Appendix A · Methodology

Appendix A: methodology

This appendix gathers the methodological detail of the study: the design, the scales, the statistical tests, the data-treatment decisions and the theoretical framework that underpins each hypothesis. The body of the report can be read without needing this appendix. Everything technical is documented here for anyone who wants to audit or dig deeper.

Study design

The study follows José Supo's (2013) instrument-validation methodology, applied in four stages: literature review, item construction, expert-judge validation, and pilot study with statistical analysis of item discrimination (n=20).

Two versions of the questionnaire were administered, terminologically adapted to each population ("law firm" vs. "legal department", "clients" vs. "internal clients"), in collaboration between the Laboratory of Law and Artificial Intelligence of Universidad Panamericana, KermaPartners and Lemontech. Data were collected between June and November 2025.

Sample and geographic scope

The sample is non-probabilistic and predominantly Mexican. Participants were recruited through the networks of the convening organizations, which likely overestimates the market's effective level of adoption.

PopulationMexicoChileOther LATAMOutside LATAM
Law firms (n=72)61 (84.7%)04 (5.6%)7 (9.7%)
Legal departments (n=39)18 (46.2%)10 (25.6%)8 (20.5%)3 (7.7%)
Total (n=111)79 (71.2%)10 (9.0%)12 (10.8%)10 (9.0%)

Mexico has the largest subgroup, but sample size alone does not justify country-representative inference. Tests describe associations within this network-recruited sample. Chilean observations (10 departments) are exploratory.

Measurement scales

The instrument uses three main scales:

ScaleRange1 / 035
Agreement1–5Strongly disagreeNeutralStrongly agree
Frequency1–5NeverHalf the timeAlways
Tool usage0–5Not contracted / NeverOccasionalDaily

Other scalar variables: KPI maturity (1–5), knowledge (1–5), % of digitalization (0–4, by ranges), training frequency (0–4).

Statistical tests

  • Likert scales vs. neutral point: Wilcoxon signed-rank test against the neutral value (3.0), two-sided. Used because of the sample size and the ordinal nature of the data.
  • Comparison between two groups (e.g. law firms vs. legal departments): Mann-Whitney U.
  • Comparison between more than two groups (e.g. by country or size): Kruskal-Wallis.
  • Categorical variables: Chi-square (or Fisher's exact if any cell < 5).
  • Relationship between variables: Spearman correlation (ρ), because of the ordinal nature of the scales.
  • Significance level: α = 0.05.

Effect sizes (r): r < 0.1 negligible, 0.1–0.3 small, 0.3–0.5 medium, > 0.5 large. Correlation strength (Evans, 1996): 0.00–0.19 very weak, 0.20–0.39 weak, 0.40–0.59 moderate, 0.60–0.79 strong, 0.80–1.00 very strong.

Data treatment: skip-logic and AI imputation

Nine respondents in law firms and eight in legal departments left the AI-implementation question blank, yet did answer the questions on AI tool usage. This is attributed to a survey-platform artifact, and "Yes, implemented AI" was inferred for those cases.

Since this decision mechanically raises the adoption rate (the report's headline figure), a sensitivity analysis is reported with and without imputation:

PopulationRate with imputation (%)Rate without imputation (%)Difference (pp)
Law firms 73.6 69.8 3.8
Legal departments 69.4 60.7 8.7

With imputation: 53/72 firms and 25/36 departments. Without imputation: 44/63 and 17/28 explicit answers. Three departments have no valid adoption answer. Numerators and denominators both change.

Note on the Adoption Index and part-whole correlations

The Adoption Index is a summative scale of components. Correlating a component with the full index produces a part-whole correlation, inflated by construction. For that reason, whenever the report compares which factors relate to the index, it uses the corrected item-total correlation, which excludes each component from its own total. Both versions (raw and corrected) are reported in the Correlations appendix.

Hypotheses: statement, test and result

There are 18 statements (H01–H18). Item results are available for 17 rows, but H15 and H17 share a test and H01 does not directly measure the existence of policies. H16 lacks a conclusive test.

No.ChapterHypothesis Mean Dp (D) Mean Gp (G)Result
H01 AI · Ethical framework Organizations lack internal policies for ethical AI use 3.87 <0.0001 2.86 0.5601 Not directly testable
H02 AI · Ethical framework The practical application of ethical considerations is inconsistent 3.23 0.2447 3.11 0.7423 No soportada
H03 Capacidad · Infraestructura The digital divide is a barrier to technology adoption 3.36 0.0267 3.15 0.4552 Soportada
H04 Capacidad · Infraestructura There is a lack of technical understanding for effective AI use 3.34 0.0172 3.60 0.0087 No soportada
H05 Capacidad · Infraestructura Technological fragmentation is a significant barrier 2.73 0.0888 3.23 0.2705 No soportada
H06 People · Adaptation Human skills retain their centrality in legal practice 4.16 <0.0001 3.97 <0.0001 Consensus finding
H07 People · Adaptation Human oversight of AI work is a core competency 4.65 <0.0001 4.26 <0.0001 Consensus finding
H08 People · Adaptation Resistance to change is the main barrier to technology adoption 3.54 0.0003 3.64 0.0055 Consensus finding
H09 People · Efficiency Adopting digital technologies increases operational efficiency 3.96 <0.0001 4.11 <0.0001 Soportada
H10 People · Efficiency Task automation reduces operating costs 3.35 0.0208 3.62 0.0040 Soportada
H11 People · Efficiency Automation frees up time for higher-value work 3.88 <0.0001 3.97 <0.0001 Soportada
H12 Personas · Cliente Technology implementation improves client satisfaction 3.43 0.0042 3.77 0.0002 Soportada
H13 Personas · Cliente Technology increases perceived transparency 3.78 <0.0001 3.97 <0.0001 Soportada
H14 Personas · Cliente There is a growing preference for digital interactions with clients 2.64 0.0105 3.23 0.1452 No soportada
H15 IA · Uso Organizations perceive a tangible efficiency benefit from using AI 3.92 <0.0001 3.61 0.0167 Soportada
H16 IA · Uso Those who use AI perceive an improvement in the quality of their services — — — — Indeterminado
H17 IA · Uso AI effectively reduces the time spent on routine tasks 3.92 <0.0001 3.61 0.0167 Soportada
H18 IA · Uso AI tools increase the capacity to handle more cases 3.60 0.0013 3.54 0.0311 Soportada

