman, Chapter 3: Behavioral economics, thinking processes, decision-making, and investment behavior

    M. Altman, Chapter 3: Behavioral economics, thinking processes, decision-making, and investment behavior. In: H.K. Baker, V. Ricciardi (Eds.), Investor behavior: the psychology of financial planning and investing. John Wiley & Sons, 2014, 43–61

  36. [44]

    Chen, Ethics and discrimination in artificial intelligence-enabled recruitment practices, Hum

    Z. Chen, Ethics and discrimination in artificial intelligence-enabled recruitment practices, Hum. Soc. Sci. Commun. 10(1) (2023). https://doi.org/10.1057/s41599-023-02079-x

  37. [45]

    Bartram, J

    S.M. Bartram, J. Branke, M. Motahari, Artificial intelligence in asset management, SSRN Electronic J. (2020). https://doi.org/10.2139/ssrn.3692805

  38. [46]

    LeCun, Y

    Y. LeCun, Y. Bengio, G. Hinton, Deep learning, Nature. 521(7553) (2015) 436–444. https://doi.org/10.1038/nature14539

  39. [47]

    Fatouros, K

    G. Fatouros, K. Metaxas, J. Soldatos, D. Kyriazis, Can large language models beat Wall Street? Unveiling the potential of AI in stock selection, arXiv Cornell University, 2024. https://doi.org/10.48550/arxiv.2401.03737

  40. [48]

    Dellermann, N

    D. Dellermann, N. Lipusch, P. Ebel, K.M. Popp, J.M. Leimeister, Finding the unicorn: predicting early stage startup success through a hybrid intelligence method, International Conference on Information Systems (ICIS), Seoul, South Korea, SSRN Electronic J. (2017). https://doi....

  41. [49]

    Y.Q. Ang, A. Chia, S. Saghafian, Using machine learning to demystify startups’ funding, post- money valuation, and success, in: V. Babich, J.R. Birge, G. Hilary (Eds.), Innovative technology at the interface of finance and operations, Springer Series in Supply Chain Management...

  42. [50]

    Sherk, M

    T. Sherk, M. Tran, T.V. Nguyen, SharkTank deal prediction: dataset and computational model, 11th International Conference on Knowledge and Systems Engineering (2019) 1–6. https://doi.org/10.1109/kse.2019.8919477

  43. [51]

    Deodhar, S

    O. Deodhar, S. Bhatkar, P. Dixit, P. Kulkarni, A. Oak, S. Londhe, Shark Tank deal prediction using machine learning techniques, 2022 IEEE 7th International Conference for Convergence in Technology (I2CT), 2024. https://doi.org/10.1109/i2ct61223.2024.10543

  44. [52]

    Smith, A

    B. Smith, A. Viceisza, Bite me! ABC’s Shark Tank as a path to entrepreneurship, Small Bus. Econ. 50(3) (2017) 463–479. https://doi.org/10.1007/s11187-017-9880-8

  45. [53]

    Lavanchy, P

    M. Lavanchy, P. Reichert, A. Joshi, Blood in the water: an abductive approach to startup valuation on ABC’s Shark Tank. J. Bus. Ventur. Insights. 17 (2022) e00305. https://doi.org/10.1016/j.jbvi.2022.e00305

  46. [54]

    Maxwell, Failing fast: how and why business angels rapidly reject most investment opportunities, Master’s thesis, University of Waterloo, Canada, 2009

    A.L. Maxwell, Failing fast: how and why business angels rapidly reject most investment opportunities, Master’s thesis, University of Waterloo, Canada, 2009. https://uwspace.uwaterloo.ca/handle/10012/4253

  47. [55]

    Bhattarai, A ‘Shark Tank’ appearance can lead to millions — but is it real? The Washington Post, 17 January 2014

    A. Bhattarai, A ‘Shark Tank’ appearance can lead to millions — but is it real? The Washington Post, 17 January 2014. https://www.washingtonpost.com/business/capitalbusiness/a-shark-tank- appearance-can-lead-to-millions--but-is-it-real/2014/01/17/67ccc262-7c9a-11e3-93c1- 0e8881...

  48. [56]

    White, S

    C. White, S. Dooley, M. Roberts, A. Pal, B. Feuer, S. Jain, R. Shwartz-Ziv, N. Jain, K. Saifullah, S. Dey, S. Agrawal, S.S. Sandha, S. Naidu, C. Hegde, Y. LeCun, T. Goldstein, W. Neiswanger, M. Goldblum, LiveBench: a challenging, contamination-free LLM benchmark. arXiv, Cornel...

  49. [57]

    https://www.kaggle.com/datasets/thirumani/shark-tank-us-dataset

    Shark Tank US Dataset, 2024. https://www.kaggle.com/datasets/thirumani/shark-tank-us-dataset

  50. [248]

    https://doi.org/10.1016/0883-9026(89)90014-1

  51. [2024]

    https://doi.org/10.48550/arxiv.2406.19314

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.