Correction notes on the hypotheses table

  • H01: not directly testable. The item measures frequency of risk assessment, not policy existence. Means of 3.87 and 2.86 remain descriptive item results. A p-value of 0.5601 does not establish missing policies in departments. The supported verdict and its inclusion in the previous supported-hypothesis count are withdrawn.
  • H15 and H17 (AI usage). They share the same item (time reduction on routine tasks) and therefore identical values. They are kept as distinct theoretical statements, but statistically they are a single test.
  • H16 (perceived quality with AI). It was left indeterminate: it requires comparing adopters vs. non-adopters, and the non-adopter group is too small for a conclusive test.

Theoretical framework by block

The literature that underpins each block of hypotheses is summarized below. It is kept here, outside the body, so as not to overload the general reading.

Infrastructure. Zolea (2023) documents that digitalization broadened access to legal education, but a digital divide persists: many students lack the necessary technology or connectivity. Syahril & Hasan (2024) warn that the mere availability of infrastructure does not guarantee its effective use: it requires technical understanding and organizational capability.

Professional adaptation. Kubicki (cited in Berry, 2024) holds that critical thinking and human connections remain the most valued competencies, even with high technological adoption. Armour et al. (2022) describe three effects of technology on legal employment (substitution, complement and role creation). Human oversight of AI corresponds to the complement effect. The EY (2024) barometer identifies resistance to change —along with the lack of financial and human resources— as the main barrier to legal digital transformation.

Efficiency. Rebenchuk (cited in Berry, 2024) estimates that ~44% of legal tasks are automatable. Armour et al. (2022) link cost reduction to the substitution effect on routine tasks.

Client experience. Clio (2023) identifies client satisfaction as one of the most direct outcomes of technological adoption, with transparency (visibility of the matter's status and timelines) as the central lever.

AI usage. Jackowski & Araszkiewicz (2023) report that 51% of legal firms have already implemented some form of AI and that 38.2% of tasks in legal firms are routine and automatable, which translates into the perception of freed-up professional time.

Ethical framework. Bowan & Katzav (cited in Berry, 2024) argue that AI systems operate without genuine ethical understanding, which opens a gap between technical capability and professional responsibility. Pierce & Goutos (cited in Berry, 2024) note that, since 2012, the ABA Model Rules have required lawyers to keep abreast of the benefits and risks of relevant technology, a duty that now extends to the use of AI. Logvinova (cited in Berry, 2024) warns that regulatory-governance principles are still lacking and that standards are only just being formulated worldwide. In Mexico, where the specific framework for legal AI is still developing, this gap is more pronounced.

Bibliography

References cited in this appendix. Secondary citations (Bowan & Katzav, Kubicki, Logvinova, Pierce & Goutos, Rebenchuk) are drawn from Berry (2024).

  • Armour, J., Parnham, R., & Sako, M. (2022). Augmented lawyering. University of Illinois Law Review, 2022(1), 71.
  • Berry, S. (2024). AI and the Legal Profession: Transforming the future of law (ed. Alex Davies, 2023; Globe Law and Business). Legal Information Management, 24(1), 61–63.
  • Clio. (2023). 2023 Legal Trends Report [Report]. Clio. https://www.clio.com/resources/legal-trends/
  • Evans, J. D. (1996). Straightforward statistics for the behavioral sciences. Brooks/Cole Publishing.
  • EY. (2024). Segundo Barómetro de gerencias legales: retos y oportunidades. Una radiografía de los equipos jurídicos [Report]. EY Servicios Profesionales de Auditoría y Asesorías Ltda.
  • Jackowski, M., & Araszkiewicz, M. (2023). First Global Report on the State of Artificial Intelligence in Legal Practice. Liquid Legal Institute. https://doi.org/10.38023/5501c854-14eb-4529-b6b9-44343e477f44
  • Supo, J. (2013). Cómo validar un instrumento. Lima: Biblioteca Nacional del Perú. https://cvonline.uaeh.edu.mx/Cursos/Maestria/MGIEV/MGIEV15/Unidad_2/lec_0514_como%20validar%20un%20instrumento.pdf
  • Syahril, M. A. F., & Hasan, N. (2024). Legal Analysis of the Application of Cyber Notary in the Notary Profession in Indonesia. Jurnal Litigasi Amsir, 11(3), 352–358.
  • Zolea, S. (2023). Digitalization of Legal Education. Uzbek Journal of Law and Digital Policy, 1.

How to cite

Prince Tritto, P., Villa Berger, P., Salmon, L., Muñoz Ponce, M., & Limón González, J. B. (2026). Technology Maturity Index for the Legal Industry: pilot study (data from Mexico, 2025). Law and Artificial Intelligence Laboratory, Universidad Panamericana; KermaPartners; Lemontech. https://labderia.up.edu.mx/en/legaltech-report